Screening for sexually transmitted diseases in rural women in Papua New Guinea: are WHO therapeutic algorithms appropriate for case detection?

Screening for sexually transmitted diseases in rural women in Papua New Guinea: are WHO therapeutic algorithms appropriate for case detection?* M. Passey,1 C.S. Mgone,2 S. Lupiwa,3 S. Tiwara,4 T. Lupiwa,5 & M.P. Alpers6 The presence of a large reservoir of untreated sexually transmitted diseases (STDs) in developing countries has prompted a number of suggestions for improving case detection, including the use of clinical algorithms and risk assessments to identify women likely to be infected when they present to clinics for other reasons. We used data from a community-based study of STDs to develop and evaluate algorithms for detection of cervical infection with Chlamydia trachomatis or Neisseria gonorrhoeae, and for detection of vaginal infection with Trichomonas vaginalis or bacterial vaginosis. The algorithms were derived using data from 192 randomly selected women, then evaluated on 200 self-selected women. We evaluated the WHO algorithm for vaginal discharge in both groups. The prevalences of cervical and vaginal infection in the randomly selected group were 27% and 50%, respec- tively, and 23% and 52%, respectively, in the self-selected group. The derived algorithms had high sensitivi- ties in both groups, but poor specificities in the self-selected women, and the positive predictive values were unacceptably low. The WHO algorithms had extremely low sensitivity for detecting either vaginal or cervical infection because relatively few women reported vaginal discharge. Simple algorithms and risk assessments are not valid for case detection in this population. Introduction Sexually transmitted diseases (STDs) are an impor- tant cause of morbidity and mortality worldwide, particularly in resource-poor settings. High rates of STDs have been documented among apparently low-risk women, with reported prevalences in ante- natal clinics, family planning clinics and rural community-based surveys in the range 1-29% for Chlamydia trachomatis, 3-49% for Trichomonas vaginalis, 0.3-22% for Neisseria gonorrhoeae, and 0.3-18% for syphilis (1-3). The complications and sequelae of these infections affect women more than men, and include ectopic pregnancies, infertility, chronic pelvic pain, postpartum endometritis, cervi- cal cancer, fetal wastage, low birth weight, and con- genital or perinatal infections (1-4). More recently there has been increasing evidence that STDs *From: Papua New Guinea Institute of Medical Research, P.O. Box 60, Goroka, E.H.P. 441, Papua New Guinea. Research Fellow. Requests for reprints should be sent to Dr Passey at "Ronyerra", MSF 2002, Armidale NSW, 2350 Australia. 2 Deputy Director and Research Fellow. 3Registered Nurse. 4Medical Laboratory Technician. 5Principal Technical Officer. 6 Director. Reprint No. 5876 increase sexual transmission of human immuno- deficiency virus (HIV) (1, 5-7). Although early detection and treatment of STDs can prevent complications and minimize the severity of long-term sequelae, many infections go untreated. Utilization of specialized services for the management of STDs is often low, in part because infections are frequently asymptomatic or produce vague, nonspecific symptoms, particularly among women. Cultural barriers, as well as poor under- standing of the significance of symptoms may also reduce care-seeking by women. In order to make services more widely available, it has been suggested that STD screening and management should be in- corporated into other primary health services, in- cluding family planning, antenatal, and maternal and child health (MCH) clinics (1-4, 8-14). Training staff and providing the necessary resources for appropri- ate management of symptomatic patients presenting for treatment at every level of the health system is clearly better than not having these services avail- able. However, the vast majority of infected women (and possibly men) appear not to present for treat- ment at all, and there is thus a need for active case detection if infected people are to be identified and treated. A further constraint is that laboratory diagnosis for STDs, even for syphilis or trichomonal infection, is frequently unavailable in peripheral health facili- ties in developing countries. In such settings, tech- Bulletin of the World Health Organization, 1998, 76 (4): 401-411 © World Health Organization 1998 401 M. Passey et al. niques for detection of chlamydial and gonococcal infection are prohibitively expensive and technically demanding (14). In response to the limited availabil- ity of laboratory tests, WHO has developed algo- rithms for the syndromic management of patients presenting for STD treatment (15). These algorithms are presented as flow charts (see Figs. 1 and 2), which indicate appropriate management based on the patient's symptoms, signs and risk assessment, and have been developed for various levels of clinical and microscopic capacity. A standard risk assess- ment has been developed, but it is recommended that this should be tailored to local risk factors if the information is available. However, these algorithms are intended for case management of people specifi- cally presenting for treatment, and were not devel- oped as a screening tool for case detection. For case detection, screening based on risk as- sessment (with or without clinical information), followed by presumptive treatment, has been sug- gested (1, 9, 10, 12, 16). Several studies have ap- peared in which risk scores for the detection of chlamydial and gonococcal infection in various clini- cal settings in Africa have been developed and evalu- ated (9, 10, 12, 17). It has also been suggested that the WHO algorithms could be used as a screening tool in family planning, antenatal, and MCH clinics (18). This approach has