Organisation mondiale de la santé (OMS) · Technical Documents

WHO guidelines for screening and treatment of precancerous lesions for cervical cancer prevention: supplemental material: GRADE evidence-to-recommendation tables and evidence profiles for each recommendation

9 décembre 2013 Organisation mondiale de la santé

Le texte intégral est hébergé par l’organisation qui le publie. lawenc.com indexe les métadonnées et renvoie vers la source officielle.

Texte intégral

WHO guidelines for screening and treatment of precancerous lesions for cervical cancer prevention Supplemental material: GRADE evidence-to-recommendation tables and evidence profiles for each recommendation

WHO/RHR/13.21 © World Health Organization 2013 All rights reserved. Publications of the World Health Organization are available on the WHO web site (www.who.int) or can be purchased from WHO Press, World Health Organization, 20 Avenue Appia, 1211 Geneva 27, Switzerland (tel.: +41 22 791 3264; fax: +41 22 791 4857; e-mail: bookorders@who.int). Requests for permission to reproduce or translate WHO publications –whether for sale or for non-commercial distribution– should be addressed to WHO Press through the WHO web site (www.who.int/about/licensing/copyright_form/en/index.html). The mention of specific companies or of certain manufacturers’ products does not imply that they are endorsed or recommended by the World Health Organization in preference to others of a similar nature that are not mentioned. Errors and omissions excepted, the names of proprietary products are distinguished by initial capital letters. All reasonable precautions have been taken by the World Health Organization to verify the information contained in this publication. However, the published material is being distributed without warranty of any kind, either expressed or implied. The responsibility for the interpretation and use of the material lies with the reader. In no event shall the World Health Organization be liable for damages arising from its use. This publication contains the collective views of an international group of experts [or give name of group] and does not necessarily represent the decisions or the policies of the World Health Organization.

Contents Introduction iv Acronyms and abbreviations v

Section A. GRADE evidence-to-recommendation tables and evidence profiles for each recommendation (negative or unknown HIV status) Recommendation 1 Recommendation 2 Recommendation 3 Recommendation 4 Recommendation 5 Recommendation 6 Recommendation 7 Recommendation 8 Recommendation 9 1 2 10 24 38 53 61 69 83

Section B. GRADE evidence-to-recommendation tables and evidence profiles for each recommendation (HIV-positive status or unknown HIV status in areas with high endemic HIV infection) Recommendation 1 Recommendation 2 Recommendation 3 Recommendation 4 Recommendation 5 Recommendation 6 Recommendation 7 Recommendation 8 Recommendation 9 98 99 106 119 132 145 152 159 172

Introduction This document includes the judgements and evidence for each recommendation as presented and used by the Guideline Development Group to make recommendations for the WHO guidelines for screening and treatment of precancerous lesions for cervical cancer prevention.1 In each section and for each recommendation, we provide: recommendation and remarks, which relate to the strength of the recommendation and the quality of the evidence; an evidence-to-recommendation table, describing the judgements made by the Guideline Development Group; evidence for each recommendation, including: –– flowchart for the screen-and-treat strategies that were compared; –– evidence used for decision-making: 1. diagnostic test accuracy evidence profile; 2. GRADE evidence table for patient-important outcomes following different screen-and-treat strategies (based on model); 3. GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age. –– references.

Available at: http://www.who.int/reproductivehealth/publications/cancers/screening_and_treatment_of_precancerous_lesions/en/index.html

Acronyms and abbreviations ASCUS atypical squamous cells of undetermined significance CI confidence interval CIN cervical intraepithelial neoplasia CKC cold knife conization colp colposcopic impression; women who have abnormal results on colposcopy would not be treated cryo cryotherapy cyto cytology, Papanicolaou test (using conventional or liquid-based cytology); cut-off for screen-positive test is ASCUS DTA diagnostic test accuracy FN false negative, calculated from sensitivity of screening test; women who receive FN screening test results will not receive the treatment they need (because their positive status was undetected) FP false positive, calculated from specificity of screening test; women who receive FP screening test results will receive unnecessary treatment GRADE Grading of Recommendations, Assessment, Development and Evaluation HIV human immunodeficiency virus HPV test human papillomavirus screening test; cut-off for screen-positive test is 1 pg/mL HPVVIA a sequence of screening tests in which an HPV test is followed by VIA; the HPV test is used first and only those women who screen positive for HPV are screened with VIA as a second screening test; women who screen positive on VIA are then treated, while women who screen negative on VIA are not treated HPV +/– CKC strategy in which an HPV screening test is used and screen-positive women are treated with CKC, but screen-negative women are not treated; it should be noted that all screen-and-treat strategies follow this format (also HPVVIA +/– LEEP) LEEP loop electrosurgical excision procedure (also LLETZ, large loop excision of the transformation zone) QoE quality of evidence QUADAS QUality Assessment for Diagnostic Accuracy Studies TN true negative, calculated from specificity of screening test; women who receive TN screening test results will not receive treatment and do not need treatment TP true positive, calculated from sensitivity of screening test; women who receive TP screening test results will receive the treatment they need VIA visual inspection (of the cervix) with acetic acid; can be used (i) as a cervical screening test; or (ii) to assess whether a patient is eligible for cryotherapy

Section A. GRADE evidence-to-recommendation tables and evidence profiles for each recommendation (negative or unknown HIV status)

Recommendation 1 The expert panel recommends against the use of CKC as treatment in a screen-and-treat strategy (strong recommendation, evidence) Remarks: The screen-and-treat strategies considered by the panel with CKC as treatment included an HPV test, VIA, or an HPV test followed by VIA as screening. Although the benefits were similar for CKC compared with cryotherapy or LEEP for all screen-and-treat strategies, the harms were greater with CKC. This recommendation applies to women regardless of HIV status.

Evidence-to-recommendation table Decision domain Quality of evidence Is there high- or moderate-quality evidence? Judgement Yes  Balance of benefits versus harms and burdens Are you confident that the benefits outweigh the harms and burdens for the recommended strategy? Values and preferences Are you confident about the assumed or identified relative values and are they similar across the target population? Resource implications Is the cost small relative to the net benefits for the recommended strategy? No x Summary of reason for judgement There is high- to moderate-quality evidence for the diagnostic test accuracy data for VIA and HPV test. There is low- to verylow-quality evidence for the effects of treatment and the natural progression of CIN from observational studies often with inconsistent results across studies. The link between test accuracy data and treatment effects is very uncertain. The benefits of HPV screen-and-treat strategy (reduction in CIN recurrence, cervical cancer, and related mortality) may be greater than VIA, and the harms may be similar. There may also be slightly greater overtreatment and slightly fewer cancers detected with HPV test compared to VIA. High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was also placed on a reduction in cervical cancer and related mortality versus complications from treatment (e.g. major bleeding or infection requiring hospitalization). Low value was placed on minor infections or bleeding, and the small number of cancers detected at screening or of women overtreated. HPV testing is resource-dependent. Where HPV testing is available, affordable and implementable, the overall net benefit over VIA is worth the resources. But where not available, HPV test may not be worth the benefits.

Yes x

No 

Yes x

No 

Yes 

No x

This recommendation was made using the data from recommendations 2 to 9, in which the outcomes after use of CKC were compared to LEEP and cryotherapy (e.g. HPVCKC in evidence for recommendation 2). Refer to the following recommendations as presented in this section.

Recommendation 2 Where resources permit, the expert panel suggests a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with VIA and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, evidence) In resource-constrained settings, where screening with an HPV test is not feasible, the expert panel suggests a strategy of screen with VIA and treat with cryotherapy (or LEEP when not eligible) over a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, evidence) Remarks: The benefits of screen-and-treat with an HPV test or VIA, compared to no screening, outweighed the harms, but the reductions in cancer and related mortality were greater with an HPV test when compared to VIA. The availability of HPV testing is resource-dependent and, therefore, the expert panel suggests that an HPV test over VIA be provided where it is available, affordable, implementable, and sustainable over time. This recommendation applies to women regardless of HIV status.

Yes x

No 

Yes x

No 

Resources for cold knife conization are greater than for cryotherapy or LEEP. Yes x No 

Evidence for HPV test compared to VIA to screen for CIN2+ 1. Flowchart of screen-and-treat strategies Asymptomatic women

HPV test

VIA

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Cryo eligible?

Yes

Treat with LEEP

Treat with CKC

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

* Outcomes are: mortality from cervical cancer, rate of cervical cancer detection, rate of CIN2+ detection, major bleeding, premature delivery, infertility, STI detection, major infections, and minor infections

2. Evidence used for decision-making: HPV test compared to VIA to screen for CIN2+ Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test 95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92)

2.1 Diagnostic test accuracy (DTA) evidence profile No. of studies (No. of patients) 5 studies (8921 patients) 5 studies (8921 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies Limitations Indirectness Inconsistency Nonea None Noneb Imprecision None Publication bias Undetected DTA QoE Effect per 1000 patients/year for pretest probability of 2% HPV test 19 (17 to 20) 5 more Nonea None Seriousb Nonec Undetected moderate 823 (706 to 892) 30 fewer 5 studies (8921 patients) Cross-sectional and cohort studies Nonea None Seriousb Nonec Undetected 157 (88 to 274) 30 more 5 studies (8921 patients) Cross-sectional and cohort studies Nonea None Noneb None Undetected 1 (0 to 3) 5 fewer 6 (4 to 9) CRITICAL 127 (78 to 206) CRITICAL 853 (774 to 902) CRITICAL VIA 14 (11 to 16) Importance CRITICAL

Outcome True positives (patients with CIN2+) TP absolute difference True negatives (patients without CIN2+) TP absolute difference False positives (patients incorrectly classified as having CIN2+) FP absolute difference False negatives (patients incorrectly classified as not having CIN2+) FP absolute difference

high

moderate

high

5 Footnotes: a We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. This was not downgraded and this was a borderline judgement. b Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values, and could not be explained by the quality of studies. For TP and FN this was a borderline judgement. We downgraded TN and FP and considered this in the context of other factors, in particular imprecision. c Wide CI for TN and FP that may lead to different decisions depending on which of the confidence limits is assumed.

6 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to VIA Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Outcomes Mortality from cervical cancer1 Cervical cancer incidence2 CIN2+ recurrence3 Undetected CIN2+ (FN) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 Unnecessarily treated (FP) Cancer found at first-time screening9 1511 712 – 156 1649 HPV +/– CKC 20 28 1088 HPV +/–LEEP 30 43 1677 1000 397 575 – 225 1061 157 000 2454 60 610 – 24 1139 1210 670 – 125 1321 HPV +/– cryo 30 43 1677 VIA +/– CKC 81 112 4328 VIA +/– LEEP 88 124 4762 6000 318 560 – 180 850 127 000 3168 48 588 – 19 913 VIA +/– cryo 88 124 4762 No screen10 250 350 13 400 – 0 500 0 0 0 – –

7 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) The overall QoE for each of these outcomes is very low . Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

8 2.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPV +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 a

HPV +/– LEEP

HPV +/– cryo

VIA +/– CKC

VIA +/– LEEP

VIA +/– cryo

No screen10

6 8 1109

9 12 1698

23 32 4457

25 35 4891

71 100 13 829

37 52 1062

57 79 1651

150 209 4174

164 229 4608

464 650 12 886

68 96 1015

105 146 1604

276 386 3891

303 424 4325

857 1200 11 943

1511 712 – 156 1649

397 575 – 225 1061

60 610 – 24 1139

1210 670 – 125 1321

318 560 – 180 850

48 588 – 19 913

0 500 – 0 0

Footnotes: Events were calculated in a similar way to that used for Table 2.2. However, events for cervical cancer are based on incidence of cervical cancer in Eastern Africa of 100/1 000 000 cervical cancers per year in women age 15–39 years; 650 for age 40–49 years; and 1200 for age 50–74 years, provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). It is assumed that the recurrence of CIN is constant across age groups but the incidence of cervical cancer and associated mortality increases, thus reducing the overall recurrence of CIN (a proportion of those with CIN will develop cancer or die). These data should be used primarily to make comparisons across screen-and-treat strategies, not within columns for different age groups. 4,5,6,7,8,10 See footnotes for Table 2.2.

3. References to studies included in meta-analysis of diagnostic test accuracy Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynecology & Obstetrics, 2005, 89(2):120–126. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

Recommendation 3 The expert panel suggests a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with cytology followed by colposcopy (with or without biopsy) and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, evidence) Remarks: The reductions in cancer and related mortality were slightly greater with an HPV test only compared to cytology followed by colposcopy. Although there may be overtreatment of populations with high HPV prevalence and consequently more harms, as well as fewer cancers seen at first-time screening with an HPV test, there are greater resources required in cytology programmes due to quality control, training, and waiting time. The addition of colposcopy also requires a second visit. However, in countries where an appropriate/high-quality screening strategy with cytology (referring women with ASCUS or greater results) followed by colposcopy already exists, either an HPV test or cytology followed by colposcopy could be used.

Yes 

No x

Yes x

No 

High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the resources required and lower value on the harms. There may be additional resources required in cytology programmes due to increased training of providers, quality control, and waiting time. Colposcopy following cytology also requires a second visit. However, in countries where an appropriate/high-quality screening strategy with cytology exists, resources would be required to change over to HPV test.

Yes x

No 

Evidence for HPV test compared to cytology followed by colposcopy to screen for CIN2+ 1. Flowchart of screen-and-treat strategies Asymptomatic women

HPV test

Cytology

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Cryo eligible?

Colposcopy

Yes

Test + (TP)

Test – (TN)

Treat with LEEP

Treat with CKC

Treat with cryo

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

*Outcomes are: mortality from cervical cancer, rate of cervical cancer detection, rate of CIN2+ detection, major bleeding, premature delivery, infertility, STI detection, major infections, and minor infections

2. Evidence used for decision-making: HPV test compared to cytology (ASCUS) followed by colposcopic impression to screen for CIN2+ Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test 94% (95% CI: 89 to 97) 90% (95% CI: 86 to 93) Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS) 70% (95% CI: 57 to 81) 95% (95% CI: 92 to 97) Pooled sensitivity colposcopic impresssion Pooled specificity colposcopic impression 95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

(Reference standard: colposcopy with biopsy when indicated)

2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to cytology (ASCUS) followed by colposcopic impression Factors that may decrease quality of evidence Effect per 1000 patients/year for pretest probability of 2% Cytology followed by colposcopic impression 13 6 more 11 studies (39 050 patients) Cross-sectional and cohort studies Serious b Nonec Seriousd Nonee Undetected low 882 (843 to 911) 70 fewer 11 studies (39 050 patients) Cross-sectional and cohort studies Serious b Nonec Seriousd Nonee Undetected 98 (69 to 137) 70 more 11 studies (39 050 patients) Cross-sectional and cohort studies Serious b Nonec Seriousd Nonee Undetected 1 (1 to 2) 6 fewer 7 CRITICAL 28 CRITICAL 952 CRITICAL

No. of studies (No. of patients)a 11 studies (39 050 patients)

Study design Cross-sectional and cohort studies

Limitations Indirectness Inconsistency Serious b Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

DTA QoE

HPV test 19 (18 to 19)

Importance CRITICAL

low

13 Footnotes: a This is the number of studies that assessed data for HPV test and cytology. b We used QUADAS to assess risk of bias. Half of the studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for cytology followed by colposcopy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of HPV test, cytology (ASCUS) and colposcopy sensitivity and specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

14 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to cytology (ASCUS) followed by colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Cyto colp imp +/– CKC 89 125 4782 Cyto colp imp +/– LEEP 96 135 5194 7000 40 573 – 16 757 358 550 – 37 391 94 518 – 53 251 28 000 4794 14 520 – 6 270 Cyto colp imp +/– cryo 96 135 5194

Outcomes Mortality from cervical cancer1 Cervical cancer incidence2 CIN2+ recurrence3 Undetected CIN2+ (FN) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 Unnecessarily treated (FP) Cancer found at first-time screening9

HPV +/– CKC 20 28 1088

HPV +/– LEEP 30 20 1677 1000

HPV +/– cryo 30 43 1677

No screen10 250 350 13 400 – 0 500 – 0 0 – 0

1004 641 – 104 1096

264 550 – 150 705 98 000 2454

15 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% HPV test: pooled sensitivity 94% (95% CI: 89 to 97), pooled specificity 90% (95% CI: 86 to 93) Cytology (ASCUS): pooled sensitivity 70% (95% CI: 57 to 81), pooled specificity 95% (95% CI: 92 to 97) Colposcopic impression: pooled sensitivity 95% (95% CI: 86 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low . Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/ history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

16 2.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPV +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 1004 641 – 104 1096 264 550 – 150 705 40 573 – 16 757 358 550 – 37 391 94 518 – 53 251 14 520 – 6 270 0 500 – 0 0 68 96 1015 105 146 1604 105 146 1604 305 427 4293 330 462 4706 330 462 4706 857 1200 11 943 37 52 1062 57 79 1651 57 79 1651 165 231 4609 179 250 5022 179 250 5022 464 650 12 886 6 8 1109 9 12 1698 9 12 1698 25 36 4925 28 39 5337 28 39 5337 71 100 13 829 HPV +/– LEEP HPV +/– cryo Cytocolp imp +/– CKC Cytocolp imp +/–LEEP Cytocolp imp +/– cryo No screen10

Footnotes: a Events were calculated similar to Table 2.2. However, events for cervical cancer are based on incidence of cervical cancer in Eastern Africa of 100/1 000 000 cervical cancers per year in women age 15–39 years; 650 for age 40–49 years; and 1200 for age 50–74 years, provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). It is assumed that the recurrence of CIN is constant across age groups but the incidence of cervical cancer and associated mortality increases, thus reducing the overall recurrence of CIN (a proportion of those with CIN will develop cancer or die). These data should be used primarily to make comparisons across screen-and-treat strategies, not within columns for different age groups. 4,5,6,7,8,10 See footnotes for Table 2.2.

