Consolidated guidelines on HIV testing services, 2019: web annex B. GRADE table: should HIV self-testing be offered as an additional HIV testing approach?

Web Annex B 1 CONSOLIDATED GUIDELINES ON HIV TESTING SERVICES 2019 Web Annex B. GRADE table: should HIV self-testing be offered as an additional HIV testing approach? ISBN 978-92-4-001174-8 (electronic version) Web Annex B 2 Consolidated guidelines on HIV testing services, 2019. Web Annex B. GRADE table: should HIV self-testing be offered as an additional HIV testing approach? ISBN 978-92-4-001174-8 (electronic version) This publication was originally published under WHO reference number WHO/UCN/HHS/19.41. © World Health Organization 2020 Some rights reserved. This work is available under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 IGO licence (CC BY- NC-SA 3.0 IGO; https://creativecommons.org/licenses/by-nc-sa/3.0/igo). Under the terms of this licence, you may copy, redistribute and adapt the work for non-commercial purposes, provided the work is appropriately cited, as indicated below. In any use of this work, there should be no suggestion that WHO endorses any specific organization, products or services. The use of the WHO logo is not permitted. If you adapt the work, then you must license your work under the same or equivalent Creative Commons licence. If you create a translation of this work, you should add the following disclaimer along with the suggested citation: “This translation was not created by the World Health Organization (WHO). WHO is not responsible for the content or accuracy of this translation. The original English edition shall be the binding and authentic edition”. Any mediation relating to disputes arising under the licence shall be conducted in accordance with the mediation rules of the World Intellectual Property Organization (http://www.wipo.int/amc/en/mediation/rules/). Suggested citation. Web Annex B. GRADE table: should HIV self-testing be offered as an additional HIV testing approach? In: Consolidated guidelines on HIV testing services, 2019. Geneva: World Health Organization; 2020. Licence: CC BY-NC-SA 3.0 IGO. Cataloguing-in-Publication (CIP) data. CIP data are available at http://apps.who.int/iris. Sales, rights and licensing. To purchase WHO publications, see http://apps.who.int/bookorders. To submit requests for commercial use and queries on rights and licensing, see http://www.who.int/about/licensing. Third-party materials. If you wish to reuse material from this work that is attributed to a third party, such as tables, figures or images, it is your responsibility to determine whether permission is needed for that reuse and to obtain permission from the copyright holder. The risk of claims resulting from infringement of any third-party-owned component in the work rests solely with the user. General disclaimers. The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of WHO concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. Dotted and dashed lines on maps represent approximate border lines for which there may not yet be full agreement. The mention of specific companies or of certain manufacturers’ products does not imply that they are endorsed or recommended by WHO 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 WHO 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 WHO be liable for damages arising from its use. This publication forms part of the WHO guideline entitled Consolidated guidelines on HIV testing services, 2019. It is being made publicly available for transparency purposes and information, in accordance with the WHO handbook for guideline development, 2nd edition (2014). Web Annex B 3 Contents Table B1. GRADE table for HIVST compared to SOC for HTS ......................................................................................................................................................................... 4 Table B2. GRADE table for HIVST + linkage intervention compared to HIVST only for linkage to HIV CARE or ART ..................................................................... 