the advantage that training health staff to use the WHO algorithms as a screen- ing tool could be readily combined with training them in their use for symptomatic patients, and would avoid the potential confusion that might arise if different risk assessment tools are used for differ- ent purposes. Since these algorithms were developed for the management of symptomatic patients, they are hierarchical in nature, requiring symptoms for entry. While this clearly limits their usefulness for asymptomatic infections, where even mild symptoms are recognized by women attending clinics for other reasons, the algorithms are potentially of benefit. To the best of our knowledge, evaluation of the WHO algorithms as a screening tool for detection of chlamydial and/or gonococcal infection has only been reported from one non-African country (19). This study, among urban, married women in Turkey, found a low sensitivity, particularly for situations in which vaginal examination was not possible. However, the population studied had a relatively low prevalence of chlamydial infection (4.9%). To date, published studies evaluating various methods for screening for STDs have focused on chlamydial infection with or without concomitant gonorrhoea. Despite high reported prevalences of infection with T. vaginalis, as well as the increased risk of pre-term birth, low birth weight (1) and HIV transmission (7) associated with trichomonal infection, in addition to the immediate morbidity suffered by infected women, no studies have evalu- ated the vaginal discharge algorithm as a screening tool for detection of trichomonal infection. We used data collected for a community-based STD study in the highlands of Papua New Guinea to evaluate clinical algorithms as a screening tool for detection of chlamydial, gonococcal, and trichomonal infection in women. Based on a previous analysis of risk factors (20), we developed our own algorithms using data from a randomly se- lected group of rural women and then tested these algorithms in a different group of women from the same community. Additionally we evaluated the WHO algorithms for vaginal discharge (with and without speculum examination) as potential screen- ing tools in both groups of women. The present arti- cle describes our findings. Methods In 1995 we conducted a cross-sectional survey of a rural and peri-urban population in the Asaro Valley in the highlands of Papua New Guinea to estimate the prevalence of STDs in an adult population of reproductive age; to determine the risk factors for STDs; and to assess treatment seeking. The data on prevalence and risk factors have been reported in detail elsewhere (20). The majority of the people in this area are subsistence farmers who grow coffee as a cash crop or labourers on coffee plantations, many of whom are migrants from other parts of the country. The highlands highway, the major route running between the coastal ports and the highland population centres, runs through this valley. The nearest town, Goroka, has a hospital which provides both outpatient and inpatient services and acts as a referral centre for the province's health centres. The hospital has STD, family planning, and antenatal clinics, as well as a weekly outpatient gynaecological clinic. Ethical clearance for the study was obtained from the Medical Research Advisory Committee of Papua New Guinea. Recruitment of study subjects Two groups of women were recruited, as outlined below. Randomly selected women. A cluster-sampling scheme was used to select a random sample of women aged 15-45 years from a population of ap- proximately 20 000 that had been recently censused by the Papua New Guinea Institute of Medical Re- search (IMR) for another study. Details of partici- WHO Bulletin OMS. Vol 76 1998402 Screening for STDs in rural Papua New Guinea pant recruitment have been described elsewhere (20). A total of 201 women (75% of those selected) were interviewed, examined and had laboratory re- sults obtained for them. For the present study, women who were menstruating at the time of the examination were excluded, leaving 192 participants. Self-selected women. Any woman from the selected study villages who had not been randomly selected, but wished to participate in the study, was included in a second group. These women were interviewed, examined, and tested in the same manner, and im- mediately following the randomly selected women. In this group there were 222 women who completed the study protocol, with 200 remaining after exclu- sion of those who were menstruating. All participants gave their free and informed consent after they were provided with a detailed ex- planation of the study. Data and specimen collection Demographic data, obstetric and sexual history, and current symptomatology were collected from each participant during an interview by a female nurse or physician. Each participant was then seen by a fe- male physician who carried out a brief general and a full gynaecological examination, including speculum and bimanual examinations. Specimens collected in- cluded the following: blood for syphilis serology; a cotton-tipped high vaginal swab for detection of bac- terial vaginosis by Gram staining; a cotton-tipped endocervical swab for Gram stain and culture of N. gonorrhoeae; a Dacron endocervical swab for detection of C. trachomatis by polymerase chain reaction (PCR); and amine test, pH determination, and wet mount preparation of vaginal secretions from the speculum, for detection of bacterial vaginosis and T vaginalis. Pelvic inflammatory disease (PID) was diagnosed when three of the fol- lowing were present: adnexal tenderness, cervical motion tenderness, uterine tenderness, and lower abdominal tenderness. Laboratory methods Wet mount preparation of vaginal secretions. These were immediately prepared from the speculum using normal saline, after determination of