3. Evidence used for decision-making: HPV test compared to cytology (ASCUS) followed by colposcopic impression and biopsy when indicated to screen for CIN2+ Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test (Reference standard: colposcopy with biopsy when indicated)

94% (95% CI: 89 to 97) 90% (95% CI: 86 to 93)

Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS)

70% (95% CI: 57 to 81) 95% (95% CI: 92 to 97)

3.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to cytology (ASCUS) followed by colposcopic impressed and biopsy when indicated Factors that may decrease quality of evidence Effect per 1000 patients/year for pretest probability of 2% Cytology followed by colposcopy with biopsy 14 5 more 11 studies (39 050 patients) Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected low 882 (843 to 911) 98 fewer 11 studies (39 050 patients) Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected 98 (69 to 137) 98 more 11 studies (39 050 patients) Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected 1 (1 to 2) 5 fewer 6 CRITICAL 0 CRITICAL 980 CRITICAL

No. of studies (No. of patients)a 11 studies (39 050 patients)

Study design Cross-sectional and cohort studies

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

DTA QoE

HPV test 19 (18 to 19)

Importance CRITICAL

low

18 Footnotes: a This is the number of studies that assessed data for HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. This was downgraded one level in the context of other factors, in particular indirectness. c Data for cytology followed by colposcopy +/– biopsy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of HPV test and cytology (ASCUS) sensitivity and specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e

Wide CI for sensitivity and specificity of cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

19 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to cytology (ASCUS) followed by colposcopy impressed and biopsy when indicated Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Cytocolp biopsy +/– CKC 81 113 4328 Cytocolp biopsy +/–LEEP 88 124 4762 6000 40 573 – 16 757 120 517 – 12 131 32 506 – 18 84 0 4794 5 509 – 2 91 Cytocolp biopsy +/– cryo 88 124 4762

HPV +/– CKC 20 28 1088

HPV +/– LEEP 30 20 1677 1000

HPV +/– cryo 30 43 1677

No screen10 250 350 13 400 – 0 500 – 0 0 – 0

1004 641 – 104 1096

264 550 – 150 705 98 000 2454

20 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% HPV test: pooled sensitivity 94% (95% CI: 89 to 97), pooled specificity 90% (95% CI: 86 to 93) Cytology (ASCUS): Pooled sensitivity 70% (95% CI: 57 to 81), pooled specificity 95% (95% CI: 92 to 97) The overall QoE for each of these outcomes is very low . Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/ history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

21 3.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPV +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 1004 641 – 104 1096 264 550 – 150 705 40 573 – 16 757 120 517 – 12 131 32 506 – 18 84 5 509 – 2 91 0 500 – 0 0 68 96 1015 105 146 1604 105 146 1604 276 386 3891 303 424 4325 303 424 4325 857 1200 11 943 37 52 1062 57 79 1651 57 79 1651 150 209 4174 164 229 4608 164 229 4608 464 650 12 886 6 8 1109 9 12 1698 9 12 1698 23 32 4457 25 35 4891 25 35 4891 71 100 13 829 HPV +/– LEEP HPV +/– cryo Cytocolp biopsy +/– CKC Cytocolp biopsy +/–LEEP Cytocolp biopsy +/– cryo No screen10

Footnotes: a Events were calculated similar to Table 3.2. However, events for cervical cancer are based on incidence of cervical cancer in Eastern Africa of 100/1 000 000 cervical cancers per year in women age 15–39 years; 650 for age 40–49 years; and 1200 for age 50–74 years, provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). It is assumed that the recurrence of CIN is constant across age groups but the incidence of cervical cancer and associated mortality increases, thus reducing the overall recurrence of CIN (a proportion of those with CIN will develop cancer or die). These data should be used primarily to make comparisons across screen-and-treat strategies, not within columns for different age groups. 4,5,6,7,8,10 See footnotes for Table 3.2.

4. References 4.1 References to studies included in meta-analysis of diagnostic test accuracy Agorastos T et al. Human papillomavirus testing for primary screening in women at low risk of developing cervical cancer. The Greek experience. Gynecologic Oncology, 2005, 96(3):714–720. Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Bigras G, De Marval F. The probability for a Pap test to be abnormal is directly proportional to HPV viral load: Results from a Swiss study comparing HPV testing and liquid-based cytology to detect cervical cancer precursors in 13 842 women. British Journal of Cancer, 2005, 93(5):575–581. Cardenas-Turanzas M et al. The performance of human papillomavirus high-risk DNA testing in the screening and diagnostic settings. Cancer Epidemiology Biomarkers and Prevention, 2008, 17(10):2865–2871. de Cremoux P et al. Efficiency of the hybrid capture 2 HPV DNA test in cervical cancer screening. A study by the French Society of Clinical Cytology. American Journal of Clinical Pathology, 2003, 120(4):492–499. Depuydt CE et al. BD-ProExC as adjunct molecular marker for improved detection of CIN2+ after HPV primary screening. Cancer Epidemiology Biomarkers and Prevention, 2011, 20(4):628–637. Hovland S et al. A comprehensive evaluation of the accuracy of cervical pre-cancer detection methods in a high-risk area in East Congo. British Journal of Cancer, 2010, 102(6):957–965. Mahmud SM et al. Comparison of human papillomavirus testing and cytology for cervical cancer screening in a primary health care setting in the Democratic Republic of the Congo. Gynecologic Oncology, 2012, 124(2):286–291. Monsonego J et al. Evaluation of oncogenic human papillomavirus RNA and DNA tests with liquid-based cytology in primary cervical cancer screening: the FASE study. International Journal of Cancer, 2011, 129(3):691–701. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Petry KU et al. Inclusion of HPV testing in routine cervical cancer screening for women above 29 years in Germany: results for 8466 patients. British Journal of Cancer, 2003, 88(10):1570–1577. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936.

23 4.2 References to studies included for diagnostic test accuracy of colposcopic impression Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cantor SB et al. Accuracy of colposcopy in the diagnostic setting compared with the screening setting. Obstetrics & Gynecology, 2008, 111(1):7–14. Cremer ML et al. Digital assessment of the reproductive tract versus colposcopy for directing biopsies in women with abnormal Pap smears. Journal of Lower Genital Tract Disease, 2010, 14(1):5–10. Cristoforoni PM et al. Computerized colposcopy: results of a pilot study and analysis of its clinical relevance. Obstetrics & Gynecology, 1995, 85(6):1011–1016. Durdi GS et al. Correlation of colposcopy using Reid colposcopic index with histopathology – a prospective study. Journal of the Turkish German Gynecology Association, 2009, 10(4):205–207. Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651. Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: Is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

Recommendation 4 The expert panel recommends a strategy of screen with VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with cytology followed by colposcopy (with or without biopsy) and treat with cryotherapy (or LEEP when not eligible) (strong recommendation, evidence) Remarks: The benefits and harms of the two screen-and-treat strategies are similar, but there are fewer harms with cytology followed by colposcopy with biopsy when indicated. Despite overtreatment with VIA and fewer cancers detected at first-time screening, more resources are required for cytology programmes with colposcopy (with or without biopsy) due to quality control, training, and waiting time, as well as a second visit. The recommendation for VIA over cytology followed by colposcopy can be applied in countries that are currently considering either strategy or countries that currently have both strategies available. This recommendation applies to women regardless of HIV status.

Yes 

No x

Yes x

No 

High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the resources required and lower value on the harms. Fewer resources are required for VIA. There may be additional resources required in cytology programmes due to increased training of providers, quality control, and waiting time. Colposcopy following cytology also requires a second visit.

Yes x

No 

Evidence for VIA compared to cytology followed by colposcopy to screen for CIN2+ 1. Flowchart of screen-and-treat strategies Asymptomatic women

VIA

Cytology

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Colposcopy Eligible for cryo Not eligible for cryo

Test + (TP & FP) Treat with cryo Treat with CKC Treat with LEEP

Test – (TN & FN)

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

*Outcomes are: mortality from cervical cancer, rate of cervical cancer detection, rate of CIN2+ detection, major bleeding, premature delivery, infertility, STI detection, major infections, and minor infections

2. Evidence used for decision-making: VIA compared to cytology followed by colposcopic impression Diagnostic test accuracy Pooled sensitivity VIA Pooled specificity VIA 77% (95% CI: 65 to 85) 82% (95% CI: 67 to 91) Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS) 84% (95% CI: 76 to 90) 88% (95% CI: 79 to 93) Pooled sensitivity colposcopic impression Pooled specificity colposcopic impression 95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

(Reference standard: colposcopy with biopsy when indicated)

2.1 Diagnostic test accuracy (DTA) evidence profile: VIA compared to cytology followed by colposcopic impression No. of studies (No. of patients) a 11 studies (12 089 patients) 11 studies (12 089 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 2% Cytology followed by colposcopic impression 16 1 fewer Seriousb Nonec Seriousd Nonee Undetected 804 (657 to 892) 108 fewer 11 studies (12 089 patients) Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected 176 (88 to 323) 108 more 11 studies (12 089 patients) Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected 5 (3 to 7) 1 more 4 CRITICAL 68 CRITICAL 912 CRITICAL

Outcome True positives (patients with CIN2+) TP absolute difference True negatives (patients without CIN2+) TN absolute difference False positives (patients incorrectly classified as having CIN2+) FP absolute difference False negatives (patients incorrectly classified as not having CIN2+) FN absolute difference

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

VIA 15 (13 to 17)

Importance CRITICAL

low

27 Footnotes: a This is the number of studies that assessed data for HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for cytology followed by colposcopy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of VIA, cytology (ASCUS) and colposcopy sensitivity and specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e

Wide CI for sensitivity and specificity of cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

28 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: VIA compared to cytology followed by colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Cytocolp imp +/– CKC 54 76 2935 Cytocolp imp +/–LEEP 63 89 3435 4000 36 566 – 14 680 726 602 – 75 792 191 536 – 108 510 68 000 4794 29 553 – 11 548 Cytocolp imp +/– cryo 63 89 3435

VIA +/– CKC 44 62 2384

VIA +/– LEEP 54 75 2911 5000

VIA +/– cryo 54 75 2911

No screen10 250 350 13 400 – 0 500 – 0 0 – 0

901 627 – 93 984

237 545 – 134 633 176 000 3168

29 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% VIA: pooled sensitivity 77% (95% CI: 66 to 85), pooled specificity 82% (95% CI: 67 to 91) Cytology (ASCUS): Pooled sensitivity 84% (95% CI: 76 to 90), pooled specificity 88% (95% CI: 79 to 93) Colposcopic impression: Pooled sensitivity 95% (95% CI: 86 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low . Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/ history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

30 2.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age VIA +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 901 627 – 93 984 237 545 – 134 633 36 566 – 14 680 726 602 – 75 792 191 536 – 108 510 29 553 – 11 548 0 500 – 0 0 151 212 2165 187 261 2692 184 257 2692 184 257 2654 217 304 3155 217 304 3155 857 1200 11 943 82 115 2307 101 142 2836 100 139 2834 100 139 2834 118 165 3336 118 165 3336 464 650 12 886 13 18 2448 16 22 3017 15 21 2975 15 21 2975 18 25 3518 18 25 3518 71 100 13 829 VIA +/– LEEP VIA +/– cryo Cytocolp imp +/– CKC Cytocolp imp +/–LEEP Cytocolp imp +/– cryo No screen10

3. Evidence used for decision-making: VIA compared to cytology followed by colposcopic impression and biopsy when indicated Diagnostic test accuracy Pooled sensitivity VIA Pooled specificity VIA (Reference standard: colposcopy with biopsy when indicated)

77% (95% CI: 66 to 85) 82% (95% CI: 67 to 91)

Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS)

84% (95% CI: 76 to 90) 88% (95% CI: 79 to 93)

3.1 Diagnostic test accuracy (DTA) evidence profile: VIA compared to cytology followed by colposcopic impression and biopsy when indicated No. of studies (No. of patients) a 11 studies (12 089 patients) 11 studies (12 089 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 2% Cytology followed by colposcopy with biopsy 17 2 fewer Seriousb Nonec Seriousd Nonee Undetected

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

VIA 15 (13 to 17)

Importance CRITICAL

⊕⊕⊝⊝ low

804 (657 to 892) 176 fewer

980

CRITICAL

11 studies (12 089 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

176 (88 to 323) 176 more

CRITICAL

11 studies (12 089 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

5 (3 to 7) 2 more

CRITICAL

32 Footnotes: a This is the number of studies that assessed data for HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. This was downgraded one level in the context of other factors, in particular indirectness. c Data for cytology followed by colposcopy +/– biopsy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of VIA and cytology (ASCUS) sensitivity and specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

33 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: VIA compared to cytology followed by colposcopic impression and biopsy when indicated Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Cyto colp biopsy +/– CKC 44 62 2384 Cyto colp biopsy +/–LEEP 54 75 2911 3000 36 566 – 14 680 146 520 – 15 159 38 507 – 22 102 0 4794 6 511 – 2 110 Cyto colp biopsy +/– cryo 54 75 2911

VIA +/– CKC 44 62 2384

VIA +/– LEEP 54 75 2911 5000

VIA +/– cryo 54 75 2911

No screen10 250 350 13 400 – 0 500 – 0 0 – 0

901 627 – 93 984

237 545 – 134 633 176 000 3168

34 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% VIA: pooled sensitivity 77% (95% CI: 66 to 85), pooled specificity 83% (95% CI: 68 to 92) Cytology (ASCUS): Pooled sensitivity 84% (95% CI: 76 to 90), pooled specificity 88% (95% CI: 79 to 93) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see previously for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

35 3.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age VIA +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 901 627 – 93 984 237 545 – 134 633 36 566 – 14 680 146 520 – 15 159 38 507 – 22 102 6 511 – 2 110 0 500 – 0 0 151 212 2165 187 261 2692 184 257 2692 151 212 2165 184 257 2692 184 257 2692 857 1200 11 943 82 115 2307 101 142 2836 100 139 2834 82 115 2307 100 139 2834 100 139 2834 464 650 12 886 13 18 2448 16 22 3017 15 21 2975 13 18 2448 15 21 2975 15 21 2975 71 100 13 829 VIA +/– LEEP VIA +/– cryo Cyto colp biopsy +/– CKC Cyto colp biopsy +/–LEEP Cyto colp biopsy +/– cryo No screen10

4. References 4.1 References to studies included in meta-analysis of diagnostic test accuracy Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cremer M et al. Adequacy of visual inspection with acetic acid in women of advancing age. International Journal of Gynaecology & Obstetrics, 2011, 113(1):68–71. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynaecology & Obstetrics, 2005, 89(2):120–126. Elit L et al. Assessment of 2 cervical screening methods in Mongolia: cervical cytology and visual inspection with acetic acid. Journal of Lower Genital Tract Disease, 2006, 10(2):83–88. Ghaemmaghami F et al. Visual inspection with acetic acid as a feasible screening test for cervical neoplasia in Iran. International Journal of Gynecological Cancer, 2004, 14(3):465–469. Goel A et al. Visual inspection of the cervix with acetic acid for cervical intraepithelial lesions. International Journal of Gynaecology & Obstetrics, 2005, 88(1):25–30. Hedge D et al. Diagnostic value of acetic acid comparing with conventional Pap smear in the detection of colposcopic biopsy-proved CIN. Journal of Cancer Research & Therapeutics, 2011, 7(4):454–458. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Qiao YL et al. A new HPV-DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Sahasrabuddhe VV et al. Comparison of visual inspection with acetic acid and cervical cytology to detect high-grade cervical neoplasia among HIV-infected women in India. International Journal of Cancer, 2012, 130(1):234–240. Sankaranarayanan R et al. Test characteristics of visual inspection with 4% acetic acid (VIA) and Lugol’s iodine (VILI) in cervical cancer screening in Kerala, India. International Journal of Cancer, 2003, 106(3):404–408. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

37 4.2 References to studies included for diagnostic test accuracy of colposcopic impression Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cantor SB et al. Accuracy of colposcopy in the diagnostic setting compared with the screening setting. Obstetrics & Gynecology, 2008, 111(1):7–14. Cremer ML et al. Digital assessment of the reproductive tract versus colposcopy for directing biopsies in women with abnormal Pap smears. Journal of Lower Genital Tract Disease, 2010, 14(1):5–10. Cristoforoni PM et al. Computerized colposcopy: results of a pilot study and analysis of its clinical relevance. Obstetrics & Gynecology, 1995, 85(6):1011–1016. Durdi GS et al. Correlation of colposcopy using Reid colposcopic index with histopathology – a prospective study. Journal of the Turkish German Gynecology Association, 2009, 10(4):205–207. Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica. 1994, 73(8):648–651. Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

Recommendation 5 The expert panel suggests a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with an HPV test followed by colposcopy (with or without biopsy) and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The reductions in cancer and related mortality with either strategy outweigh the harms and costs of no screening, and were similar between the two strategies. Although overtreatment and, consequently, harms are reduced with the addition of colposcopy (with or without biopsy), there are more resource implications with colposcopy due to increased training of providers, quality control, waiting time, and the potential for more women to be lost to follow-up. The addition of colposcopy to an HPV test would also require a second visit. In countries without an existing screening strategy, an HPV test followed by colposcopy is not recommended. This recommendation applies to women regardless of HIV status.