pH using standard test strips, and carrying out the amine test using potassium hydroxide (21). The wet mount was used to detect T. vaginalis, whose presence was con- sidered to be diagnostic for trichomonal vaginitis, and for the laboratory confirmation of bacterial vaginosis, which was considered positive if three of the following criteria were present: clue cells on wet mount or Gram stain; pH > 4.5; positive amine test; and the absence of normal flora on a Gram stain of the vaginal fluid. Gram stain. Smears were stained using standard techniques (22). The vaginal swab was used for de- tection of bacterial vaginosis (see above) and the endocervical swab for detection of Gram-negative intracellular diplococci. Syphilis serology. The rapid plasma reagin test (RPR) (Murex Diagnostics, Dartford, England) was performed following the manufacturer's instruc- tions and using a cut-off of 1:4. The results of the RPR were confirmed using a Treponema pallidum haemagglutination test (TPHA)(Wellcome Diagnostics, Dartford, England). If both tests were positive the woman was considered to have active syphilis. Culture for N. gonorrhoeae. Endocervical swabs were immediately smeared onto GC media (Difco, Detroit, MI, USA) containing vancomycin inhibitor and onto chocolate agar, and subsequently trans- ported to the laboratory in candle jars, where they were placed in a carbon dioxide incubator at 37 °C and read after 24 h and 48 h. Colonies were Gram- stained and identified further using standard tech- niques (22). Detection of C. trachomatis using PCR. Swabs were immediately placed into phosphate-buffered saline and then transported to the laboratory for process- ing or storage at -70 'C until they were analysed. Following crude DNA extraction, PCR was per- formed using the method described by Hayes et al. (23) - a two-stage (nested) PCR, which amplifies the gene encoding the chlamydial major outer membrane protein (MOMP). The gold standards used for comparisons with the clinical algorithms were as follows: for cervical infection, the detection of C. trachomatis by PCR or N. gonorrhoeae by culture; and for vaginal infection, the detection of T. vaginalis by wet mount or bacte- rial vaginosis by wet mount and Gram stain. Treatment All women were treated immediately based on the clinical findings and the wet mount examination, fol- lowing the standard treatment used in Papua New Guinea. Additionally, any infections subsequently detected by laboratory tests were treated at a follow- up visit. WHO Bulletin OMS. Vol 76 1998 403 M. Passey et al. Statistical analysis and evaluation of algorithms To develop our own clinical algorithms, we used risk factors for chlamydial and trichomonal infections that we had identified in a previous analysis of the data for the randomly selected women (20). For this, purpose we developed univariate and multivariate logistic regression models using the method of gen- eralized estimating equations developed by Liang & Zeger for analysing clustered binary data (24). This method gives robust variance estimates for the re- gression parameters taking into account the clus- tered sampling design. Odds ratios (OR) and 95% confidence intervals (CI) were calculated. Risk factors identified in the univariate analysis were assessed singly and in various combinations to determine their sensitivity (proportion of true infections classified as positive by the algorithm), specificity (proportion of truly negative women clas- sified as negative by the algorithm) and positive pre- dictive value (proportion of true positives among those classified as positive by the algorithm) for the diagnosis of cervical infection and vaginal infection, compared to the results obtained using the gold standards (see above). Since we were developing an algorithm that could be used as a screening tool, only variables obtained by interview (and not those from examination or laboratory tests) were considered. Using the sensitivity, specificity and positive predic- tive value results, we identified the most useful algo- rithm for each of the outcomes (cervical infection and vaginal infection). These algorithms were then tested on the self-selected group of women by deter- mining their sensitivity, specificity, and positive pre- dictive value for this group. Evaluated also were the WHO algorithms for vaginal discharge (15). Two levels of the algorithm were evaluated (see Figs. 1 and 2); the first level is used in settings where no pelvic or speculum exami- nation is possible; and the second level assumes that speculum and bimanual examinations are possible. The algorithms attempt to differentiate between women who have only vaginal infections and those who also have cervical infections with C. trachomatis and/or N. gonorrhoeae, by using a risk assessment. We evaluated the WHO algorithms using two risk assessments: the standard assessment (partner symp- tomatic, or any two of the following: age < 21 years; single; > one sex partner; or new partner in previous 3 months); and the assessment we developed our- selves. For our own risk assessment we used the variables included in our derived algorithm for the detection of cervicitis. The WHO algorithms were applied to the women in each of the two groups in a simulation based on the interview and examination Fig. 1. WHO algorithm for vaginal discharge for use when a speculum is not available. Vaginal discharge vaginal discharge s asssment positive - No Yes * Treat for vaginitis only * Educate *Counsel if needed * Promote/provide condoms * Treat for cervictis and vaginitis * Educate 0 Counsel if needed * Promote/provide condoms * Partner management * Return if necessary Positive = partner symptomatic or any two of: age <2lyears; single; >1 partner; new partner in past 3 months WHO 98130 data, and sensitivity, specificity, and positive predic- tive values were calculated. Potential differences between the two groups of women were examined using Pearson's x2 test, with Fisher's exact test used when appropriate. All data were analysed using Epi Info 5 software (25). Differ- ences were regarded as statistically significant at the P < 0.05 level. Results Sociodemographic factors and prevalence of STDs The demographic, behavioural, and clinical charac- teristics of the two groups are shown in Table 1. In the univariate analysis, the self-selected group was significantly more likely to have had some formal education, have no living children, report a history of STD, and have a symptomatic partner (P < 0.05); they were also significantly more likely to report a current abnormal vaginal discharge and to have an abnormal vaginal discharge on examination. There were no differences in marital status between the two groups, with the majority of women in both groups being married as the only wife. A small pro- portion were co-wives, being married to polygamous men. The prevalences of STDs and other reproduc- tive tract infections (RTIs) for the two groups are WHO Bulletin OMS. Vol 76 1998404 Screening for STDs in rural Papua New Guinea Fig. 2. WHO algorithm for vaginal discharge for use with a speculum. Vaginal discharge (with speculum) Patient complains of vaginal discharg( Risk assessment positive? - Yes * Treat for cervicitis and vaginitis * Educate * Counsel if needed * Promotelprovide condoms * Partner management * Return if necessary No Mucupus from cervix Pouevginal discharge? Curd-like vaginal dischare No discharge? *Treat for cervicitis and vaginitis 0 Treat for trichonomas 0 Treat for candida 0 Educate * Educate and bacterial vaginosis 0 Educate 0 Counsel if needed * Counsel if needed * Educate * Counsel if needed 0 Promote/provide condoms * Promote/provide condorm *Counsel if needed 0 Promote/provide condoms * Partner management 0 Promote/provide condoms 0 Return if necessary * Return if necessary 0 Partner management * Return if necessary aPositive = partner symptomatic or any two of: age <21years; single; >1 partner; new partner in past 3 months WHO 98131 Table 1: Selected characteristics of the two groups of study women, Asaro Valley, Papua New Guinea No. in random group No. in self-selected group Characteristic (n = 192) (n = 200) P value Age - 25 years 69 (36)a 84 (42) 0.26 Marital status 0.80 Married Only wife 141 (73) 146 (73) Co-wife 14 (7) 18 (9) Single/separated/widowed 37 (19) 36 (18) Any formal educationb 81 (45) 102 (58) 0.026 No living children 42 (22) 83 (42) <0.0001 <4 living children 146 (76) 156 (78) 0.73 Currently pregnant 25 (13) 14 (7) 0.068 Infertilec 50 (26) 69 (35) 0.087 Reported STD, ever 61 (32) 90 (45) 0.010 Reported STD in last 3 months 31 (16) 46 (23) 0.11 Genital symptom treated in last 3 monthsd 19 (10) 31 (16) 0.13 >1 partner in last 3 months 12 (6) 10 (5) 0.59 >1 partner in last 12 months 18 (9) 28 (14) 0.13 Partner symptomatic 18 (9) 34 (17) 0.038 Ever used a condom 14 (7) 10 (5) 0.344 Reported current abnormal discharge 42 (22) 63 (32) 0.042 Vaginal discharge on examination 135 (70) 162 (81) 0.019 Profuse vaginal discharge on examination 115 (60) 136 (68) 0.12 Mucopurulent cervical discharge 17 (9) 28 (14) 0.16 a Figures in parentheses are percentages. b A total of 36 women had data missing on education: 13 in the random group, 23 in the self-selected group. c Couples were considered infertile if the woman reported that they wanted more children, were trying to conceive, and had had unprotected intercourse for ¢2 years. d Genital symptoms treated in last 3 months, regardless of whether or not the woman perceived them to be sexually transmitted. WHO Bulletin OMS. Vol 76 1998 405 M. Passey et al. Table 2: Prevalence of sexually transmitted diseases (STDs) and other repro- ductive tract infections among the two groups of study women, Asaro Val- ley, Papua New Guinea No. in random group No. in self-selected group (n= 192) (n= 200) Pvalue Chlamydia trachomatis 51 (27)a 45 (23) 0.41 Neisseria gonorrhoeae 3 (2) 2 (1) 0.96 Trichomonas vaginalis 89 (46) 91 (46) 0.95 Syphilisb 8 (4) 10 (5) 0.88 Bacterial vaginosisc 18 (9) 26 (13) 0.32 Pelvic inflammatory diseased 21 (13) 34 (19) 0.17 Any laboratory-confirmed STDe 113 (59) 114 (57) 0.79 a Figures in parentheses are percentages. b RPR and TPHA tests both positive. c A total of 3 women had missing data: 1 in the random group and 2 in the self-selected group. dWomen who were currently pregnant or who had had a hysterectomy were excluded (28 in the random group and 17 in the self-selected group); a further 2 women in the self-selected group had missing data. e Any laboratory-confirmed STD includes C. trachomatis, N. gonorrhoeae, T. vaginalis and positive syphilis serology. shown in Table 2. Both groups of women had high levels of disease, with 52 (27%) of the randomly selected and 46 (23%) of the self-selected women having cervical infection with either C. trachomatis or N. gonorrhoeae or both. The prevalence of vaginal infection with T. vaginalis and/or bacterial vaginosis was 96 (50%) and 103 (52%) in the two groups, respectively. Nearly 60% of women in each group had some sort of STD, with no significant differences in the level of infection in the two groups, although the self-selected group had a slightly higher level of PID. Development of algorithms Cervical infection. In the univariate analysis the fol- lowing variables, obtained by interview, were signifi- cantly associated (positively or negatively) with chlamydial infection: age - 25 years (odds ratio, (OR) = 4.5; 95% confidence limits (CL) = 2.5, 7.9), being married as the only wife (OR = 0.2; 95% CL = 0.1, 0.5), having fewer than four living children (OR = 10.8; 95% CL = 3.4,34.4); current use of a modern contraceptive method (OR = 0.4; 95% CL = 0.2, 0.8) and reported STD in the previous 3 months (OR = 2.1; 95% CL = 1.1, 4.1). Having had more than