Evidence-to-recommendation table Decision domain Quality of evidence Is there high- or moderate-quality evidence? Judgement Yes  Balance of benefits versus harms and burdens Are you confident that the benefits outweigh the harms and burdens for the recommended strategy? Values and preferences Are you confident about the assumed or identified relative values and are they similar across the target population? Resource implications Is the cost small relative to the net benefits for the recommended strategy? No x Summary of reason for judgement There is low-quality evidence for the diagnostic test accuracy of HPV test followed by colposcopy and we did not have a direct comparison of this triage test to HPV test alone. There is low- to very-low-quality evidence for the effects of treatment and the natural progression of CIN from observational studies often with inconsistent results across studies. The link between test accuracy data and treatment effects is very uncertain. The benefits of HPV test followed by colposcopy and HPV test alone may be similar. However, there were greater harms with HPV test alone (due to overtreatment with HPV test alone). There may also be slightly fewer cancers detected with HPV test followed by colposcopy. High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the resources required and lower value on the harms. There may be additional resources required with the addition of colposcopy (with or without biopsy), there are more resource implications with colposcopy due to increased training of providers, quality control, waiting time, and potential for more women lost to follow up. The addition of colposcopy to HPV test would also require a second visit.

Yes 

No x

Yes x

No 

Yes x

No 

Evidence for HPV test compared to HPV test followed by colposcopy to screen for CIN2+ 1. Flowchart of screen-and-treat strategies Asymptomatic women

HPV test

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Cryo eligible?

Colposcopy

Yes

Test + (TP & FP)

Test – (TN & FN)

Treat with LEEP

Treat with CKC

Treat with cryo

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

*Outcomes are: mortality from cervical cancer, rate of cervical cancer detection, rate of CIN2+ detection, major bleeding, premature delivery, infertility, STI detection, major infections, and minor infections

2. Evidence used for decision-making: HPV test compared to HPV test followed by colposcopic impression Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test (Reference standard: colposcopy with biopsy when indicated)

93% (95% CI: 87 to 96) 88% (95% CI: 82 to 91)

Pooled sensitivity colposcopic impression Pooled specificity colposcopic impression

95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to HPV test followed by colposcopic impression No. of studies (No. of patients)a 15 studies (45 783 patients) 15 studies (45 783 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 2% HPV test followed by colposcopic impression 18 1 more Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

HPV test 19 (17 to 19)

Importance CRITICAL

⊕⊕⊝⊝ low

862 (813 to 892) 49 fewer

911

CRITICAL

15 studies (45 783 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

118 (88 to 167) 50 more

CRITICAL

15 studies (45 783 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

1 (1 to 3) 1 fewer

CRITICAL

41 Footnotes: a This is the number of studies that assessed data for HPV test. b We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by colposcopy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. e Few participants contributed to colposcopy data. Therefore there are wide confidence intervals for colposcopy specificity, which may lead to different decisions depending on which confidence limits are assumed.

42 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to HPV test followed by colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPV colp imp +/– CKC 31 44 1704 HPV colp imp +/–LEEP 41 58 2263 2000 46 586 – 18 887 743 604 – 77 810 195 537 – 111 521 68 000 3545 29 554 – 12 560 HPV colp imp +/– cryo 41 58 2263

HPV +/– CKC 20 28 1088

HPV +/– LEEP 30 43 1677 1000

HPV +/– cryo 30 43 1677

No screen10 250 350 13 400 – 0 500 – 0 0 – 0

1176 665 – 122 1283

309 558 – 175 826 118 000 2454

43 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% HPV test: pooled sensitivity 93% (95% CI: 87 to 96), pooled specificity 88% (95% CI: 82 to 91) Colposcopic impression: pooled sensitivity 95% (95% CI: 82 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

44 2.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPV +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 a

HPV +/– LEEP

HPV +/– cryo

HPV colp imp +/– CKC

HPV colp imp +/–LEEP

HPV colp imp +/– cryo

No screen10

6 8 1109

9 13 1745

9 12 1698

12 17 2305

71 100 13 829

37 52 1062

58 82 1653

57 79 1651

77 108 2213

464 650 12 886

68 96 1015

108 199 1604

105 151 1604

105 146 1562

142 146 2121

142 199 2121

857 1200 11 943

1176 665 – 122 1283

309 558 – 175 826

46 586 – 18 887

743 604 – 77 810

195 537 – 111 521

29 554 – 12 560

0 500 – 0 0

Footnotes: Events were calculated similar to Table 2.2. However, events for cervical cancer are based on incidence of cervical cancer in Eastern Africa of 100/1 000 000 cervical cancers per year in women age 15–39 years; 650 for age 40–49 years; and 1200 for age 50–74 years, provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). It is assumed that the recurrence of CIN is constant across age groups but the incidence of cervical cancer and associated mortality increases, thus reducing the overall recurrence of CIN (a proportion of those with CIN will develop cancer or die). These data should be used primarily to make comparisons across screen-and-treat strategies, not within columns for different age groups. 4,5,6,7,8,10 See footnotes for Table 2.2.

3. Evidence used for decision-making: HPV test compared to HPV test followed by colposcopic impression Diagnostic test accuracy Pooled sensitivity HPV test (Reference standard: colposcopy with biopsy when indicated)

93% (95% CI: 87 to 96)

Pooled specificity HPV test

88% (95% CI: 82 to 91)

3.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to HPV test followed by colposcopic impression No. of studies (No. of patients)a 15 studies (45 783 patients) 15 studies (45 783 patients) Factors that may decrease quality of evidence DTA QoE Effect per 1000 patients/year for pretest probability of 2% HPV test followed by colposcopy with biopsy 19 0 more Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected

Study design Cross-sectional and cohort studies

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

HPV test 19 (17 to 19)

Importance CRITICAL

⊕⊕⊝⊝ low

862 (804 to 892) 118 fewer

980

CRITICAL

15 studies (45 783 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

118 (88 to 176) 118 more

CRITICAL

15 studies (45 783 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

1 (1 to 3) 0

CRITICAL

46 Footnotes: a This is the number of studies that assessed data for HPV test. b We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by colposcopy +/– biopsy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. e Few participants contributed to colposcopy data. Therefore there are wide confidence intervals for colposcopy specificity, which may lead to different decisions depending on which confidence limits are assumed.

47 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to HPV test followed by colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPV colp biopsy +/– CKC 20 28 1088 HPV colp biopsy +/–LEEP 30 43 1677 1000 46 586 – 18 887 163 523 – 17 178 43 508 – 24 115 0 3545 6 512 – 3 123 HPV colp biopsy +/– cryo 30 43 1677

HPV +/– CKC 20 28 1088

HPV +/– LEEP 30 43 1677 1000

HPV +/– cryo 30 43 1677

No screen10 250 350 13 400 – 0 500 – 0 0 – 0

1176 665 – 122 1283

309 558 – 175 826 118 000 2454

48 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% HPV test: pooled sensitivity 93% (95% CI: 87 to 96), pooled specificity 88% (95% CI: 82 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

49 3.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPV +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 a

HPV +/– LEEP

HPV +/– Cryo

HPV colp biopsy +/– CKC

HPV colp biopsy +/–LEEP

HPV colp biopsy +/– cryo

No screen10

6 8 1109

9 12 1698

6 8 1109

9 12 1698

71 100 13 829

37 52 1062

57 79 1651

37 52 1062

57 79 1651

464 650 12 886

68 96 1015

105 146 1604

68 96 1015

105 146 1604

857 1200 11 943

1176 665 – 122 1283

309 558 – 175 826

46 586 – 18 887

163 523 – 17 178

43 508 – 24 115

6 512 – 3 123

0 500 – 0 0

Footnotes: Events were calculated similar to Table 3.2. However, events for cervical cancer are based on incidence of cervical cancer in Eastern Africa of 100/1 000 000 cervical cancers per year in women age 15–39 years; 650 for age 40–49 years; and 1200 for age 50–74 years, provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). It is assumed that the recurrence of CIN is constant across age groups but the incidence of cervical cancer and associated mortality increases, thus reducing the overall recurrence of CIN (a proportion of those with CIN will develop cancer or die). These data should be used primarily to make comparisons across screen-and-treat strategies, not within columns for different age groups. 4,5,6,7,8,10 See footnotes for Table 3.2.

4. References 4.1 References to studies included in meta-analysis of diagnostic test accuracy Agorastos T et al. Human papillomavirus testing for primary screening in women at low risk of developing cervical cancer. The Greek experience. Gynecologic Oncology, 2005, 96(3):714–720. Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Bigras G, De Marval F. The probability for a Pap test to be abnormal is directly proportional to HPV viral load: Results from a Swiss study comparing HPV testing and liquid-based cytology to detect cervical cancer precursors in 13 842 women. British Journal of Cancer. 2005, 93(5):575–581. Blumenthal PD et al. Adjunctive testing for cervical cancer in low resource settings with visual inspection, HPV, and the Pap smear. International Journal of Gynecology & Obstetrics, 2001, 72(1):47–53. Cardenas-Turanzas M et al. The performance of human papillomavirus high-risk DNA testing in the screening and diagnostic settings. Cancer Epidemiology Biomarkers and Prevention, 2008, 17(10):2865–2871. de Cremoux P et al. Efficiency of the hybrid capture 2 HPV DNA test in cervical cancer screening. A study by the French Society of Clinical Cytology. American Journal of Clinical Pathology, 2003, 120(4):492–499. Depuydt CE et al. BD-ProExC as adjunct molecular marker for improved detection of CIN2+ after HPV primary screening. Cancer Epidemiology Biomarkers and Prevention, 2011, 20(4):628–637. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynecology & Obstetrics, 2005, 89(2):120–126. Hovland S et al. A comprehensive evaluation of the accuracy of cervical pre-cancer detection methods in a high-risk area in East Congo. British Journal of Cancer, 2010, 102(6):957–965. Mahmud SM et al. Comparison of human papillomavirus testing and cytology for cervical cancer screening in a primary health care setting in the Democratic Republic of the Congo. Gynecologic Oncology, 2012, 124(2):286–291. Monsonego J et al. Evaluation of oncogenic human papillomavirus RNA and DNA tests with liquid-based cytology in primary cervical cancer screening: the FASE study. International Journal of Cancer, 2011, 129(3):691–701.

51 Petry KU et al. Inclusion of HPV testing in routine cervical cancer screening for women above 29 years in Germany: results for 8466 patients. British Journal of Cancer, 2003, 88(10):1570–1577. Qiao YL et al. A new HPV-DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

4.2 References to studies included for diagnostic test accuracy of colposcopic impression Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cantor SB et al. Accuracy of colposcopy in the diagnostic setting compared with the screening setting. Obstetrics & Gynecology 2008, 111(1):7–14. Cremer ML et al. Digital assessment of the reproductive tract versus colposcopy for directing biopsies in women with abnormal Pap smears. Journal of Lower Genital Tract Disease, 2010, 14(1):5–10. Cristoforoni PM et al. Computerized colposcopy: results of a pilot study and analysis of its clinical relevance. Obstetrics & Gynecology, 1995, 85(6):1011–1016. Durdi GS et al. Correlation of colposcopy using Reid colposcopic index with histopathology – a prospective study. Journal of the Turkish German Gynecology Association, 2009, 10(4):205–207. Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651.

52 Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: is VIA superseding Pap smear? Journal of SAFOG. 2011, 3(3):131–134.

Recommendation 6 The expert panel suggests either a strategy of screen with an HPV test followed by VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) or a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The reductions in cancer and related mortality were greater with an HPV test used as a single screening test than with an HPV test followed by VIA, and this reduction was even greater in women of HIV-positive status. However, there may be overtreatment, and thus potentially greater harms with screen-and-treat when using an HPV test as a single test. There is also some uncertainty about the effects of an HPV test followed by VIA and how VIA performs after a positive HPV test because there was no direct evidence about this strategy. There is also the potential for additional resources that are required to refer women for VIA testing after a positive HPV test, the need for a second visit to perform VIA, and increased training to perform both tests. For these reasons, the recommendation is for either an HPV test followed by VIA or an HPV test only, and it is conditional. It is to be noted that benefits are more pronounced compared to harms in women of HIV-positive status when using an HPV test only.

Evidence-to-recommendation table Decision domain Quality of evidence Is there high- or moderate-quality evidence? Judgement Yes  Balance of benefits versus harms and burdens Are you confident that the benefits outweigh the harms and burdens for the recommended strategy? No x Summary of reason for judgement There is low-quality evidence for the diagnostic test accuracy of HPV test followed by VIA and compared to HPV test alone. There is low- to very-low-quality evidence for the effects of treatment and the natural progression of CIN from observational studies often with inconsistent results across studies. The link between test accuracy data and treatment effects is very uncertain. There may be fewer major harms with HPV test followed by VIA than with HPV test alone due to less overtreatment. There may also be slightly greater cancers detected with HPV test followed by VIA than with HPV test alone. However, there may be slightly greater CIN recurrence, cervical cancer, and related mortality with HPV test followed by VIA. In women of HIV-positive status there were still fewer harms, less overtreatment and greater cancers detected at first-time screening. However, there was even greater CIN recurrence, cervical cancer and related mortality with HPV test followed by VIA in women of HIV-positive status than in women of unknown status. High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was also placed on reducing overtreatment and resulting complications, and resource use. Greater resources may be required for HPV test followed by VIA due to adding on an additional test. However, there is less overtreatment (fewer treatments provided) and fewer complications requiring hospitalization.

Yes 

No x

Values and preferences Are you confident about the assumed or identified relative values and are they similar across the target population? Resource implications Is the cost small relative to the net benefits for the recommended strategy?

Yes x

No 

Yes 

No x

Evidence for an HPV test followed by VIA compared to an HPV test to screen for CIN2+ 1. Flowchart of screen-and-treat strategies Asymptomatic women

HPV test

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

VIA

Cryo eligible?

Test + (TP & FP)

Test – (TN & FN)

Yes

Eligible for cryo

Not eligible for cryo

Treat with LEEP

Treat with CKC

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

*Outcomes are: mortality from cervical cancer, rate of cervical cancer detection, rate of CIN2+ detection, major bleeding, premature delivery, infertility, STI detection, major infections, and minor infections

2. Evidence used for decision-making: HPV test followed by VIA compared to HPV test Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test 95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92)

moderate

⊕⊕⊕⊝

moderate 137 more

⊕⊕⊕⊝

960

823 (706 to 892)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

Noned

Undetected

moderate 137 fewer

⊕⊕⊕⊝

157 (88 to 274)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

None

Undetected

moderate

⊕⊕⊕⊝

7 6 more

1 (0 to 3)

CRITICAL

56 Footnotes: a We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. The decision to downgrade was a borderline judgement and was considered in the context of other factors. b Data for HPV test followed by VIA were calculated based on sensitivity and specificity of the two tests. Direct data were unavailable. c Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. d Wide CI for HPV test sensitivity and VIA specificity, and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

57 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA compared to HPV test Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVVIA +/– CKC 91 128 4905 HPVVIA +/– LEEP 99 138 5311 7000 288 540 – 30 314 76 514 – 43 202 20 000 3168 11 521 – 5 217 1511 712 – 156 1649 HPVVIA +/– cryo 99 138 5311

HPV +/– CKC 20 28 1088

HPV +/– LEEP 30 43 1677 1000 397 575 – 225 1061 157 000 2454

HPV +/– cryo 30 43 1677

No screen10 250 350 13 400 –

60 610 – 24 1139

0 500 – 0 0 – 0

58 Footnotes: The colours in the table: I n each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

59 2.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPVVIA +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 288 540 – 30 314 76 514 – 43 202 11 521 – 5 217 1511 712 – 156 1649 397 575 – 225 1061 60 610 – 24 1139 0 500 – 0 0 313 438 4403 338 473 4809 338 473 4809 68 96 1015 105 146 1604 105 146 1604 857 1200 11 943 170 237 4728 183 256 5134 183 256 5134 37 52 1062 57 79 1651 57 79 1651 464 650 12 886 26 37 5052 28 39 5459 28 39 5459 6 8 1109 9 12 1698 9 12 1698 71 100 13 829 HPVVIA +/– LEEP HPVVIA +/– cryo HPV +/– CKC HPV +/–LEEP HPV +/– cryo No screen10

Recommendation 7 The expert panel suggests a strategy of screen with an HPV test followed by VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with VIA and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The reductions in cancer and related mortality with an HPV test followed by VIA or with VIA alone outweighed the harms. However, the harms may be greater when using VIA only, which is likely due to overtreatment. Although, a slightly larger number of cancers may be detected on initial screen with VIA only. This recommendation is conditional due to the uncertain costs of providing the sequence of two tests (HPV test followed by VIA) over the single VIA test. In countries where an HPV test is not available, we suggest screening with VIA only. This recommendation applies to women regardless of HIV status.