one sexual partner in the previous 12 months (OR = 3.2; 95% CL = 0.9, 10.6) was almost statisti- cally significant, as was reporting an abnormal vaginal discharge (OR = 1.6; 95% CL = 1.0, 2.5). All of these variables, with the exception of currently using a modern contraceptive method were assessed among the randomly selected women for inclusion in our algorithm, although having had an STD in the previous 3 months was modified slightly to account for treatment obtained (i.e. STD in the previous 3 months with no treatment). Selected results for these variables singly and in combination are shown in Table 3. Not all the combinations that were assessed are shown. The last algorithm in Table 3 (any one of the following: age - 25 years; not being an "only wife", or having more than one sexual partner in the previ- ous 12 months) was considered to be most useful, with good sensitivity (81%), although the positive predictive value was low (45%). This algorithm was evaluated in the self-selected women and was also used as the risk assessment for the WHO algorithms. Vaginal infection. The following variables were asso- ciated (positively or negatively) with trichomonal infection in the univariate analysis: being married as the only wife (OR = 0.4; 95% CL = 0.2, 0.8); having some formal education (OR = 0.5; 95% CL = 0.3, 0.8); having no living children (OR = 2.3,95% CL = 1.1, 4.7); infertility (OR = 2.2; 95% CL = 1.0, 4.5); current use of a modern contraceptive method (OR = 0.5; 95% CL = 0.3, 0.9); having more than one sexual partner in the previous 12 months (OR = 3.3; 95% CL = 1.3, 8.7); and treatment of any genital symptoms in the previous 3 months (whether consid- ered sexually transmitted or not) (OR = 2.7; 95% CL = 1.5, 5.0). All of these variables except use of modern contraceptive methods and treatment in the previous 3 months were assessed for inclusion in our algorithm. Although not significant in the univariate analysis, we also assessed reported vaginal discharge (OR = 1.6; 95% CL = 0.8, 3.1). Sensitivities, WHO Bulletin OMS. Vol 76 1998406 Screening for STDs in rural Papua New Guinea Table 3: Selected results obtained with the screening algorithm for variables (single and combined) for predicting cervical infection with Chiamydia trachomatis/Neisseria gonorrhoeae in 192 randomly selected study women Screening algorithm % positive on algorithm Sensitivity (%) Specificity (%) ppVa (%) Age - 25 years 36 62 74 46 Not an "only wife" 27 50 82 51 <4 living children 76 96 31 34 Untreated STD in previous 3 months 14 21 89 42 >1 sexual partner in previous 12 months 9 17 94 50 Reported vaginal discharge 22 27 80 33 Any one of the age, marriage, STD or partner 53 83 58 42 variables shown aboveb Any two of the age, marriage, STD or partner 22 48 87 58 variables shown aboveb Any one of the age, marriage, or partner 49 81 63 45 variables shown abovec a Positive predictive value. b The following variables were included: age - 25 years; not an "only wife"; reporting an untreated STD in the previous 3 months; and reporting more than one sexual partner in the previous 12 months. c The following variables were included: age - 25 years; not an "only wife"; and reporting more than one sexual partner in the previous 12 months. specificities and positive predictive values for these variables, singly and in combination, as predictors of vaginal infection in the randomly selected women are shown in Table 4. Not all combinations assessed are shown. The last algorithm in Table 4 (any one of the following: not being an "only wife"; having no living children; being infertile; or reporting an abnor- mal vaginal discharge), with a sensitivity of 72%, specificity of 57%, and positive predictive value of 63%, was chosen and evaluated in the self-selected group of women. The results of the evaluation of our algorithms on the self-selected women are shown in Table 5, together with the evaluation of the WHO algorithms on both groups of women. Discussion None of the algorithms performed well, although in general, they all performed better at detecting vaginal than cervical infection. The poor perform- ance of algorithms or risk scores as a screening tool for detecting cervical infection has been noted previ- ously (12, 17, 19, 26, 27). Other workers have found them useful, although all such studies have included clinical signs or simple tests in the algorithms, an option that may not always be feasible in primary health care settings (9, 10, 28, 29). Since the WHO algorithms are dependent on the woman reporting a vaginal discharge, they had a low sensitivity for detecting either cervical or vaginal Table 4: Selected results obtained with the screening algorithm for variables (single and combined) for predicting vaginal infection with Trichomonas vaginalislbacterial vaginosis in 192 randomly selected study women Screening algorithm % positive on algorithm Sensitivity (%) Specificity (%) PPVa (%) Not an "only wife" 27 36 83 69 No formal education 55 61 51 54 No living children 22 27 83 62 Infertile 26 33 81 64 >1 partner in last 12 months 9 15 96 78 Reported vaginal discharge 22 27 83 62 Any one of the above 83 92 26 55 Any two of the above 44 61 74 70 Any one of the marriage, children, infertility or 57 72 57 63 vaginal discharge variables aboveb a Positive predictive value. bThe following variables were included: not an "only wife", no living children, infertility and reported vaginal discharge. WHO Bulletin OMS. Vol 76 1998 407 M. Passey et al. Table 5: Selected results obtained using clinical algorithms for predicing cervical infection with Chiamydia trachomatisiNeisseria gonorrhoeae, or vaginal infection with Trachoma vaginalislbacterial vaginosis among randomly selected and self-selected study women Randomly selected women (n = 192) Self-selected women (n = 200) % positive Sensitivity Specificity PpVa % positive Sensitivity