Yes x

No 

Yes x

No 

High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the greater number of complications and the number of women overtreated. Greater resources with overtreatment with VIA alone. However there may be additional resources required to refer women for VIA testing after a positive HPV test, the need for a second visit, and increased training to perform both tests.

Yes 

No x

Evidence for an HPV test followed by VIA compared to VIA to screen for CIN2+ 1. Flowchart of screen-and-treat strategies Asymptomatic women

HPV test

VIA

Test + (TP & FP)

Test – (TN & FN)

Suspect cervical cancer

Test + (TP & FP)

Test – (TN & FN)

VIA

Eligible for cryo

Not eligible for cryo

Test + (TP & FP)

Test – (TN & FN) Treat with cryo Treat with CKC Treat with LEEP

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

*Outcomes are: mortality from cervical cancer, rate of cervical cancer detection, rate of CIN2+ detection, major bleeding, premature delivery, infertility, STI detection, major infections, and minor infections

2. Evidence used for decision-making: HPV test followed by VIA compared to VIA Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test 95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92)

moderate

⊕⊕⊕⊝

moderate 107 more

⊕⊕⊕⊝

960

853 (774 to 902)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

Noned

Undetected

moderate 107 fewer

⊕⊕⊕⊝

127 (78 to 206)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

Noned

Undetected

moderate

⊕⊕⊕⊝

7 1 more

6 (4 to 9)

CRITICAL

64 Footnotes: a We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. The decision to downgrade was a borderline judgement and was considered in the context of other factors. b Data for HPV test followed by VIA were calculated based on sensitivity and specificity of the two tests. Direct data were unavailable. c Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. d Wide CI for HPV test sensitivity and VIA specificity, and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

65 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA compared to VIA Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVVIA +/– CKC 91 128 4905 HPVVIA +/– LEEP 99 138 5311 7000 288 540 – 30 314 76 514 – 43 202 20 000 2454 11 521 – 5 217 1210 670 – 125 1321 HPVVIA +/– cryo 99 138 5311

VIA +/– CKC 81 113 4328

VIA +/–LEEP 88 124 4762 6000 318 560 – 180 850 127 000 3168

VIA +/– cryo 88 124 4762

No screen10 250 350 13 400 –

48 588 – 19 913

0 500 – 0 0 – 0

66 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

67 2.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPVVIA +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 a

HPVVIA +/– LEEP

HPVVIA +/– cryo

VIA +/– CKC

VIA +/–LEEP

VIA +/– cryo

No screen10

26 37 5052

28 39 5459

23 32 4457

25 35 4891

71 100 13 829

170 237 4728

183 256 5134

150 209 4174

164 229 4608

464 650 12 886

313 438 4403

338 473 4809

276 386 3891

303 424 4325

857 1200 11 943

288 540 – 30 314

76 514 – 43 202

11 521 – 5 217

1210 670 – 125 1321

318 560 – 180 850

48 588 – 19 913

0 500 – 0 0

3. References to studies included in meta-analysis of diagnostic test accuracy

Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynaecology & Obstetrics, 2005, 89(2):120–126. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

Recommendation 8 The expert panel suggests a strategy of screen with an HPV test followed by VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with cytology followed by colposcopy (with or without biopsy) and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The benefits of the two screen-and-treat strategies are similar. However, there may be higher resources required in cytology programmes due to quality control, training, and waiting time. The addition of colposcopy requires a second visit. This recommendation applies to women regardless of HIV status.

Yes 

No x

Yes x

No 

High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the resources required. Fewer resources may be required for HPV test followed by VIA as there may be additional resources required in cytology programmes due to increased training of providers, quality control, and waiting time. Colposcopy following cytology also requires a second visit.

Yes x

No 

Evidence for an HPV test followed by VIA compared to cytology followed by colposcopy to screen for CIN2+ 1. Flowchart of screen-and-treat strategies Asymptomatic women

HPV test

Cytology

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

VIA

Colposcopy

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Eligible for cryo

Not eligible for cryo

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

*Outcomes are: mortality from cervical cancer, rate of cervical cancer detection, rate of CIN2+ detection, major bleeding, premature delivery, infertility, STI detection, major infections, and minor infections

2. Evidence used for decision-making: HPV test followed by VIA compared to cytology (ASCUS) and colposcopic impression Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS) (Reference standard: colposcopy with biopsy when indicated)

94% (95% CI: 89 to 97) 90% (95% CI: 86 to 93) 70% (95% CI: 57 to 81) 95% (95% CI: 92 to 97)

Pooled sensitivity VIA Pooled specificity VIA Pooled sensitivity colposcopic impression Pooled specificity colposcopic impression

69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92) 95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

72 2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to cytology (ASCUS) and colposcopic impression No. of studies (No. of patients)a 14 studies (34 584 patients) 14 studies (34 584 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 2% HPV test followed by VIA 13 0 Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected Cytology followed by colposcopic impression 13

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

Importance CRITICAL

⊕⊕⊝⊝ low

967 15 more

952

CRITICAL

14 studies (34 584 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

13 15 fewer

CRITICAL

14 studies (34 584 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

7 0

CRITICAL

Footnotes: a This is the number of studies that assessed DTA data for: 1. HPV test and VIA; and 2. HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by VIA and cytology followed by colposcopy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of HPV test, VIA, cytology (ASCUS) and colposcopy sensitivity/specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of HPV test followed by VIA and cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

73 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA compared to cytology (ASCUS) and colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVVIA +/– CKC 91 128 4905 HPVVIA +/– LEEP 99 138 5311 7000 222 531 – 23 242 58 511 – 33 156 13 000 3168 9 516 – 3 167 358 550 – 37 391 HPVVIA +/– cryo 99 138 5311 Cytocolp imp +/– CKC 89 125 4782 Cytocolp imp +/– LEEP 96 135 5194 7000 94 518 – 53 251 28 000 4794 14 526 – 6 270 Cytocolp imp +/– cryo 96 135 5194

No screen10 250 350 13 400 – 0 500 – 0 0 – 0

74 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% VIA: Pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) Cytology (ASCUS): Pooled sensitivity 70% (95% CI: 57 to 81), pooled specificity 95% (95% CI: 92 to 97) Colposcopy: Pooled sensitivity 95% (95% CI: 86 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

75 2.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPVVIA +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 a

HPVVIA +/– LEEP

HPVVIA +/– cryo

Cytocolp imp +/– CKC

Cytocolp imp +/–LEEP

Cytocolp imp +/– cryo

No screen10

26 37 5052

28 39 5459

25 36 4925

28 39 5337

71 100 13 829

170 237 4728

183 256 5134

165 231 4609

179 250 5022

464 650 12 886

313 438 4403

338 473 4809

305 427 4293

330 462 4706

857 1200 11 943

222 531 – 23 242

58 511 – 33 156

9 516 – 3 167

358 550 – 37 391

94 518 – 53 251

14 526 – 6 270

0 500 – 0 0

3. Evidence used for decision-making: HPV test followed by VIA compared to cytology (ASCUS) and colposcopic impression with biopsy when indicated Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test 94% (95% CI: 89 to 97) 90% (95% CI: 86 to 93) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92) Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS) 70% (95% CI: 57 to 81) 95% (95% CI: 92 to 97)

(Reference standard: colposcopy with biopsy when indicated)

77 3.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to cytology (ASCUS) and colposcopy with biopsy when indicated Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 2% Cytology followed by colposcopy with biopsy 14 1 fewer 14 studies (34 584 patients) Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected

No. of studies (No. of patients)a 14 studies (34 584 patients)

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

HPV test followed by VIA 13

Importance CRITICAL

⊕⊕⊝⊝ low

967 13 fewer

980

CRITICAL

14 studies (34 584 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

13 13 more

CRITICAL

14 studies (34 584 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

7 1 more

CRITICAL

Footnotes: a This is the number of studies that assessed DTA data for: 1. HPV test and VIA, and 2. HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by VIA and cytology followed by colposcopy +/– biopsy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of HPV test, VIA and cytology (ASCUS) sensitivity/specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of HPV test followed by VIA and cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

78 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA compared to cytology (ASCUS) and colposcopy with biopsy when indicated Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVVIA +/– CKC 91 128 4905 HPVVIA +/– LEEP 99 138 5311 7000 222 531 – 23 242 58 511 – 33 156 13 000 3168 9 516 – 3 167 120 517 – 12 131 HPVVIA +/– cryo 99 138 5311 Cytocolp biopsy +/– CKC 81 113 4328 Cytocolp biopsy +/–LEEP 88 124 4762 6000 32 506 – 18 84 0 3545 5 509 – 2 91 Cytocolp biopsy +/– cryo 88 124 4762

No screen10 250 350 13 400 – 0 500 – 0 0 – 0

79 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% HPV test: Pooled sensitivity 94% (95% CI: 89 to 97), pooled specificity 90% (95% CI: 86 to 93) VIA: Pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) Cytology (ASCUS): Pooled sensitivity 70% (95% CI: 57 to 81), pooled specificity 95% (95% CI: 92 to 97) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

80 3.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPVVIA +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 222 531 – 23 242 58 511 – 33 156 9 516 – 3 167 120 517 – 12 131 32 506 – 18 84 5 509 – 2 91 0 500 – 0 0 313 438 4403 338 473 4809 338 473 4809 276 386 3891 303 424 4325 303 424 4325 857 1200 11 943 170 237 4728 183 256 5134 183 256 5134 150 209 4174 164 229 4608 164 229 4608 464 650 12 886 26 37 5052 28 39 5459 28 39 5459 23 32 4457 25 35 4891 25 35 4891 71 100 13 829 HPVVIA +/– LEEP HPVVIA +/– cryo Cyto colp biopsy +/– CKC Cyto colp biopsy +/– LEEP Cyto colp biopsy +/– cryo No screen10

4. References 4.1 References to studies included in meta-analysis of diagnostic test accuracy Agorastos T et al. Human papillomavirus testing for primary screening in women at low risk of developing cervical cancer. The Greek experience. Gynecologic Oncology, 2005, 96(3):714–720. Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Bigras G, De Marval F. The probability for a Pap test to be abnormal is directly proportional to HPV viral load: Results from a Swiss study comparing HPV testing and liquid-based cytology to detect cervical cancer precursors in 13 842 women. British Journal of Cancer, 2005, 93(5):575–581. Cardenas-Turanzas M et al. The performance of human papillomavirus high-risk DNA testing in the screening and diagnostic settings. Cancer Epidemiology Biomarkers and Prevention, 2008, 17(10):2865–2871. de Cremoux P et al. Efficiency of the hybrid capture 2 HPV DNA test in cervical cancer screening. A study by the French Society of Clinical Cytology. American Journal of Clinical Pathology, 2003, 120(4):492–499. Depuydt CE et al. BD-ProExC as adjunct molecular marker for improved detection of CIN2+ after HPV primary screening. Cancer Epidemiology Biomarkers and Prevention, 2011, 20(4):628–637. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynaecology & Obstetrics, 2005, 89(2):120–126.Hovland S et al. A comprehensive evaluation of the accuracy of cervical pre-cancer detection methods in a high-risk area in East Congo. British Journal of Cancer, 2010, 102(6):957–965. Mahmud SM et al. Comparison of human papillomavirus testing and cytology for cervical cancer screening in a primary health care setting in the Democratic Republic of the Congo. Gynecologic Oncology, 2012, 124(2):286–291. Monsonego J et al. Evaluation of oncogenic human papillomavirus RNA and DNA tests with liquid-based cytology in primary cervical cancer screening: the FASE study. International Journal of Cancer, 2011, 129(3):691–701. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Petry KU et al. Inclusion of HPV testing in routine cervical cancer screening for women above 29 years in Germany: results for 8466 patients. British Journal of Cancer, 2003, 88(10):1570–1577.

82 Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

4.2 References to studies included for diagnostic test accuracy of colposcopic impression Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cantor SB et al. Accuracy of colposcopy in the diagnostic setting compared with the screening setting. Obstetrics & Gynecology, 2008, 111(1):7–14. Cremer ML et al. Digital assessment of the reproductive tract versus colposcopy for directing biopsies in women with abnormal Pap smears. Journal of Lower Genital Tract Disease, 2010, 14(1):5–10. Cristoforoni PM et al. Computerized colposcopy: results of a pilot study and analysis of its clinical relevance. Obstetrics & Gynecology, 1995, 85(6):1011–1016. Durdi GS et al. Correlation of colposcopy using Reid colposcopic index with histopathology – a prospective study. Journal of the Turkish German Gynecology Association, 2009, 10(4):205–207. Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651. Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: Is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

Recommendation 9 The expert panel suggests a strategy of screen with an HPV test followed by VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with an HPV test followed by colposcopy (with or without biopsy) and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The reductions in cancer and related mortality of screen-and-treat with an HPV test followed by colposcopy (with or without biopsy) may be slightly greater compared to an HPV test followed by VIA. The panel agreed that the benefits of either strategy outweigh the harms and costs; however, the difference in costs between the strategies is uncertain. There may be more resource implications with colposcopy due to increased training of providers, quality control, waiting time, and the potential for more women to be lost to follow-up. It is also unclear whether women would perceive a difference between VIA and colposcopy; however, a biopsy during colposcopy may be less acceptable than VIA. This recommendation applies to women regardless of HIV status.

Evidence-to-recommendation table Decision domain Quality of evidence Is there high- or moderate-quality evidence? Judgement Yes  Balance of benefits versus harms and burdens Are you confident that the benefits outweigh the harms and burdens for the recommended strategy? Values and preferences Are you confident about the assumed or identified relative values and are they similar across the target population? Resource implications Is the cost small relative to the net benefits for the recommended strategy? No x Summary of reason for judgement There is low-quality evidence for the diagnostic test accuracy of both triage tests and a comparison between the strategies. There is low- to very-low-quality evidence for the effects of treatment and the natural progression of CIN from observational studies often with inconsistent results across studies. Also the link between test accuracy data and treatment effects is very uncertain. The benefits of HPV test followed by colposcopy (reduction in CIN recurrence, cervical cancer, and related mortality) may be greater than with HPV test followed by VIA. But there may be greater overtreatment with HPV test followed by colposcopy without biopsy. Little or no difference in cancers detected. High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the greater number of women overtreated and potential complications. High value was placed on women finding a biopsy less acceptable than visual inspection. There may be greater resource implications by adding colposcopy than with adding VIA to the HPV test due to increased training of providers, quality control, waiting time, and potential for more women lost to follow up.

Yes 

No x

Yes x

No 

Yes 

No x

Evidence for an HPV test followed by VIA compared to an HPV test followed by colposcopy to screen for CIN2+ 1. Flowchart of screen-and-treat strategies Asymptomatic women

HPV test

HPV

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

VIA

Colposcopy

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Eligible for cryo

Not eligible for cryo

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

*Outcomes are: mortality from cervical cancer, rate of cervical cancer detection, rate of CIN2+ detection, major bleeding, premature delivery, infertility, STI detection, major infections, and minor infections

2. Evidence used for decision-making: HPV test followed by VIA compared to HPV test followed by colposcopic impression Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test 95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92) Pooled sensitivity colposcopic impression Pooled specificity colposcopic impression 95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

(Reference standard: colposcopy with biopsy when indicated)

86 2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to HPV test followed by colposcopic impression No. of studies (No. of patients)a 5 studies (8921 patients) 5 studies (8921 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 2% HPV test followed by VIA 13 5 fewer Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected HPV test followed by colposcopic impression 18

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

Importance CRITICAL

⊕⊕⊝⊝ low

960 71 more

889

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

20 71 fewer

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

7 5 more

CRITICAL

Footnotes: a This is the number of studies that assessed DTA data for HPV test and VIA. b We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by VIA and for HPV test followed by colposcopic impression were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. e Wide CI for HPV test sensitivity and VIA specificity, and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

87 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA compared to HPV test followed by colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVàVIA +/– CKC 91 128 4905 HPVàVIA +/– LEEP 99 138 5311 7000 288 540 – 30 314 76 514 – 43 202 20 000 3168 11 521 – 5 217 937 631 – 97 1022 HPVàVIA +/– cryo 99 138 5311 HPV colp imp +/– CKC 31 44 1704 HPV colp imp +/–LEEP 41 58 2263 2000 246 546 – 140 658 91 000 3545 37 568 – 15 706 0 500 – 0 0 – 0 HPV colp imp +/– cryo 42 58 2263

No screen10 250 350 13 400

88 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) Colposcopy: pooled sensitivity 95% (95% CI: 86 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

89 2.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPVàVIA +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 a

HPVàVIA +/– LEEP

HPVàVIA +/– cryo

HPV colp imp +/– CKC

HPV colp imp +/– LEEP

HPV colp imp +/– cryo

No screen10

26 37 5052

28 39 5459

9 13 1745

12 17 2305

71 100 13 829

170 237 4728

183 256 5134

58 82 1653

77 108 2213

464 650 12 886

313 438 4403

338 473 4809

108 151 1562

142 199 2121

857 1200 11 943

288 540 – 30 314

76 514 – 43 202

11 521 – 5 217

937 631 – 97 1022

246 546 – 140 658

37 568 – 15 706

0 500 – 0 0

3. Evidence used for decision-making: HPV test followed by VIA compared to HPV test followed by colposcopy with biopsy Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test (Reference standard: colposcopy with biopsy when indicated)

95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91)

Pooled sensitivity VIA Pooled specificity VIA

69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92)

3.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA vs HPV test followed by colposcopy with biopsy No. of studies (No. of patients)a 5 studies (8921 patients) 5 studies (8921 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 2% HPV test followed HPV test followed by colposcopy with by VIA biopsy Importance 13 6 fewer Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected 19 CRITICAL

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

⊕⊕⊝⊝ low

960 20 fewer

980

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

20 20 more

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

7 6 more

CRITICAL

91 Footnotes: a This is the number of studies that assessed DTA data for HPV test and VIA. b We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by VIA and for HPV test followed by colposcopy with biopsy when indicated were calculated based on sensitivity and specificity of the two tests. Direct data were not available. d Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. e Wide CI for HPV test sensitivity and VIA specificity, and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

92 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA compared to HPV test followed by colposcopy with biopsy Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVàVIA +/– CKC 91 128 4905 HPVàVIA +/– LEEP 99 138 5311 7000 288 540 – 30 314 76 514 – 43 202 20 000 3168 11 521 – 5 217 163 523 – 17 178 HPVàVIA +/– cryo 99 138 5311 HPV colp biopsy +/– CKC 20 28 1088 HPV colp biopsy +/–LEEP 30 43 1677 1000 43 508 – 24 115 0 3545 6 512 – 3 123 0 500 – 0 0 – 0 HPV colp biopsy +/– cryo 31 43 1677

No screen10 250 350 13 400

93 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 2% VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low-quality evidence for treatment effects and natural progression/history data. 1 We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer, we assumed 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 2 We assume no cervical cancer in TN or FP. To calculate cervical cancer incidence in women with persistent CIN2+, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). 3 We assume no CIN2+ in TN and FP. Our calculations in the model are based on 70% natural persistence of CIN2+ with no treatment (30% regression) in FN. The incidence of cervical cancer and mortality are also subtracted from the CIN2+ in FN (see above for calculations). TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. 4 We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. 5 We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. 6 We did not identify any data about the risk of infertility after treatment for CIN2+. 7 We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. 8 We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. 9 Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for tests with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is conducted). 10 ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

94 3.3 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies by age HPVVIA +/– CKC 15–39 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 40–49 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence 50–74 yearsa Mortality from cervical cancer Cervical cancer incidence CIN2+ recurrence Complications (same across all groups) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 a

HPVVIA +/– LEEP

HPVVIA +/– cryo

HPV colp biopsy +/– CKC

HPV colp biopsy +/–LEEP

HPV colp biopsy +/– cryo

No screen10

26 37 5052

28 39 5459

6 8 1109

9 12 1698

71 100 13 829

170 237 4728

183 256 5134

37 52 1062

57 79 1651

464 650 12 886

313 438 4403

338 473 4809

68 96 1015

105 146 1604

857 1200 11 943

288 540 – 30 314

76 514 – 43 202

11 521 – 5 217

163 523 – 17 178

43 508 – 24 115

6 512 – 3 123

0 500 – 0 0

96 Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetric Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651. Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: Is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

Section B. GRADE evidence-to-recommendation tables and evidence profiles for each recommendation (HIV-positive status or unknown HIV status in areas with high endemic HIV infection)

Recommendation 1 The expert panel recommends against the use of CKC as treatment in a screen-and-treat strategy (strong recommendation, ⊕⊝⊝⊝ evidence) Remarks: The screen-and-treat strategies considered by the panel with CKC as treatment included the HPV test, VIA, or an HPV test followed by VIA as screening. Although the benefits were similar for CKC compared with cryotherapy or LEEP for all screen-and-treat strategies, the harms were greater with CKC. This recommendation applies to women regardless of HIV status.

Yes 

No 

Yes 

No 

Resources for cold knife conization are greater than for cryotherapy or LEEP. Yes  No 

This recommendation was made using the data from recommendations 1 to 8, in which the outcomes after use of CKC were compared to LEEP and cyrotherapy. Refer to the following recommendations as presented in this section.

Recommendation 2 Where resources permit, the expert panel suggests a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with VIA and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) In resource-constrained settings, where screening with an HPV test is not feasible, the expert panel suggests a strategy of screen with VIA and treat with cryotherapy (or LEEP when not eligible) over a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The benefits of screen-and-treat with an HPV test or VIA, compared to no screening, outweighed the harms, but the reductions in cancer and related mortality were greater with an HPV test when compared to VIA. The availability of HPV testing is resource-dependent and, therefore, the expert panel suggests that an HPV test over VIA be provided where it is available, affordable, implementable, and sustainable over time. This recommendation applies to women regardless of HIV status.

Evidence-to-recommendation table Decision domain Quality of evidence Is there high- or moderate-quality evidence? Judgement Yes  Balance of benefits versus harms and burdens Are you confident that the benefits outweigh the harms and burdens for the recommended strategy? Values and preferences Are you confident about the assumed or identified relative values and are they similar across the target population? Resource implications Is the cost small relative to the net benefits for the recommended strategy? No  Summary of reason for judgement There is high- to moderate-quality evidence for the diagnostic test accuracy data for VIA and the HPV test. There is low- to very-low-quality evidence for the effects of treatment and the natural progression of CIN from observational studies often with inconsistent results across studies. The link between test accuracy data and treatment effects is very uncertain. The benefits of HPV screen-and-treat strategy (reduction in CIN recurrence, cervical cancer, and related mortality) may be greater than VIA, and the harms may be similar. There may also be slightly greater overtreatment and slightly fewer cancers detected with an HPV test compared to VIA. High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was also placed on a reduction in cervical cancer and related mortality versus complications from treatment (e.g. major bleeding or infection requiring hospitalization). Low value was placed on minor infections or bleeding, and the small number of cancers detected at screening or of women overtreated. HPV testing is resource dependent. Where HPV testing is available, affordable and implementable, the overall net benefit over VIA is worth the resources. But where not available, an HPV test may not be worth the benefits.

Yes 

No 

Yes 

No 

Yes 

No 

100

Evidence for an HPV test compared to VIA to screen for CIN2+ in women of HIV-positive status 1. Flowchart of screen-and-treat strategies Asymptomatic HIV-positive women

HPV test

VIA

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Cryo eligible?

Yes

Treat with LEEP

Treat with CKC

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

101

2. Evidence used for decision-making: HPV test compared to VIA Diagnostic test accuracy (data based on women with unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test 95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92)

2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to VIA No. of studies (No. of patients) 5 studies (8921 patients) 5 studies 8921 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies Limitations Indirectness Inconsistency Nonea Noneb Nonec Imprecision None Publication bias Undetected DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test 95 (84 to 98) 26 more Nonea Noneb Seriousc Noned Undetected VIA 69 (11 to 81) Importance CRITICAL

⊕⊕⊕⊕ high

moderate

⊕⊕⊕⊝

756 (648 to 819) 27 fewer

783 (711 to 828)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

Noned

Undetected

moderate

⊕⊕⊕⊝

144 (81 to 252) 27 more

117 (72 to 189)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Nonec

None

Undetected

⊕⊕⊕⊕ high

5 (2 to 16) 26 fewer

31 (19 to 46)

CRITICAL

102 Footnotes: a We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. The decision not to downgrade this was a borderline judgement. b Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. c Estimates of HPV and VIA sensitivity and specificity were variable despite similar cut-off values; and could not be explained by quality of studies. For TP and FN this was a borderline judgement. We downgraded TN and FP and considered this in the context of other factors, in particular, imprecision. d Wide CI for TN and FP that may lead to different decisions depending on which of the confidence limits is assumed.

103 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to VIA Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Outcomes Mortality from cervical cancer1 Cervical cancer incidence2 CIN2+ recurrence3 Undetected CIN2+ (FN) Major bleeding4 Premature delivery5 Infertility6 Major infections7 Minor infections8 Unnecessarily treated (FP) Cancer found at first-time screening9 2052 788 – 212 2239 HPV +/– CKC 318 445 6069 HPV +/–LEEP 460 644 9014 5000 539 602 – 306 1440 144 000 2454 81 649 – 32 1547 1597 724 – 165 1742 HPV +/– cryo 460 644 9014 VIA +/– CKC 1481 1986 26 190 VIA +/– LEEP 1521 2130 28 329 31 000 420 579 – 238 1121 117 000 3168 63 616 – 25 1204 VIA +/– cryo 1521 2130 28 329 No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

104 Footnotes: The colours in the table: : In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

105

3. References to studies included in meta-analysis of diagnostic test accuracy 3.1 References to studies included in meta-analysis of diagnostic test accuracy Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynecology & Obstetrics, 2005, 89(2):120–126. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

3.2 Additional references Denny L et al. Human papillomavirus infection and cervical disease in human immunodeficiency virus-1-infected women. Obstetrics & Gynecology, 2008, 111(6):1380–1387. De Vuyst H et al. HIV, human papillomavirus, and cervical neoplasia and cancer in the era of highly active antiretroviral therapy. European Journal of Cancer Prevention, 2008, 17(6):545–554. De Vuyst H et al. Prevalence and determinants of human papillomavirus infection and cervical lesions in HIV-positive women in Kenya. British Journal of Cancer, 2012, 107(9):1624–1630. Joshi S et al. Screening of cervical neoplasia in HIV-infected women in India. AIDS, 2013, 27(4):607–615. Sankaranarayanan R et al.; Osmanabad District Cervical Screening Study Group. A cluster randomized controlled trial of visual, cytology and human papillomavirus screening for cancer of the cervix in rural India. International Journal of Cancer, 2005, 116(4):617–623. Zhang HY et al. HPV prevalence and cervical intraepithelial neoplasia among HIV-infected women in Yunnan Province, China: a pilot study. Asian Pacific Journal of Cancer Prevention, 2012, 13(1):91–96.

106

Recommendation 3 The expert panel suggests a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with cytology followed by colposcopy (with or without biopsy) and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The reductions in cancer and related mortality were slightly greater with an HPV test only compared to cytology followed by colposcopy. Although there may be overtreatment of populations with high HPV prevalence and consequently more harms, as well as fewer cancers seen at first-time screening with an HPV test, there are greater resources required in cytology programmes due to quality control, training, and waiting time. The addition of colposcopy also requires a second visit. However, in countries where an appropriate/high-quality screening strategy with cytology (referring women with ASCUS or greater results) followed by colposcopy already exists, either an HPV test or cytology followed by colposcopy could be used.

Yes 

No 

Yes 

No 

Yes 

No 

107

Evidence for an HPV test compared to cytology followed by colposcopy to screen for CIN2+ in women of HIV-positive status 1. Flowchart of screen-and-treat strategies Asymptomatic HIV-positive women

HPV test

Cytology

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Cryo eligible?

Colposcopy

Yes

Test + (TP)

Test – (TN)

Treat with LEEP

Treat with CKC

Treat with cryo

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

108

2. Evidence used for decision-making: HPV test compared to cytology (ASCUS) and colposcopic impression Diagnostic test accuracy (data based on women of unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test 94% (95% CI: 89 to 97) 90% (95% CI: 86 to 93) Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS) 70% (95% CI: 57 to 81) 95% (95% CI: 92 to 97) Pooled sensitivity colposcopic impresssion Pooled specificity colposcopic impression 95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

(Reference standard: colposcopy with biopsy when indicated)

2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to cytology (ASCUS) and colposcopic impression No. of studies (No. of patients)a 11 studies (39 050 patients) 11 studies (39 050 patients) Factors that may decrease quality of evidence DTA QoE Effect per 1000 patients/year for pretest probability of 10% Cytology followed by colposcopic impression 67 27 more Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected

Study design Cross-sectional and cohort studies

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

HPV test 94 (89 to 97)

Importance CRITICAL

⊕⊕⊝⊝ low

810 (774 to 837) 64 fewer

874

CRITICAL

11 studies (39 050 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

90 (63 to 126) 64 more

CRITICAL

11 studies (39 050 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

6 (3 to 11) 28 fewer

CRITICAL

109 Footnotes: a This is the number of studies that assessed data for HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for cytology followed by colposcopy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of HPV test, cytology (ASCUS) and colposcopy sensitivity and specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

110 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to cytology (ASCUS) and colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Cyto colp imp +/– CKC 1524 2134 28 124 Cyto colp imp +/– LEEP 1624 2273 30 186 34 000 62 615 – 25 1191 795 612 – 82 867 209 539 – 118 558 26 000 4794 31 558 – 13 599 Cyto colp imp +/– cryo 1624 2273 30 186

HPV +/– CKC 360 504 6843

HPV +/– LEEP 501 701 9757 6000

HPV +/– cryo 501 701 9757

No screen10 4350 6075 79 575 – 0 500 – 0 0 – –

1580 722 – 163 1724

415 578 – 235 1109 90 000 2454

111 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% of women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) HPV test: pooled sensitivity 94% (95% CI: 89 to 97), pooled specificity 90% (95% CI: 86 to 93) Cytology (ASCUS): pooled sensitivity 70% (95% CI: 57 to 81), pooled specificity 95% (95% CI: 92 to 97) Colposcopic impression: pooled sensitivity 95% (95% CI: 86 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

112

3. Evidence used for decision-making: HPV test compared to cytology (ASCUS) and colposcopic impression with biopsy when indicated Diagnostic test accuracy (data based on women with unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test (Reference standard: colposcopy with biopsy when indicated)

94% (95% CI: 89 to 97) 90% (95% CI: 86 to 93)

Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS)

70% (95% CI: 57 to 81) 95% (95% CI: 92 to 97)

3.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to cytology (ASCUS) and colposcopic impression with biopsy when indicated No. of studies (No. of patients)a 11 studies (39 050 patients) 11 studies (39 050 patients) Factors that may decrease quality of evidence DTA QoE Effect per 1000 patients/year for pretest probability of 10% Cytology followed by colposcopy with biopsy 70 24 more Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected

Study design Cross-sectional and cohort studies

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

HPV test 94 (89 to 97)

Importance CRITICAL

⊕⊕⊝⊝ low

810 (774 to 837) 90 fewer

900

CRITICAL

11 studies (39 050 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

90 (63 to 126) 90 more

CRITICAL

11 studies (39 050 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

6 (3 to 11) 24 fewer

CRITICAL

113 Footnotes: a This is the number of studies that assessed data for HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. This was downgraded one level in the context of other factors, in particular indirectness. c Data for cytology followed by colposcopy +/– biopsy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of HPV test and cytology (ASCUS) sensitivity and specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

114 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: Cytology (ASCUS) and colposcopy with biopsy when indicated Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Cytocolp biopsy +/– CKC 1376 1926 25 416 Cytocolp biopsy +/–LEEP 1481 2073 27 586 30 000 62 615 – 25 1191 601 584 – 62 656 158 530 – 90 422 0 4794 24 544 – 9 453 Cytocolp biopsy +/– cryo 1481 2073 27 586

HPV +/– CKC 360 504 6843

HPV +/– LEEP 501 701 9757 6000

HPV +/– cryo 501 701 9757

No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

1580 722 – 163 1724

415 578 – 235 1109 90 000 2454

115 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) HPV test: pooled sensitivity 94% (95% CI: 89 to 97), pooled specificity 90% (95% CI: 86 to 93) Cytology (ASCUS): Pooled sensitivity 70% (95% CI: 57 to 81), pooled specificity 95% (95% CI: 92 to 97) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

116

117 4.2 References to studies included for diagnostic test accuracy of colposcopic impression Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cantor SB et al. Accuracy of colposcopy in the diagnostic setting compared with the screening setting. Obstetrics & Gynecology, 2008, 111(1):7–14. Cremer ML et al. Digital assessment of the reproductive tract versus colposcopy for directing biopsies in women with abnormal Pap smears. Journal of Lower Genital Tract Disease, 2010, 14(1):5–10. Cristoforoni PM et al. Computerized colposcopy: results of a pilot study and analysis of its clinical relevance. Obstetrics & Gynecology, 1995, 85(6):1011–1016. Durdi GS et al. Correlation of colposcopy using Reid colposcopic index with histopathology – a prospective study. Journal of the Turkish German Gynecology Association, 2009, 10(4):205–207. Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651. Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

4.3 References to studies included for diagnostic test accuracy of colposcopic impression Denny L et al. Human papillomavirus infection and cervical disease in human immunodeficiency virus-1-infected women. Obstetrics & Gynecology, 2008, 111(6):1380–1387. De Vuyst H et al. HIV, human papillomavirus, and cervical neoplasia and cancer in the era of highly active antiretroviral therapy. European Journal of Cancer Prevention, 2008, 17(6):545–554.

118 De Vuyst H et al. Prevalence and determinants of human papillomavirus infection and cervical lesions in HIV-positive women in Kenya. British Journal of Cancer, 2012, 107(9):1624–1630. Joshi S et al. Screening of cervical neoplasia in HIV-infected women in India. AIDS, 2013, 27(4):607–615. Sankaranarayanan R et al.; Osmanabad District Cervical Screening Study Group. A cluster randomized controlled trial of visual, cytology and human papillomavirus screening for cancer of the cervix in rural India. International Journal of Cancer, 2005, 116(4):617–623. Zhang HY et al. HPV prevalence and cervical intraepithelial neoplasia among HIV-infected women in Yunnan Province, China: a pilot study. Asian Pacific Journal of Cancer Prevention, 2012, 13(1):91–96.