Specificity ppVa Screening algorithm on algorithm (%) (%) (%) on algorithm (%) (%) (%) Cervical infection WHO algorithms Standard risk assessmentb No speculum 8 10 93 33 11 17 91 36 Speculum 9 12 91 33 16 26 87 38 Derived risk assessmentc No speculum 14 23 90 46 23 37 82 38 Speculum 15 23 89 43 23 37 81 37 Derived algorithmd 49 81 63 45 61 76 44 29 Vaginal infection WHO algorithmse No speculum 22 27 83 62 32 34 71 56 Speculum 19 25 88 67 30 32 72 55 Derived algorithmf 57 72 57 63 77 83 30 56 a Positive predictive value. b The standard risk assessment is provided by WHO with the flow charts and is partner-symptomatic or any two of the following: age < 21 years; single; >1 partner; new partner in past 3 months. c The derived risk assessment is the same as the derived algorithm for cervical infection (see footnote d). dThe derived algorithm for cervical infection is any one of the following: age - 25 years; not an "only wife"; more than one partner in the previous 12 months. e Only the results using the derived risk assessment are shown here. f The derived algorithm for vaginal infection is any one of the following: not an "only wife"; no living children; infertility; and reported vaginal discharge. infection in both groups of study women, although the self-selected women more frequently reported a discharge, with a concomitantly higher sensitivity. The majority of women had an abnormal discharge on examination, suggesting that the low reporting, in part reflects a lack of recognition of the discharge by many women. While it is well known that cervical infections are frequently asymptomatic (1, 4, 9, 10, 12), it is less well recognized that trichomonal infec- tions are also often asymptomatic or that infection correlates poorly with reported vaginal discharge. Education and awareness campaigns aimed at im- proving people's understanding of STDs and their symptoms may help improve this situation. When the derived risk assessment was used in the WHO algorithms, the sensitivity of the algorithm for detecting cervical infection improved consider- ably, but was still unacceptably low at 23% and 37% for the randomly selected and self-selected women, respectively. Use of the derived risk assessment also improved the positive predictive value for cervical infection in the random group but not in the self- selected women. Although the use of a speculum to visualize the cervix slightly improved the sensitivity of the WHO algorithm to detect cervical infection when the standard risk assessment was used, there was no im- provement in sensitivity when the derived risk assessment was used. With these algorithms, the advantage of using a speculum lies in the observation of a mucopurulent cervical discharge in women who are negative on the risk assessment. Since only a few women had such a discharge (9% and 14% in the randomly selected and self-selected groups, resp.), there was little benefit gained from this. Our derived algorithms for both cervical and vaginal infection were much more sensitive than the corresponding WHO algorithms, largely because they are not dependent on the reporting of vaginal discharge. However, our algorithms resulted in an increased proportion of positives with a reduced specificity. The resultant positive predictive values were similar to those for the WHO algorithms, with the exception that the positive predictive value for cervical infection among self-selected women was lower for the derived algorithm. As a result, al- though the majority of women with cervical or vaginal infections would be detected and appropri- ately treated if our algorithms were used for case detection, there would also be a considerable number treated needlessly. For both of the derived algorithms (cervical and vaginal), the specificity and positive predictive value were lower for the self-selected than for the ran- WHO Bulletin OMS. Vol 76 1998408 Screening for STDs in rural Papua New Guinea domly selected women. This arose because in the self-selected group there were more women with some of the identified risk factors (for cervical infec- tion: age - 25 years and > one sexual partner in the previous 12 months; and for vaginal infection: no living children, infertility, and reported vaginal dis- charge) but not more infection. The similar rates of infection in the two groups, despite apparently higher risk factors in the self-selected group, was in part accounted for by a greater proportion of the self-selected women having been treated for genital symptoms in the previous 3 months (Table 1). Addi- tionally, the self-selected women who sought treat- ment were more likely to have attended an STD clinic, rather than a health centre or other health facility (data not shown), and at the time of the study STD clinics were the only ones in Papua New Guinea routinely using doxycycline. This recent treatment presumably reduced the level of infection in the self-selected group. Thus, although the risk factors were identified for the group that most closely represents the general population (randomly selected) it cannot be assumed that the same risks will apply in a self-selected group. Conversely, in other settings where risk factors have been based on a clinic population, they may not be applicable for screening a general population. The effect of prior treatment on rates of infec- tion also needs to be considered if risk assessments are to be used for screening STDs. Like most other risk assessments, our algorithms were derived from data about prevalent and not incident disease. Many factors can affect prevalence, including both inci- dence and treatment. Even if a risk assessment is valid when it is first applied to an individual, if that individual is determined to be at risk and conse- quently treated the infection usually disappears. Re- peated application of the risk assessment would increasingly identify people who