119

Recommendation 4 The expert panel recommends a strategy of screen with VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with cytology followed by colposcopy (with or without biopsy) and treat with cryotherapy (or LEEP when not eligible) (strong recommendation, ⊕⊝⊝⊝ evidence) Remarks: The benefits and harms of the two screen-and-treat strategies are similar, but there are fewer harms with cytology followed by colposcopy with biopsy when indicated. Despite overtreatment with VIA and fewer cancers detected at first-time screening, more resources are required for cytology programmes with colposcopy (with or without biopsy) due to quality control, training, and waiting time, as well as a second visit. The recommendation for VIA over cytology followed by colposcopy can be applied in countries that are currently considering either strategy or countries that currently have both strategies available. This recommendation applies to women regardless of HIV status.

Yes 

No 

Yes 

No 

High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the resources required and lower value on the harms. Fewer resources are required for VIA. There may be additional resources required in cytology programmes due to increased training of providers, quality control, and waiting time. Colposcopy following cytology also requires a second visit.

Yes 

No 

120

Evidence for VIA compared to cytology followed by colposcopy to screen for CIN2+ in women of HIV-positive status 1. Flowchart of screen-and-treat strategies Asymptomatic HIV-positive women

VIA

Cytology

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Colposcopy Eligible for cryo Not eligible for cryo

Test + (TP & FP) Treat with cryo Treat with CKC Treat with LEEP

Test – (TN & FN)

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

121

2. Evidence used for decision-making: VIA compared to cytology (ASCUS) followed by colposcopic impression Diagnostic test accuracy (data based on women with unknown HIV status) Pooled sensitivity VIA Pooled specificity VIA 77% (95% CI: 65 to 85) 82% (95% CI: 67 to 91) Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS) 84% (95% CI: 76 to 90) 88% (95% CI: 79 to 93) Pooled sensitivity colposcopic impression Pooled specificity colposcopic impression 95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

(Reference standard: colposcopy with biopsy when indicated)

2.1 Diagnostic Test Accuracy (DTA) evidence profile: VIA compared to cytology (ASCUS) followed by colposcopic impression No. of studies (No. of patients) a 11 studies (12 089 patients) 11 studies (12 089 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 10% Cytology followed by colposcopic impression 80 3 fewer Seriousb Nonec Seriousd Nonee Undetected

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

VIA 77 (65 to 85)

Importance CRITICAL

⊕⊕⊝⊝ low

738 (603 to 819) 99 fewer

837

CRITICAL

11 studies (12 089 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

162 (81 to 297) 99 more

CRITICAL

11 studies (12 089 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

23 (15 to 35) 3 more

CRITICAL

122 Footnotes: a This is the number of studies that assessed data for HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for cytology followed by colposcopy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of VIA, cytology (ASCUS) and colposcopy sensitivity and specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

123 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: VIA compared to cytology (ASCUS) followed by colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Cytocolp imp +/– CKC 961 1346 17832 Cytocolp imp +/–LEEP 1080 1513 20 306 20 000 81 649 – 32 1547 1223 672 – 126 1334 321 561 – 182 858 63 000 4794 48 589 – 19 922 Cytocolp imp +/– cryo 1080 1513 20 306

VIA +/– CKC 1080 1512 19 999

VIA +/– LEEP 1195 1673 22 386 23 000

VIA +/– cryo 1195 1673 22 386

No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

2052 788 – 212 2239

539 602 – 306 1440 162 000 3168

124 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) VIA: pooled sensitivity 77% (95% CI: 66 to 85), pooled specificity 82% (95% CI: 67 to 91) Cytology (ASCUS): Pooled sensitivity 84% (95% CI: 76 to 90), pooled specificity 88% (95% CI: 79 to 93) Colposcopic impression: Pooled sensitivity 95% (95% CI: 86 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

125

3. Evidence used for decision-making: VIA compared to cytology (ASCUS) followed by colposcopic impression and biopsy when indicated Diagnostic test accuracy (data based on women of unknown HIV status) Pooled sensitivity VIA Pooled specificity VIA (Reference standard: colposcopy with biopsy when indicated)

77% (95% CI: 66 to 85) 82% (95% CI: 67 to 91)

Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS)

84% (95% CI: 76 to 90) 88% (95% CI: 79 to 93)

3.1 Diagnostic test accuracy (DTA) evidence profile: VIA compared to cytology (ASCUS) followed by colposcopic impression and biopsy when indicated No. of studies (No. of patients) a 11 studies (12 089 patients) 11 studies (12 089 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 10% Cytology followed by colposcopy with biopsy 84 7 fewer Seriousb Nonec Seriousd Nonee Undetected

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

VIA 77 (65 to 85)

Importance CRITICAL

⊕⊕⊝⊝ low

738 (603 to 819) 162 fewer

900

CRITICAL

11 studies (12 089 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

162 (81 to 297) 162 more

CRITICAL

11 studies (12 089 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

23 (15 to 35) 7 more

CRITICAL

126 Footnotes: a This is the number of studies that assessed data for HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. This was downgraded one level in the context of other factors, in particular indirectness. c Data for cytology followed by colposcopy +/– biopsy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of VIA and cytology (ASCUS) sensitivity and specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

127 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: VIA compared to cytology (ASCUS) followed by colposcopic impression and biopsy when indicated Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) Cyto colp biopsy +/– CKC 783 1097 14 582 Cyto colp biopsy +/–LEEP 908 1273 17 186 16 000 81 649 – 32 1547 721 601 – 75 787 190 536 – 107 506 0 4794 28 553 – 11 544 Cyto colp biopsy +/– cryo 909 1273 17 186

VIA +/– CKC 1080 1512 19 999

VIA +/– LEEP 1195 1673 22 386 23 000

VIA +/– cryo 1195 1673 22 386

No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

2052 788 – 212 2239

539 602 – 306 1440 162 000 3168

128 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) VIA: pooled sensitivity 77% (95% CI: 66 to 85), pooled specificity 83% (95% CI: 68 to 92) Cytology (ASCUS): Pooled sensitivity 84% (95% CI: 76 to 90), pooled specificity 88% (95% CI: 79 to 93) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

129

4. References 4.1 References to studies included in meta-analysis of diagnostic test accuracy Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cremer M et al. Adequacy of visual inspection with acetic acid in women of advancing age. International Journal of Gynecology & Obstetrics, 2011, 113(1):68–71. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynecology & Obstetrics, 2005, 89(2):120–126. Elit L et al. Assessment of 2 cervical screening methods in Mongolia: cervical cytology and visual inspection with acetic acid. Journal of Lower Genital Tract Disease, 2006, 10(2):83–88. Ghaemmaghami F et al. Visual inspection with acetic acid as a feasible screening test for cervical neoplasia in Iran. International Journal of Gynecological Cancer, 2004, 14(3):465–469. Goel A et al. Visual inspection of the cervix with acetic acid for cervical intraepithelial lesions. International Journal of Gynecology & Obstetrics, 2005, 88(1):25–30. Hedge D et al. Diagnostic value of acetic acid comparing with conventional Pap smear in the detection of colposcopic biopsy-proved CIN. Journal of Cancer Research & Therapeutics, 2011, 7(4):454–458. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Sahasrabuddhe VV et al. Comparison of visual inspection with acetic acid and cervical cytology to detect high-grade cervical neoplasia among HIV-infected women in India. International Journal of Cancer, 2012, 130(1):234–240. Sankaranarayanan R et al. Test characteristics of visual inspection with 4% acetic acid (VIA) and Lugol's iodine (VILI) in cervical cancer screening in Kerala, India. International Journal of Cancer, 2003, 106(3):404–408. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

130 4.2 References to studies included for diagnostic test accuracy of colposcopic impression Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cantor SB et al. Accuracy of colposcopy in the diagnostic setting compared with the screening setting. Obstetrics & Gynecology, 2008, 111(1):7–14. Cremer ML et al. Digital assessment of the reproductive tract versus colposcopy for directing biopsies in women with abnormal Pap smears. Journal of Lower Genital Tract Disease, 2010, 14(1):5–10. Cristoforoni PM et al. Computerized colposcopy: results of a pilot study and analysis of its clinical relevance. Obstetrics & Gynecology, 1995, 85(6):1011–1016. Durdi GS et al. Correlation of colposcopy using Reid colposcopic index with histopathology – a prospective study. Journal of the Turkish German Gynecology Association, 2009, 10(4):205–207. Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651. Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: Is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

131 De Vuyst H et al. Prevalence and determinants of human papillomavirus infection and cervical lesions in HIV-positive women in Kenya. British Journal of Cancer, 2012, 107(9):1624–1630. Joshi S et al. Screening of cervical neoplasia in HIV-infected women in India. AIDS, 2013, 27(4):607–615. Sankaranarayanan R et al.; Osmanabad District Cervical Screening Study Group. A cluster randomized controlled trial of visual, cytology and human papillomavirus screening for cancer of the cervix in rural India. International Journal of Cancer, 2005, 116(4):617–623. Zhang HY et al. HPV prevalence and cervical intraepithelial neoplasia among HIV-infected women in Yunnan Province, China: a pilot study. Asian Pacific Journal of Cancer Prevention, 2012, 13(1):91–96.

132

Evidence-to-recommendation table Decision domain Quality of evidence Is there high- or moderate-quality evidence? Judgement Yes  Balance of benefits versus harms and burdens Are you confident that the benefits outweigh the harms and burdens for the recommended strategy? Values and preferences Are you confident about the assumed or identified relative values and are they similar across the target population? Resource implications Is the cost small relative to the net benefits for the recommended strategy? No  Summary of reason for judgement There is low-quality evidence for the diagnostic test accuracy of HPV test followed by colposcopy and we did not have a direct comparison of this triage test to HPV test alone. There is low- to very-low-quality evidence for the effects of treatment and the natural progression of CIN from observational studies often with inconsistent results across studies. The link between test accuracy data and treatment effects is very uncertain. The benefits of HPV test followed by colposcopy and HPV test alone may be similar. However, there were greater harms with HPV test alone (due to overtreatment with HPV test alone). There may also be slightly fewer cancers detected with HPV test followed by colposcopy. High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the resources required and lower value on the harms. There may be additional resources required with the addition of colposcopy (with or without biopsy), there are more resource implications with colposcopy due to increased training of providers, quality control, waiting time, and potential for more women lost to follow-up. The addition of colposcopy to HPV test would also require a second visit.

Yes 

No 

Yes 

No 

Yes 

No 

133

Evidence for an HPV test compared to an HPV test followed by colposcopy to screen for CIN2+ in women of HIV-positive status 1. Flowchart of screen-and-treat strategies Asymptomatic HIV-positive women

HPV test

HPV

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Cryo eligible?

Colposcopy

Yes

Test + (TP & FP)

Test – (TN & FN)

Treat with LEEP

Treat with CKC

Treat with cryo

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

134

93% (95% CI: 87 to 96) 88% (95% CI: 82 to 91)

Pooled sensitivity colposcopic impression Pooled specificity colposcopic impression

95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to HPV test followed by colposcopic impression No. of studies (No. of patients)a 15 studies (45 783 patients) 15 studies (45 783 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test followed by colposcopic impression 88 5 more Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

HPV test 93 (87 to 96)

Importance CRITICAL

⊕⊕⊝⊝ low

792 (738 to 819) 55 fewer

837

CRITICAL

15 studies (45 783 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

108 (81 to 162) 55 more

CRITICAL

15 studies (45 783 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

7 (4 to 13) 5 fewer

CRITICAL

135 Footnotes: a This is the number of studies that assessed data for HPV test. b We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by colposcopy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. e Few participants contributed to colposcopy data. Therefore there were wide confidence intervals for colposcopy specificity, which may lead to different decisions depending on which confidence limits are assumed.

136 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to HPV test followed by colposcopy Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPV colp imp +/– CKC 599 839 11 215 HPV colp imp +/–LEEP 731 1024 13 954 12 000 68 626 – 27 1301 1296 682 – 134 1414 341 564 – 193 910 63 000 3545 51 594 – 20 977 HPV colp imp +/– cryo 731 1024 13 954

HPV +/– CKC 402 563 7617

HPV +/– LEEP 541 758 10 500 7000

HPV +/– cryo 541 758 10 500

No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

1726 742 – 179 1883

454 585 – 257 1211 108 000 2454

137 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) HPV test: pooled sensitivity 93% (95% CI: 87 to 96), pooled specificity 88% (95% CI: 82 to 91) Colposcopic impression: pooled sensitivity 95% (95% CI: 82 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

138

3. Evidence used for decision-making: HPV test compared to HPV test followed by colposcopy with biopsy when indicated Diagnostic test accuracy Pooled sensitivity HPV test Pooled specificity HPV test (Reference standard: colposcopy with biopsy when indicated)

93% (95% CI: 87 to 96) 88% (95% CI: 82 to 91)

3.1 Diagnostic test accuracy (DTA) evidence profile: HPV test compared to HPV test followed by colposcopic impression No. of studies (No. of patients)a 15 studies (45 783 patients) 15 studies (45 783 patients) Factors that may decrease quality of evidence DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test followed by colposcopy with biopsy 93 0 Cross-sectional and cohort studies Seriousb Nonec Seriousd Nonee Undetected

Study design Cross-sectional and cohort studies

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

HPV test 93 (87 to 96)

Importance CRITICAL

⊕⊕⊝⊝ low

792 (738 to 819) 108 fewer

900

CRITICAL

15 studies (45 783 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

108 (81 to 162) 108 more

CRITICAL

15 studies (45 783 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

7 (4 to 13) 0

CRITICAL

139 Footnotes: a This is the number of studies that assessed data for HPV test. b We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by colposcopy +/– biopsy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. e Few participants contributed to colposcopy data. Therefore there were wide confidence intervals for colposcopy specificity, which may lead to different decisions depending on which confidence limits are assumed.

140 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test compared to HPV test followed by colposcopy with biopsy when indicated Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPV colp biopsy +/– CKC 402 563 7617 HPV colp biopsy +/–LEEP 541 758 10 500 7000 68 626 – 27 1301 798 612 – 83 871 210 540 – 119 561 0 3545 32 558 – 13 602 HPV colp biopsy +/– cryo 541 758 10 500

HPV +/– CKC 402 563 7617

HPV +/– LEEP 541 758 10 500 7000

HPV +/– cryo 541 758 10 500

No screen10 4350 6075 79 575 – 0 500 – 0 0 – –

1726 742 – 179 1883

454 585 – 257 1211 108 000 2454

141 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) HPV test: pooled sensitivity 93% (95% CI: 87 to 96), pooled specificity 88% (95% CI: 82 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

142

4. References 4.1 References to studies included in meta-analysis of diagnostic test accuracy Agorastos T et al. Human papillomavirus testing for primary screening in women at low risk of developing cervical cancer. The Greek experience. Gynecologic Oncology, 2005, 96(3):714–720. Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Bigras G, De Marval F. The probability for a Pap test to be abnormal is directly proportional to HPV viral load: Results from a Swiss study comparing HPV testing and liquid-based cytology to detect cervical cancer precursors in 13 842 women. British Journal of Cancer, 2005, 93(5):575–581. Blumenthal PD et al. Adjunctive testing for cervical cancer in low resource settings with visual inspection, HPV, and the Pap smear. International Journal of Gynecology & Obstetrics, 2001, 72(1):47–53. Cardenas-Turanzas M et al. The performance of human papillomavirus high-risk DNA testing in the screening and diagnostic settings. Cancer Epidemiology Biomarkers and Prevention, 2008, 17(10):2865–2871. de Cremoux P et al. Efficiency of the hybrid capture 2 HPV DNA test in cervical cancer screening. A study by the French Society of Clinical Cytology. American Journal of Clinical Pathology, 2003, 120(4):492–429. Depuydt CE et al. BD-ProExC as adjunct molecular marker for improved detection of CIN2+ after HPV primary screening. Cancer Epidemiology Biomarkers and Prevention, 2011, 20(4):628–637. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynecology & Obstetrics, 2005, 89(2):120–126. Hovland S et al. A comprehensive evaluation of the accuracy of cervical pre-cancer detection methods in a high-risk area in East Congo. British Journal of Cancer, 2010, 102(6):957–965. Mahmud SM et al. Comparison of human papillomavirus testing and cytology for cervical cancer screening in a primary health care setting in the Democratic Republic of the Congo. Gynecologic Oncology, 2012, 124(2):286–291. Monsonego J, Hudgens MG, Zerat L, Zerat JC, Syrjanen K, Halfon P et al. Evaluation of oncogenic human papillomavirus RNA and DNA tests with liquid-based cytology in primary cervical cancer screening: the FASE study. International Journal of Cancer, 2011, 129(3):691–701.

143 Petry KU et al. Inclusion of HPV testing in routine cervical cancer screening for women above 29 years in Germany: results for 8466 patients. British Journal of Cancer, 2003, 88(10):1570–1577. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352

4.2 References to studies included for diagnostic test accuracy of colposcopic impression Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cantor SB et al. Accuracy of colposcopy in the diagnostic setting compared with the screening setting. Obstetrics & Gynecology, 2008, 111(1):7–14. Cremer ML et al. Digital assessment of the reproductive tract versus colposcopy for directing biopsies in women with abnormal Pap smears. Journal of Lower Genital Tract Disease, 2010, 14(1):5–10. Cristoforoni PM et al. Computerized colposcopy: results of a pilot study and analysis of its clinical relevance. Obstetrics & Gynecology, 1995, 85(6):1011–1016. Durdi GS et al. Correlation of colposcopy using Reid colposcopic index with histopathology – a prospective study. Journal of the Turkish German Gynecology Association, 2009, 10(4):205–207. Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651.

144 Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: Is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

4.3 Additional references Denny L et al. Human papillomavirus infection and cervical disease in human immunodeficiency virus-1-infected women. Obstetrics & Gynecology, 2008, 111(6):1380–1387. De Vuyst H et al. HIV, human papillomavirus, and cervical neoplasia and cancer in the era of highly active antiretroviral therapy. European Journal of Cancer Prevention, 2008, 17(6):545–554. De Vuyst H et al. Prevalence and determinants of human papillomavirus infection and cervical lesions in HIV-positive women in Kenya. British Journal of Cancer, 2012, 107(9):1624–1630. Joshi S et al. Screening of cervical neoplasia in HIV-infected women in India. AIDS, 2013, 27(4):607–615. Sankaranarayanan R et al.; Osmanabad District Cervical Screening Study Group. A cluster randomized controlled trial of visual, cytology and human papillomavirus screening for cancer of the cervix in rural India. International Journal of Cancer, 2005, 116(4):617–623. Zhang HY et al. HPV prevalence and cervical intraepithelial neoplasia among HIV-infected women in Yunnan Province, China: a pilot study. Asian Pacific Journal of Cancer Prevention, 2012, 13(1):91–96.

145

Recommendation 6 The expert panel suggests either a strategy of screen with an HPV test followed by VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) or a strategy of screen with an HPV test and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The reductions in cancer and related mortality were greater with an HPV test used as a single screening test than with an HPV test followed by VIA, and this reduction was even greater in women of HIV-positive status. However, there may be overtreatment, and thus potentially greater harms with screen-and-treat when using an HPV test as a single test. There is also some uncertainty about the effects of an HPV test followed by VIA and how VIA performs after a positive HPV test because there was no direct evidence about this strategy. There is also the potential for additional resources that are required to refer women for VIA testing after a positive HPV test, the need for a second visit to perform VIA, and increased training to perform both tests. For these reasons, the recommendation is for either an HPV test followed by VIA or an HPV test only, and it is conditional. It is to be noted that benefits are more pronounced compared to 'harms' in women of HIV-positive status when using an HPV test only.

Evidence-to-recommendation table Decision domain Quality of evidence Is there high- or moderate-quality evidence? Judgement Yes  Balance of benefits versus harms and burdens Are you confident that the benefits outweigh the harms and burdens for the recommended strategy? No  Summary of reason for judgement There is low-quality evidence for the diagnostic test accuracy of HPV test followed by VIA and compared to HPV test alone. There is low- to very-low-quality evidence for the effects of treatment and the natural progression of CIN from observational studies often with inconsistent results across studies. The link between test accuracy data and treatment effects is very uncertain. There may be fewer major harms with HPV test followed by VIA than with HPV test alone due to less overtreatment. There may also be slightly greater cancers detected with HPV test followed by VIA than with HPV test alone. However, there may be slightly greater CIN recurrence, cervical cancer, and related mortality with HPV test followed by VIA. In women of HIV-positive status, there were still fewer harms, less overtreatment and greater cancers detected at first-time screening. However, there was even greater CIN recurrence, cervical cancer and related mortality with HPV test followed by VIA in women of HIV-positive status than in women of unknown status. High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was also placed on reducing overtreatment and resulting complications, and resource use. Greater resources may be required for HPV test followed by VIA due to adding on an additional test. However, there is less overtreatment (fewer treatments provided) and fewer complications requiring hospitalization.

Yes 

No 

Yes 

No 

Yes 

No 

146

Evidence for an HPV test followed by VIA compared to an HPV test to screen for CIN2+ in women of HIV-positive status 1. Flowchart of screen-and-treat strategies Asymptomatic HIV-positive women

HPV test

HPV

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

VIA

Cryo eligible?

Test + (TP & FP)

Test – (TN & FN)

Yes

Eligible for cryo

Not eligible for cryo

Treat with LEEP

Treat with CKC

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

147

2. Evidence used for decision-making: HPV test followed by VIA compared to HIV Diagnostic test accuracy (data based on women with unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test 95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92)

2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to HIV No. of studies (No. of patients) 5 studies (8921 patients) 5 studies (8921 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies Limitations Indirectness Inconsistency Nonea Noneb Seriousc Imprecision None Publication bias Undetected DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test followed by VIA 66 29 fewer Nonea Noneb Seriousc Noned Undetected HPV test 95 (84 to 98) Importance CRITICAL

moderate

⊕⊕⊕⊝

moderate

⊕⊕⊕⊝

881 125 more

756 (648 to 819)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

Noned

Undetected

moderate

⊕⊕⊕⊝

19 125 fewer

144 (81 to 252)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

None

Undetected

moderate

⊕⊕⊕⊝

34 29 more

5 (2 to 16)

CRITICAL

148 Footnotes: a We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. The decision to downgrade was a borderline judgement and was considered in the context of other factors. b Data for HPV test followed by VIA were calculated based on sensitivity and specificity of the two tests. Direct data were unavailable. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. c Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. d Wide CI for HPV test sensitivity and VIA specificity, and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

149 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVVIA +/– CKC 1564 2190 28 859 HPVVIA +/– LEEP 1662 2327 30 891 34 000 723 602 – 75 789 190 536 – 108 508 19 000 3168 29 553 – 11 545 2052 788 – 212 2239 HPVVIA +/– cryo 1662 2327 30 891

HPV +/– CKC 318 445 6069

HPV +/– LEEP 460 644 9014 5000 539 602 – 306 1440 144 000 2454

HPV +/– cryo 460 644 9014

No screen10 4350 6075 79 575 –

81 649 – 32 1547

0 500 – 0 0 – 0

150 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

6 7

151

3. References to studies included in meta-analysis of diagnostic test accuracy 3.1 References to studies included in meta-analysis of diagnostic test accuracy Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecology Oncology, 2001, 83(2):439–444. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynecology & Obstetrics, 2005, 89(2):120–126. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

3.2 Additional references Denny L et al. Human papillomavirus infection and cervical disease in human immunodeficiency virus-1-infected women. Obstetrics & Gynecology, 2008, 111(6):1380–1387. De Vuyst H et al. HIV, human papillomavirus, and cervical neoplasia and cancer in the era of highly active antiretroviral therapy. European Journal of Cancer Prevention, 2008, 17(6):545–554. De Vuyst H et al. Prevalence and determinants of human papillomavirus infection and cervical lesions in HIV-positive women in Kenya. British Journal of Cancer, 2012, 107(9):1624–1630. Joshi S et al. Screening of cervical neoplasia in HIV-infected women in India, AIDS, 2013, 27(4):607–615. Sankaranarayanan R et al.; Osmanabad District Cervical Screening Study Group. A cluster randomized controlled trial of visual, cytology and human papillomavirus screening for cancer of the cervix in rural India. International Journal of Cancer, 2005, 116(4):617–623. Zhang HY et al. HPV prevalence and cervical intraepithelial neoplasia among HIV-infected women in Yunnan Province, China: a pilot study. Asian Pacific Journal of Cancer Prevention, 2012, 13(1):91–96.

152

Recommendation 7 The expert panel suggests a strategy of screen with an HPV test followed by VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with VIA and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The reductions in cancer and related mortality with an HPV test followed by VIA or with VIA alone outweighed the harms. However, the harms may be greater when using VIA only, which is likely due to overtreatment. Although, a slightly larger number of cancers may be detected on initial screen with VIA only. This recommendation is conditional due to the uncertain costs of providing the sequence of two tests (HPV test followed by VIA) over the single VIA test. In countries where HPV test is not available, we suggest screening with VIA only. This recommendation applies to women regardless of HIV status.

Yes 

No 

Yes 

No 

High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the greater number of complications and the number of women overtreated. Greater resources with overtreatment with VIA alone. However there may be additional resources required to refer women for VIA testing after a positive HPV test, the need for a second visit, and increased training to perform both tests.

Yes 

No 

153

Evidence for an HPV test followed by VIA compared to VIA to screen for CIN2+ in women of HIV-positive status 1. Flowchart of screen-and-treat strategies Asymptomatic HIV-positive women

HPV test

VIA

Test + (TP & FP)

Test – (TN & FN)

Suspect cervical cancer

Test + (TP & FP)

Test – (TN & FN)

VIA

Eligible for cryo

Not eligible for cryo

Test + (TP & FP)

Test – (TN & FN) Treat with cryo Treat with CKC Treat with LEEP

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

154

2. Evidence used for decision-making: HPV test followed by VIA compared to VIA Diagnostic test accuracy (data based on women with unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test 95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92)

2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to VIA No. of studies (No. of patients) 5 studies (8921 patients) 5 studies (8921 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies Limitations Indirectness Inconsistency Nonea Noneb Seriousc Imprecision Noned Publication bias Undetected DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test followed by VIA 66 3 fewer Nonea Noneb Seriousc Noned Undetected VIA 69 (11 to 81) Importance CRITICAL

moderate

⊕⊕⊕⊝

moderate

⊕⊕⊕⊝

881 98 more

783 (711 to 828)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

Noned

Undetected

moderate

⊕⊕⊕⊝

19 98 fewer

117 (72 to 189)

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Nonea

Noneb

Seriousc

Noned

Undetected

moderate

⊕⊕⊕⊝

34 3 more

31 (19 to 46)

CRITICAL

155 Footnotes: a We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. The decision to downgrade was a borderline judgement and was considered in the context of other factors. b Data for HPV test followed by VIA were calculated based on sensitivity and specificity of the two tests. Direct data were unavailable. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. c Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. d Wide CI for HPV test sensitivity and VIA specificity, and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

156 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVVIA +/– CKC 1564 2190 28 859 HPVVIA +/– LEEP 1662 2327 30 891 34 000 723 602 – 75 789 190 536 – 108 508 19 000 2454 29 553 – 11 545 1597 724 – 165 1742 HPVVIA +/– cryo 1662 2327 30 891

VIA +/– CKC 1481 1986 26 190

VIA +/–LEEP 1521 2130 28 329 31 000 420 579 – 238 1121 117 000 3168

VIA +/– cryo 1521 2130 28 329

No screen10 4350 6075 79 575 –

63 616 – 25 1204

0 500 – 0 0 – 0

157 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

6 7

158

3. References 3.1 References to studies included in meta-analysis of diagnostic test accuracy Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynecology & Obstetrics, 2005, 89(2):120–126. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

159

Yes 

No 

Yes 

No 

High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the resources required. Fewer resources may be required for HPV test followed by VIA as there may be additional resources required in cytology programmes due to increased training of providers, quality control, and waiting time. Colposcopy following cytology also requires a second visit.

Yes 

No 

160

Evidence for an HPV test followed by VIA compared to cytology followed by colposcopy to screen for CIN2+ in women of HIV-positive status 1. Flowchart of screen-and-treat strategies Asymptomatic HIV-positive women

HPV test

Cytology

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

VIA

Colposcopy

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Eligible for cryo

Not eligible for cryo

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

161

2. Evidence used for decision-making: HPV test followed by VIA compared to cytology (ASCUS) and colposcopic impression Diagnostic test accuracy (data based on women with unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS) (Reference standard: colposcopy with biopsy when indicated)

94% (95% CI: 89 to 97) 90% (95% CI: 86 to 93) 70% (95% CI: 57 to 81) 95% (95% CI: 92 to 97)

Pooled sensitivity VIA Pooled specificity VIA Pooled sensitivity colposcopic impression Pooled specificity colposcopic impression

69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92) 95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

162 2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to cytology (ASCUS) and colposcopic impression No. of studies (No. of patients)a 14 studies (34 584 patients) 14 studies (34 584 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test followed by VIA 65 2 fewer Seriousb Nonec Seriousd Nonee Undetected Cytology followed by colposcopic impression 67

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

Importance CRITICAL

⊕⊕⊝⊝ low

888 14 more

874

CRITICAL

14 studies (34 584 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

12 14 fewer

CRITICAL

14 studies (34 584 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

35 1 more

CRITICAL

Footnotes: This is the number of studies that assessed DTA data for: 1. HPV test and VIA, and 2. HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by VIA and cytology followed by colposcopy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of HPV test, VIA, cytology (ASCUS) and colposcopy sensitivity/specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of HPV test followed by VIA and cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

163 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA compared to cytology (ASCUS) and colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVVIA +/– CKC 1594 2231 29 393 HPVVIA + /– LEEP 1691 2367 31 404 35 000 657 612 – 82 867 173 539 – 118 558 12 000 3168 26 558 – 13 599 795 592 – 68 717 HPVVIA +/– cryo 1691 2367 31 404 Cytocolp imp +/– CKC 1524 2134 28 124 Cytocolp imp +/– LEEP 1624 1752 30 186 34 000 209 533 – 98 461 26 000 4794 31 548 – 10 496 Cytocolp imp +/– cryo 1624 2273 30 186 No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

164 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) VIA: Pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) Cytology (ASCUS): Pooled sensitivity 70% (95% CI: 57 to 81), pooled specificity 95% (95% CI: 92 to 97) Colposcopy: Pooled sensitivity 95% (95% CI: 86 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very low quality evidence for treatment effects and natural progression/history data 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

165

3. Evidence used for decision-making: HPV test followed by VIA compared to cytology (ASCUS) and colposcopic impression and biopsy when indicated Diagnostic test accuracy (data based on women with unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test 94% (95% CI: 89 to 97) 90% (95% CI: 86 to 93) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92) Pooled sensitivity cytology (ASCUS) Pooled specificity cytology (ASCUS) 70% (95% CI: 57 to 81) 95% (95% CI: 92 to 97)

(Reference standard: colposcopy with biopsy when indicated)

166 3.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to cytology (ASCUS) and colposcopy with biopsy when indicated No. of studies (No. of patients)a 14 studies (34 584 patients) 14 studies (34 584 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test followed by VIA 65 5 fewer Seriousb Nonec Seriousd Nonee Undetected Cytology followed by colposcopy with biopsy 70

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

Importance CRITICAL

⊕⊕⊝⊝ low

888 12 fewer

900

CRITICAL

14 studies (34 584 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

12 12 more

CRITICAL

14 studies (34 584 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

35 5 more

CRITICAL

Footnotes: a This is the number of studies that assessed DTA data for: 1. HPV test and VIA, and 2. HPV test and cytology. b We used QUADAS to assess risk of bias. Half of studies only performed one biopsy of an abnormal lesion and had unclear blinding of tests. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by VIA and cytology followed by colposcopy +/– biopsy were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of HPV test, VIA and cytology (ASCUS) sensitivity/specificity were variable despite similar cut-off values; inconsistency was not explained by quality of studies. This was downgraded one level in the context of other factors, in particular imprecision. e Wide CI for sensitivity and specificity of HPV test followed by VIA and cytology followed by colposcopy and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

167 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA and cytology (ASCUS) and colposcopy with biopsy when indicated Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVVIA +/– CKC 1594 2231 29 393 HPVVIA +/– LEEP 1691 2367 31 404 35 000 657 612 – 82 867 173 539 – 118 558 12 000 3168 26 558 – 13 599 601 584 – 62 656 HPVVIA +/– cryo 1691 2367 31 404 Cytocolp biopsy +/– CKC 1376 1926 25 416 Cytocolp biopsy +/–LEEP 1481 2073 27 586 30 000 158 530 – 90 422 0 3545 24 544 – 9 453 Cytocolp biopsy +/– cryo 1481 2073 27 586

No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

168 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) HPV test: Pooled sensitivity94% (95% CI: 89 to 97), pooled specificity 90% (95% CI: 86 to 93) VIA: Pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) Cytology (ASCUS): Pooled sensitivity 70% (95% CI: 57 to 81), pooled specificity 95% (95% CI: 92 to 97) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

169

4. References 4.1 References to studies included in meta-analysis of diagnostic test accuracy Agorastos T et al. Human papillomavirus testing for primary screening in women at low risk of developing cervical cancer. The Greek experience. Gynecologic Oncology, 2005, 96(3):714–720. Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Bigras G, De Marval F. The probability for a Pap test to be abnormal is directly proportional to HPV viral load: Results from a Swiss study comparing HPV testing and liquid-based cytology to detect cervical cancer precursors in 13 842 women. British Journal of Cancer, 2005, 93(5):575–581. Cardenas-Turanzas M et al. The performance of human papillomavirus high-risk DNA testing in the screening and diagnostic settings. Cancer Epidemiology Biomarkers and Prevention, 2008, 17(10):2865–2871. de Cremoux P et al. Efficiency of the hybrid capture 2 HPV DNA test in cervical cancer screening. A study by the French Society of Clinical Cytology. American Journal of Clinical Pathology, 2003, 120(4):492–499. Depuydt CE et al. BD-ProExC as adjunct molecular marker for improved detection of CIN2+ after HPV primary screening. Cancer Epidemiology Biomarkers and Prevention, 2011, 20(4):628–637. De Vuyst H et al. Comparison of Pap smear, visual inspection with acetic acid, human papillomavirus DNA–PCR testing and cervicography. International Journal of Gynecology & Obstetrics, 2005, 89(2):120–126.Hovland S et al. A comprehensive evaluation of the accuracy of cervical pre-cancer detection methods in a high-risk area in East Congo. British Journal of Cancer, 2010, 102(6):957–965. Mahmud SM et al. Comparison of human papillomavirus testing and cytology for cervical cancer screening in a primary health care setting in the Democratic Republic of the Congo. Gynecologic Oncology, 2012, 124(2):286–291. Monsonego J et al. Evaluation of oncogenic human papillomavirus RNA and DNA tests with liquid-based cytology in primary cervical cancer screening: the FASE study. International Journal of Cancer, 2011, 129(3):691–701. Pan Q et al. A thin-layer, liquid-based Pap test for mass screening in an area of China with a high incidence of cervical carcinoma: a cross-sectional, comparative study. Acta Cytologica, 2003, 47(1):45–50.