were no longer in- fected. Thus any such system will need periodic revi- sions if it is properly implemented. Although the derived algorithms were fairly sensitive, none had a high positive predictive value. For an algorithm to be useful for case detection, it is important that the positive predictive value be con- siderably higher than the prevalence, since otherwise the probability of an identified person having the disease is no greater after than before it is applied. When a disease is highly stigmatized, there is an additional problem associated with misclassifying a person as positive, beyond the cost of unnecessary treatment. In many societies a woman diagnosed with an STD may suffer reprisals from her husband or others, and such a diagnosis carries a considerable burden. She may also be confused by being labelled as high risk and therefore in need of treatment, when she has no symptoms, has not been examined, and has not had any laboratory tests performed. Even if an excellent case detection tool existed, its imple- mentation in primary health care settings may be problematic. This evaluation has shown that the WHO algo- rithms for vaginal discharge are not suitable for use as a screening tool for case detection in the study population. We have developed our own algorithms for detecting both cervical and vaginal infection, using readily identifiable risk criteria, and tested them in a different group of women. Although clini- cal signs may have improved the validity of these algorithms, our aim was to include only data that could be collected by asking a few quick questions in a busy clinic. Unfortunately, although fairly sensi- tive, these derived algorithms lack specificity and therefore are also not appropriate for use as a screening tool. While vaginal infections should be easily detectable if a speculum examination and wet mount are performed, these tests are not always available in rural health facilities and privacy is fre- quently lacking. None the less, a wet mount is more frequently feasible than laboratory diagnosis of chlamydial or gonococcal infection, and high priority should be given to the development and evaluation of simple, quick, self-administered tests for all STDs. Acknowledgements We thank the WHO Special Programme of Research, Development and Research Training in Human Reproduc- tion for financial assistance for this research; in particular we thank Dr P. Rowe for his on-going enthusiasm and support for our work. We are grateful to Dr N. Padian, Dr J. van de Wijgert, and Dr A. Reingold for advice and encouragement during the analysis of these data. Finally, we express our sincere gratitude to the women who par- ticipated in this study. Resume Depistage des maladies sexuellement transmissibles chez la femme, en milieu rural, en Papouasie-Nouvelle-Guinee: les algorithmes therapeutiques de l'OMS sont-ils adaptes au depistage? L'importance du reservoir de maladies sexuelle- ment transmissibles (MST) non traitees, en particulier chez la femme, a suscite sur certain nombre de propositions pour am6liorer le depistage dans les pays en d6veloppement. On peut citer l'utilisation d'algorithmes cliniques et l'evaluation WHO Bulletin OMS. Vol 76 1998 409 M. Passey et al. des risques pour rep6rer les femmes qui peuvent etre contamin6es. Ces algorithmes sont destin6s a etre utilis6s quand la femme s'adresse au centre de sant6 pour d'autres motifs. Plusieurs de ces outils ont ete evalu6s en Afrique, mais peu d'6valuations ont ete r6alis6es ailleurs. Nous avons exploit6 les donnees d'une 6tude des MST conduite en communaut6 dans un secteur rural de la Papouasie-Nouvelle-Guinee pour 6laborer et 6valuer des algorithmes de d6pistage des infections cervicales a Chlamydia trachomatis ou a Neisseria gonorrhoeae, des infections vaginales a T. vaginalis ou des vaginoses bact6riennes. Les algorithmes ont ete mis au point avec les donn6es concernant 192 cas selectionn6s au hasard, et 6valu6s ensuite sur un autre groupe auto-selectionn6 de 200 femmes. Nous avons egalement 6valu6 dans les deux groupes l'algorithme OMS pour la prise en charge des pertes vaginales en tant qu'outil de d6pistage. La pr6valence des infections du col et du vagin dans le groupe choisi au hasard 6tait respectivement de 27% et 50%; dans le groupe autos6lectionn6, les chiffres correspondants etaient de 23% et 52%. La sensibilit6 des algorithmes obtenus pour la recher- che des infections du col et du vagin 6tait 6lev6e dans les deux groupes (81% dans le groupe s6lectionne al6atoirement et 76% dans le groupe autoselectionn6, pour les infections cervicales; 72% et 83% respectivement dans les deux groupes, pour les infections vaginales). La sp6cificit6 des deux algorithmes 6tait cependant faible dans le groupe autos6lectionn6 (44% pour les infections du col et 30% pour celles du vagin), tandis que la valeur predictive positive 6tait beaucoup trop basse (29% pour les infections du col et 56% pour celles du vagin). La sensibilite des deux algorithmes etait extremement mauvaise concernant la recherche des infections du col ou du vagin, en raison de I'absence de correlation entre 6coulement vaginal et infection du col, et de la difficulte pour la femme de reperer les pertes vaginales. Les m6thodes bas6es sur l'6valuation des risques et les simples algorithmes ne sont pas des outils valides de depistage des infections cervicales et vaginales dans cette population. Les signes cliniques auraient pu ameliorer la validit6 de ces algorithmes; notre but 6tait toutefois de n'utiliser que les donnees pouvant etre obtenues a I'aide de quelques questions dans un service surcharge. Si les infections vaginales sont facilement d6celables grace a un examen au sp6culum et une pr6paration extemporanee, les examens clinique et microscopique ne sont pas toujours possibles et l'intimit6 fait souvent d6faut dans les centres ruraux. 