170 Petry KU et al. Inclusion of HPV testing in routine cervical cancer screening for women above 29 years in Germany: results for 8466 patients. British Journal of Cancer, 2003, 88(10):1570–1577. Qiao YL et al. A new HPV–DNA test for cervical-cancer screening in developing regions: a cross-sectional study of clinical accuracy in rural China. Lancet Oncology, 2008, 9(10):929–936. Shastri SS et al. Concurrent evaluation of visual, cytological and HPV testing as screening methods for the early detection of cervical neoplasia in Mumbai, India. Bulletin of the World Health Organization, 2005, 83(3):186–194. Sodhani P et al. Test characteristics of various screening modalities for cervical cancer: a feasibility study to develop an alternative strategy for resource-limited settings. Cytopathology, 2006, 17(6):348–352.

4.2 References to studies included for diagnostic test accuracy of colposcopic impression Belinson J et al. Shanxi Province Cervical Cancer Screening Study: a cross-sectional comparative trial of multiple techniques to detect cervical neoplasia. Gynecologic Oncology, 2001, 83(2):439–444. Cantor SB et al. Accuracy of colposcopy in the diagnostic setting compared with the screening setting. Obstetrics & Gynecology, 2008, 111(1):7–14. Cremer ML et al. Digital assessment of the reproductive tract versus colposcopy for directing biopsies in women with abnormal Pap smears. Journal of Lower Genital Tract Disease, 2010, 14(1):5–10. Cristoforoni PM et al. Computerized colposcopy: results of a pilot study and analysis of its clinical relevance. Obstetrics & Gynecology, 1995, 85(6):1011–1016. Durdi GS et al. Correlation of colposcopy using Reid colposcopic index with histopathology – a prospective study. Journal of the Turkish German Gynecology Association, 2009, 10(4):205–207. Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549.

171 Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651. Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

4.3 Additional references Denny L et al. Human papillomavirus infection and cervical disease in human immunodeficiency virus-1-infected women. Obstetrics & Gynecology, 2008, 111(6):1380–1387. De Vuyst H, Lillo F, Broutet N, Smith JS. HIV, human papillomavirus, and cervical neoplasia and cancer in the era of highly active antiretroviral therapy. European Journal of Cancer Prevention, 2008, 17(6):545–554. De Vuyst H et al. Prevalence and determinants of human papillomavirus infection and cervical lesions in HIV-positive women in Kenya. British Journal of Cancer, 2012, 107(9):1624–1630. Joshi S et al. Screening of cervical neoplasia in HIV-infected women in Maharashtra, India. AIDS, 2013, 27(4):607–615. Sankaranarayanan R et al.; Osmanabad District Cervical Screening Study Group. A cluster randomized controlled trial of visual, cytology and human papillomavirus screening for cancer of the cervix in rural India. International Journal of Cancer, 2005, 116(4):617–623. Zhang HY et al. HPV prevalence and cervical intraepithelial neoplasia among HIV-infected women in Yunnan Province, China: a pilot study. Asian Pacific Journal of Cancer Prevention, 2012, 13(1):91–96..

172

Recommendation 9 The expert panel suggests a strategy of screen with an HPV test followed by VIA and treat with cryotherapy (or LEEP when not eligible for cryotherapy) over a strategy of screen with HPV test followed by colposcopy (with or without biopsy) and treat with cryotherapy (or LEEP when not eligible) (conditional recommendation, ⊕⊝⊝⊝ evidence) Remarks: The reductions in cancer and related mortality of screen-and-treat with an HPV test followed by colposcopy (with or without biopsy) may be slightly greater compared to an HPV test followed by VIA. The panel agreed that the benefits of either strategy outweigh the harms and costs; however, the difference in costs between the strategies is uncertain. There may be more resource implications with colposcopy due to increased training of providers, quality control, waiting time, and the potential for more women to be lost to follow-up. It is also unclear whether women would perceive a difference between VIA and colposcopy; however, a biopsy during colposcopy may be less acceptable than VIA. This recommendation applies to women regardless of HIV status.

Evidence-to-recommendation table Decision domain Quality of evidence Is there high- or moderate-quality evidence? Judgement Yes  Balance of benefits versus harms and burdens Are you confident that the benefits outweigh the harms and burdens for the recommended strategy? Values and preferences Are you confident about the assumed or identified relative values and are they similar across the target population? Resource implications Is the cost small relative to the net benefits for the recommended strategy? No  Summary of reason for judgement There is low-quality evidence for the diagnostic test accuracy of both triage tests and a comparison between the strategies. There is low- to very-low-quality evidence for the effects of treatment and the natural progression of CIN from observational studies often with inconsistent results across studies. Also the link between test accuracy data and treatment effects is very uncertain. The benefits of HPV test followed by colposcopy (reduction in CIN recurrence, cervical cancer, and related mortality) may be greater than with HPV test followed by VIA. But there may be greater overtreatment with HPV test followed by colposcopy without biopsy. Little or no difference in cancers detected. High value was placed on a screen-and-treat strategy versus no screening, since qualitative studies have shown that once women decide to be screened they find the screening tests and immediate treatment acceptable. High value was placed on the greater number of women overtreated and potential complications. High value was placed on women finding a biopsy less acceptable than visual inspection. There may be greater resource implications by adding colposcopy then with adding VIA to the HPV test due to increased training of providers, quality control, waiting time, and potential for more women lost to follow-up.

Yes 

No 

Yes 

No 

Yes 

No 

173

Evidence for an HPV test followed by VIA compared to an HPV test followed by colposcopy to screen for CIN2+ in women of HIV-positive status 1. Flowchart of screen-and-treat strategies Asymptomatic HIV-positive women

HPV test

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

VIA

Colposcopy

Test + (TP & FP)

Test – (TN & FN)

Test + (TP & FP)

Test – (TN & FN)

Eligible for cryo

Not eligible for cryo

Eligible for cryo

Not eligible for cryo

Treat with cryo

Treat with CKC

Treat with LEEP

Treat with cryo

Treat with CKC

Treat with LEEP

Outcomes*

174

2. Evidence used for decision-making: HPV test followed by VIA compared to HPV test followed by colposcopic impression Diagnostic test accuracy (data based on women of unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test 95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91) Pooled sensitivity VIA Pooled specificity VIA 69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92) Pooled sensitivity colposcopic impression Pooled specificity colposcopic impression 95% (95% CI: 86 to 98) 42% (95% CI: 26 to 61)

(Reference standard: colposcopy with biopsy when indicated)

175 2.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to HPV test followed by colposcopic impression No. of studies (No. of patients)a 5 studies (8921 patients) 5 studies (8921 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test followed by VIA 66 24 fewer Seriousb Nonec Seriousd Nonee Undetected HPV test followed by colposcopic impression 90

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

Importance CRITICAL

⊕⊕⊝⊝ low

881 65 more

816

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

19 65 fewer

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

34 24 more

CRITICAL

Footnotes: a This is the number of studies that assessed DTA data for HPV test and VIA. b We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. Colposcopy studies had unclear blinding of index test results. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by VIA and for HPV test followed by colposcopic impression were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women of unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. e Wide CI for HPV test sensitivity and VIA specificity, and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

176 2.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies HPV test followed by VIA compared to HPV test followed by colposcopic impression Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVàVIA +/– CKC 1564 2190 28 859 HPVàVIA +/– LEEP 1662 2327 30 891 34 000 723 602 – 75 789 190 536 – 108 508 19 000 3168 29 553 – 11 545 1492 709 – 154 1628 HPVàVIA +/– cryo 1662 2327 30 891 HPV colp imp +/– CKC 519 726 9745 HPV colp imp +/–LEEP 654 915 12 543 10 000 392 574 – 222 1047 84 000 3545 59 609 – 23 1125 HPV colp imp +/– cryo 654 915 12 543

No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

177 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) Colposcopy: pooled sensitivity 95% (95% CI: 86 to 98), pooled specificity 42% (95% CI: 26 to 61) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

178

3. Evidence used for decision-making: HPV test followed by VIA compared to HPV test followed by colposcopic impression and biopsy when indicated Diagnostic test accuracy (data based on women of unknown HIV status) Pooled sensitivity HPV test Pooled specificity HPV test (Reference standard: colposcopy with biopsy when indicated)

95% (95% CI: 84 to 98) 84% (95% CI: 72 to 91)

Pooled sensitivity VIA Pooled specificity VIA

69% (95% CI: 54 to 81) 87% (95% CI: 79 to 92)

3.1 Diagnostic test accuracy (DTA) evidence profile: HPV test followed by VIA compared to HPV test followed by colposcopic impression and biopsy when indicated No. of studies (No. of patients)a 5 studies (8921 patients) 5 studies (8921 patients) Factors that may decrease quality of evidence Study design Cross-sectional and cohort studies Cross-sectional and cohort studies DTA QoE Effect per 1000 patients/year for pretest probability of 10% HPV test followed by VIA 66 29 fewer Seriousb Nonec Seriousd Nonee Undetected HPV test followed by colposcopy with biopsy 95

Limitations Indirectness Inconsistency Seriousb Nonec Seriousd

Imprecision Nonee

Publication bias Undetected

Importance CRITICAL

⊕⊕⊝⊝ low

881 19 fewer

900

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

19 19 more

CRITICAL

5 studies (8921 patients)

Cross-sectional and cohort studies

Seriousb

Nonec

Seriousd

Nonee

Undetected

⊕⊕⊝⊝ low

34 29 more

CRITICAL

179 Footnotes: a This is the number of studies that assessed DTA data for HPV test and VIA. b We used QUADAS to assess risk of bias. Many studies only performed one biopsy of an abnormal lesion. This was downgraded one level in the context of other factors, in particular indirectness. c Data for HPV test followed by VIA and for HPV test followed by colposcopy with biopsy when indicated were calculated based on sensitivity and specificity of the two tests. Direct data were not available. Diagnostic test accuracy data were based on women with unknown HIV status; the data were not considered indirect and so the quality of evidence was not downgraded. d Estimates of HPV test and VIA sensitivity and specificity were variable despite similar cut-off values; inconsistency could not be explained by quality of studies. This was downgraded. This judgement was considered in the context of other factors, in particular imprecision. e Wide CI for HPV test sensitivity and VIA specificity, and therefore wide CI for TP, TN, FP, FN, may lead to different decisions depending on which confidence limits are assumed.

180 3.2 GRADE evidence table for patient-important outcomes following different screen-and-treat strategies: HPV test followed by VIA compared to HPV test followed by colposcopic impression and biopsy when indicated Events in the screen-and-treat strategies for patient-important outcomes (numbers presented per 1 000 000 patients) HPVàVIA +/– CKC 1564 2190 28 859 HPVàVIA +/– LEEP 1662 2327 30 891 34 000 723 602 – 75 789 190 536 – 108 508 19 000 3168 29 553 – 11 545 816 614 – 84 890 HPVàVIA +/– cryo 1662 2327 30 891 HPV colp biopsy +/– CKC 318 445 6069 HPV colp biopsy +/–LEEP 460 644 9014 5000 214 540 – 122 573 0 3545 32 559 – 13 615 HPV colp biopsy +/– cryo 460 644 9014

No screen10 4350 6075 79 575 – 0 500 – 0 0 – 0

181 Footnotes: The colours in the table: In each GRADE evidence table, colour-coding is used to highlight the ‘desirability’ of the effects for that outcome relative to other outcomes. The continuum runs from dark gray (desirable) through light gray and light pink to dark pink (least desirable). The numbers in the table are based on CIN2+ pretest probability 10% in women of HIV-positive status (Denny et al., 2008; De Vuyst et al., 2012; Joshi et al., 2012; Zhang et al., 2012) VIA: pooled sensitivity 69% (95% CI: 54 to 81), pooled specificity 87% (95% CI: 79 to 92) HPV test: pooled sensitivity 95% (95% CI: 84 to 98), pooled specificity 84% (95% CI: 72 to 91) The overall QoE for each of these outcomes is very low ⊕⊝⊝⊝. Our lack of confidence in these effect estimates stems mainly from very-low-quality evidence for treatment effects and natural progression/history data. 1

We assume no mortality from cervical cancer in TN and FP. To calculate the mortality from cervical cancer in women of HIV-positive status, we assumed the same risk of mortality in women of unknown HIV status: 250 deaths per 350 women with cervical cancer. These numbers are based on Eastern Africa age-standardized rates of cervical cancer and mortality provided by WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no cervical cancer in TN or FP. The calculations for cervical cancer incidence in women of HIV-positive status with persistent CIN2+ are based on a 2.7 standardized Risk Ratio of cancer when compared to women with unknown HIV status (De Vuyst et al., 2008). For women of unknown status, we assumed 350 cervical cancers per 14 000 women who have persistent CIN2+ (i.e. FN). This incidence is based on Eastern Africa age-standardized rate of cervical cancer of 350 cervical cancers per 1 000 000 women, of whom 2% have CIN2+ (20 000 women with CIN2+, and a subsequent 30% regression for a total of 14 000 with persistent CIN2+). These data are available from WHO (http://globocan.iarc.fr/, accessed 30 October 2012). We assume no CIN2+ in TN and FP. Our calculations in the model are based on 90% natural persistence of CIN2+ with no treatment (10% regression) in FN. TP are treated and recurrence rates of CIN2+ are 5.3% in cryotherapy and LEEP, and 2.2% in CKC. We assumed major bleed would be 0 in TN and FN as they were not treated. We assumed 0.000339 of the population treated with cryotherapy, 0.002257 with LEEP, and 0.001705 with CKC, based on pooled proportions in observational studies with no independent controls, will have major bleeding. We assumed 5% population risk of premature delivery in 1% of women who become pregnant. Based on pooled meta-analysis of controlled observational studies, 0.001125 of the population treated with cryotherapy, 0.000925 with LEEP, and 0.001705 of the population treated with CKC will have premature delivery. We did not identify any data about the risk of infertility after treatment for CIN2+. We assumed major infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.000135 of the population treated with cryotherapy, 0.001279 with LEEP, and 0.000888 with CKC will have major infection. We assumed minor infection would be 0 in TN and FN as they were not treated. Based on pooled proportions from studies with no independent control, 0.006473 of the population treated with cryotherapy, 0.006027 with LEEP, and 0.009368 with CKC will have minor infection. Cancers detected at first-time screening calculated from Sankaranarayanan et al. (2005). Numbers for single screening tests were calculated as ‘screen-detected’ cancers in women who participated in the screening programme; and numbers for test with colposcopy were calculated as ‘screen-detected’ plus ‘clinically detected’ cancers. For a sequence of tests (e.g. HPV test followed by VIA), the greater number of cancers detected between tests was used. No cancers would be found in the ‘no screen’ group. This is not the annual incidence of cervical cancer (which is shown in a row above). It represents the cumulative rate of cancer development before screening started (i.e. the prevalence of cancer at the time when screening is first conducted). ‘No screen’ numbers were calculated using the same assumptions above for FN, with the exception of premature delivery which was baseline risk in the population.

6 7

182

183 Ferris DG, Miller MD. Colposcopic accuracy in a residency training program: defining competency and proficiency. Journal of Family Practice, 1993, 36(5):515–520. Homesley HD, Jobson VW, Reish RL. Use of colposcopically directed, four-quadrant cervical biopsy by the colposcopy trainee. Journal of Reproductive Medicine, 1984, 29(5):311–316. Jones DE et al. Evaluation of the atypical Pap smear. American Journal of Obstetrics & Gynecology, 1987, 157(3):544–549. Kierkegaard O et al. Diagnostic accuracy of cytology and colposcopy in cervical squamous intraepithelial lesions. Acta Obstetricia et Gynecologica Scandinavica, 1994, 73(8):648–651. Mousavi AS et al. A prospective study to evaluate the correlation between Reid colposcopic index impression and biopsy histology. Journal of Lower Genital Tract Disease, 2007, 11(3):147–150. Patil K et al. Comparison of diagnostic efficacy of visual inspection of cervix with acetic acid and Pap smear for prevention of cervical cancer: Is VIA superseding Pap smear? Journal of SAFOG, 2011, 3(3):131–134.

Editing, proofreading, design and layout: Green Ink (www.greenink.co.uk)

Informations clés

Organisation Organisation mondiale de la santé (OMS)

Type de document Technical Documents

Date d'adoption 9 décembre 2013

Source Organisation mondiale de la santé