11 reste que la preparation extemporan6e est plus souvent realisable que le diagnostic au laboratoire d'une infection a Chlamydia ou a gonocoques, et que la priorite doit etre accordee a la mise au point et a 1'6valuation d'autotests simples et rapides applicables a toutes les MST. References 1. Wasserheit JN, Holmes KK. Reproductive tract in- fections: challenges for international health policy, programs, and research. In: Germain A et al., eds. Reproductive tract infections: global impact and priori- ties for women's reproductive health. New York, Plenum Press, 1992. 2. Wasserheit JN. The significance and scope of repro- ductive tract infections among third world women. In- temationaljournal ofgynecology and obstetrics, 1989, Suppl 3:145-168. 3. Brunham RC, Embree JE. Sexually transmitted diseases: current and future dimensions of the problem in the third world. In: Germain A et al., eds. Reproductive tract infections: global impact and priori- ties for women's reproductive health. New York, Plenum Press, 1992. 4. Cates W, Wasserheit JN. Genital chlamydial infec- tions: epidemiology and reproductive sequelae. American journal of obstetrics and gynecology, 1991, 164: 1771-1781. 5. Cameron DW et al. Female-to-male transmission of human immunodeficiency virus type 1: risk factors for seroconversion in men. Lancet, 1989, 2: 403-407. 6. Plummer FA et al. Co-factors in male to female trans- mission of human immunodeficiency virus type 1. Journal of infectious diseases, 1991, 163: 233-239. 7. Laga M et al. Non-ulcerative sexually transmitted diseases as risk factors for HIV-1 transmission in women: results from a cohort study. AIDS, 1993, 7: 95-102. 8. Cates W, Stone KM. Family planning, sexually trans- mitted diseases and contraceptive choice: a literature update - part II. Family planning perspectives, 1992, 24: 122-128. 9. Braddick MR et al. Towards developing a diagnostic algorithm for Chlamydia trachomatis and Neisseria gonorrhoeae cervicitis in pregnancy. Genitourinary medicine, 1990, 66: 62-65. 10. Vuylsteke B et al. Clinical algorithms for the screen- ing of women for gonococcal and chlamydial infection: evaluation of pregnant women and prostitutes in Zaire. Clinical infectious diseases, 1993, 17: 82-88. 11. Matondo P. Antenatal interventions against sexually transmitted diseases in Africa. Lancet, 1993, 341: 1565-1566. 12. Mayaud P et al. Risk assessment and other screen- ing options for gonorrhoea and chlamydial infections in women attending rural Tanzanian antenatal clinics. Bulletin of the World Health Organization, 1995, 73: 621-630. 13. Fox LJ et al. Improving reproductive health: integrat- ing STD and contraceptive services. Journal of the 410 WHO Bulletin OMS. Vol 76 1998 Screening for STDs in rural Papua New Guinea American Medical Women's Association, 1995, 50: 129-136. 14. Piot P, Laga M. Current approaches to sexually trans- mitted disease control in developing countries. In: Wasserheit JN et al., eds. Research issues in human behaviour and sexually transmitted diseases in the AIDS era. Washington, DC, American Society for Microbiology, 1991. 15. Management of sexually transmitted diseases. Un- published WHO document GPA/TEM/94.1, 1994 (available upon request from UNAIDS, 1211 Geneva 27, Switzerland). 16. Meheus A. Women's health: importance of reproduc- tive tract infections, pelvic inflammatory disease and cervical cancer. In: Germain A, et al., eds. Reproduc- tive tract infections: global impact and priorities for women's reproductive health. New York, Plenum Press, 1992. 17. Costello Daly C et al. Risk factors for gonorrhoea, syphilis, and trichomonas infections among women attending family planning clinics in Nairobi, Kenya. Genitourinary medicine, 1994, 70: 155-161. 18. Lande R. Controlling sexually transmitted diseases. Population reports. Series L: issues in world health, 1993, 9:1-31. 19. Ronsmans C et al. Clinical algorithms for the screen- ing of Chlamydia trachomatis in Turkish women. Genitourinary medicine, 1996, 72: 182-186. 20. Passey M et al. Community-based study of sexually transmitted diseases in rural women in the highlands of Papua New Guinea: prevalence and risk factors. Sexually transmitted infections, (in press). 21. Spiegel CA. Vaginitis. In: Wentworth BB, Judson FN, eds. Laboratory methods for the diagnosis of sexually transmitted diseases. Washington, DC, American Public Health Association, 1984: 151-168. 22. Ehret JM, Judson FM, Biddle JW. Gonorrhea. In: Wentworth BB, Judson FN, eds. Laboratory methods for the diagnosis of sexually transmitted diseases. Washington, DC, American Public Health Association, 1984: 43-80. 23. Hayes JL et al. Genotyping of Chlamydia trachomatis from a trachoma-endemic village in The Gambia by a nested polymerase chain reaction: identification of strain variants. Joumal of infectious diseases, 1992, 166: 1173-1177. 24. Liang K-Y, Zeger SL. Longitudinal data analysis using generalised linear models. Biometrika, 1986, 73: 13-22. 25. Dean AG et al. Epi Info version 5: a word processing, database, and statistics program for epidemiology on microcomputers. Stone Mountain, GA, USD Inc., 1990. 26. Chout RT et al. Screening for Chlamydia trachomatis infection in pregnant women in Martinique. Sexually transmitted diseases, 1995, 22: 221-227. 27. Weinstock HS et al. Chlamydia trachomatis infection in women: a need for universal screening in high prevalence populations? American joumal of epidemi- ology, 1992, 135: 41-47. 28. Johnson BA et al. Derivation and validation of a clinical diagnostic model for chlamydial cervicitis in- fection in university women. Journal of the American Medical Association, 1990; 264: 3161-3165. 29. Handsfield HH et al. Criteria for selective screening for Chlamydia trachomatis infection in women attend- ing family planning clinics. Journal of the American Medical Association, 1986, 255: 1730-1734. WHO Bulletin OMS. Vol 76 1998 411