33 Web Annex B 4 Table B1. GRADE table for HIVST compared to SOC for HTS Author(s): Muhammad S. Jamil, T. Charles Witzel, Ingrid Wilson Question: HIVST compared to SOC for HTS Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) Uptake of HIV testing (overall) 23 a randomised trials b serious c not serious d not serious e not serious none 16439/24739 (66.4%) f 7775/19260 (40.4%) f RR 1.62 (1.43 to 1.84) 250 more per 1,000 (from 174 more to 339 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (general population) 13 g randomised trials h very serious i not serious j not serious k not serious none 12870/19308 (66.7%) l 5812/14523 (40.0%) l RR 2.09 (1.69 to 2.58) 436 more per 1,000 (from 276 more to 632 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (key populations) 10 m randomised trials n serious o not serious p not serious q not serious none 3569/5431 (65.7%) r 1963/4737 (41.4%) r RR 1.28 (1.13 to 1.45) 116 more per 1,000 (from 54 more to 186 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (men who have sex with men) 7 s randomised trials t serious u not serious v not serious not serious none 2427/3404 (71.3%) w 1432/3393 (42.2%) w RR 1.37 (1.19 to 1.57) 156 more per 1,000 (from 80 more to 241 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (female sex workers) 3 x randomised trials y serious z not serious aa not serious not serious none 1142/2027 (56.3%) ab 531/1344 (39.5%) ab RR 1.16 (0.98 to 1.36) 63 more per 1,000 (from 8 fewer to 142 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (men) Web Annex B 5 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 17 ac randomised trials ad serious ae not serious af not serious not serious none 9308/12954 (71.9%) ag 3981/9632 (41.3%) ag RR 1.81 (1.56 to 2.11) 335 more per 1,000 (from 231 more to 459 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (women) 6 ah randomised trials ai serious aj not serious ak not serious not serious none 2772/4415 (62.8%) al 1578/3820 (41.3%) al RR 1.45 (1.13 to 1.86) 186 more per 1,000 (from 54 more to 355 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (young people 15-24 years) 5 am randomised trials an very serious ao not serious ap not serious not serious none 1006/2355 (42.7%) aq 700/2057 (34.0%) aq RR 2.10 (1.36 to 3.23) 374 more per 1,000 (from 123 more to 759 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (measurement time-point: ≤6 months) 17 ar randomised trials as serious at not serious au not serious not serious none 11814/17488 (67.6%) av 4058/12267 (33.1%) av RR 1.79 (1.51 to 2.13) 261 more per 1,000 (from 169 more to 374 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (measurement time-point: >6 months) 6 aw randomised trials ax very serious ay not serious az not serious not serious none 4625/7251 (63.8%) ba 3717/6993 (53.2%) ba RR 1.27 (1.10 to 1.48) 144 more per 1,000 (from 53 more to 255 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (online and mail distribution) 5 bb randomised trials bc serious bd not serious be not serious not serious none 2161/3106 (69.6%) bf 1218/3099 (39.3%) bf RR 1.47 (1.30 to 1.66) 185 more per 1,000 (from 118 more to 259 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (facility-based distribution) Web Annex B 6 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 6 bg randomised trials bh serious bi not serious bj not serious not serious none 903/1155 (78.2%) bk 520/810 (64.2%) bk RR 1.28 (1.01 to 1.61) 180 more per 1,000 (from 6 more to 392 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (secondary distribution - women to male partners) 4 bl randomised trials bm serious bn not serious bo not serious not serious none 4477/5737 (78.0%) bp 840/2578 (32.6%) bp RR 2.63 (1.81 to 3.82) 531 more per 1,000 (from 264 more to 919 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing services (secondary distribution - HIV-positive clients to partners) 2 bq randomised trials br very serious bs not serious bt not serious not serious none 507/823 (61.6%) bu 120/369 (32.5%) bu RR 2.03 (1.01 to 4.09) 335 more per 1,000 (from 3 more to 1,000 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (secondary distribution - peers) 2 bv randomised trials bw serious bx not serious by not serious not serious none 555/612 (90.7%) bz 244/324 (75.3%) bz RR 1.19 (0.97 to 1.47) 143 more per 1,000 (from 23 fewer to 354 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (community- or home-based distribution) 3 ca randomised trials cb very serious cc not serious cd not serious not serious none 6530/9559 (68.3%) ce 4419/8516 (51.9%) ce RR 1.43 (0.95 to 2.13) 223 more per 1,000 (from 26 fewer to 586 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (HIVST at facilities) 4 cf randomised trials cg very serious ch not serious ci not serious not serious none 1344/3747 (35.9%) cj 414/3564 (11.6%) cj RR 2.14 (1.22 to 3.74) 132 more per 1,000 (from 26 more to 318 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (no or basic support) Web Annex B 7 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 4 ck randomised trials cl very serious cm not serious cn not serious not serious none 268/377 (71.1%) co 157/355 (44.2%) co RR 1.60 (1.13 to 2.28) 265 more per 1,000 (from 57 more to 566 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (IFU enhancement, video or study hotline) 8 cp randomised trials cq serious cr not serious cs not serious not serious none 6985/9448 (73.9%) ct 2096/5848 (35.8%) ct RR 2.02 (1.65 to 2.47) 366 more per 1,000 (from 233 more to 527 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (group demonstration) 3 cu randomised trials cv serious cw serious cx not serious not serious none 2124/3374 (63.0%) cy 736/2599 (28.3%) cy RR 1.26 (1.01 to 1.58) 74 more per 1,000 (from 3 more to 164 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (in-person demonstration or training) 4 cz randomised trials da very serious db not serious dc not serious not serious none 6626/9675 (68.5%) dd 4511/8630 (52.3%) dd RR 1.31 (0.93 to 1.86) 162 more per 1,000 (from 37 fewer to 450 more) ⨁⨁◯◯ LOW CRITICAL Uptake of HIV testing (in-person observation or supervision) 3 de randomised trials df serious dg not serious dh not serious not serious none 1650/243 (679.0%) di 166/1613 (10.3%) di RR 1.69 (1.07 to 2.67) 71 more per 1,000 (from 7 more to 172 more) ⨁⨁⨁◯ MODERATE CRITICAL Uptake of HIV testing (virtual real-time support or supervision) 1 randomised trials serious dj serious dk not serious not serious none 193/215 (89.8%) dl 109/215 (50.7%) dl RR 1.77 (1.54 to 2.04) 390 more per 1,000 (from 274 more to 527 more) ⨁⨁◯◯ LOW CRITICAL Web Annex B 8 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) Frequency of HIV testing (men who have sex with men) 3 dm randomised trials dn serious do not serious dp not serious not serious none Among 3 trials 1242 participants in the HIVST arm reported 6351 HIV tests over 12 - 15 months compared to 1222 participants reporting 2107 HIV tests in the SOC arm during the same period. This represents 2.6 additional tests per participant during follow-up period (mean difference: 2.6, 95% CI: 1.2 - 4.0) ⨁⨁⨁◯ MODERATE IMPORTANT HIV positivity rate among randomized (overall) 16 dq randomised trials dr serious ds not serious dt not serious not serious none 388/16636 (2.3%) du 143/10330 (1.4%) du RR 1.06 (0.76 to 1.48) 1 more per 1,000 (from 3 fewer to 7 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (overall) 17 dv randomised trials dw serious dx not serious dy not serious not serious none 444/13824 (3.2%) dz 357/9689 (3.7%) dz RR 0.97 (0.74 to 1.27) ea 1 fewer per 1,000 (from 10 fewer to 10 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (general population) 8 eb randomised trials ec serious ed not serious ee not serious not serious none 227/11088 (2.0%) ef 243/7932 (3.1%) ef RR 0.81 (0.45 to 1.47) 6 fewer per 1,000 (from 17 fewer to 14 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (key populations) 9 eg randomised trials eh serious ei not serious ej not serious ek not serious none 217/2736 (7.9%) el 114/1757 (6.5%) el RR 0.99 (0.75 to 1.30) 1 fewer per 1,000 (from 16 fewer to 19 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (men who have sex with men) 6 em randomised trials en serious ed not serious eo not serious not serious none 34/1538 (2.2%) ep 14/1152 (1.2%) ep RR 1.59 (0.87 to 2.89) 7 more per 1,000 (from 2 fewer to 23 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (female sex workers) Web Annex B 9 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 3 eq randomised trials er serious es not serious et not serious not serious none 183/1198 (15.3%) eu 100/605 (16.5%) eu RR 0.87 (0.63 to 1.21) 21 fewer per 1,000 (from 61 fewer to 35 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (men) 13 ev randomised trials ew serious ex not serious ey not serious not serious none 103/3675 (2.8%) ez 28/1647 (1.7%) ez RR 1.13 (0.74 to 1.72) 2 more per 1,000 (from 4 fewer to 12 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (women) 4 fa randomised trials fb serious fc not serious fd not serious not serious none 193/1648 (11.7%) fe 102/710 (14.4%) fe RR 0.88 (0.65 to 1.19) 17 fewer per 1,000 (from 50 fewer to 27 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (young people 15-24 years) 4 ff randomised trials fg very serious fh not serious fi not serious not serious none 93/6011 (1.5%) fj 220/6973 (3.2%) fj RR 0.67 (0.33 to 1.34) 10 fewer per 1,000 (from 21 fewer to 11 more) ⨁⨁◯◯ LOW CRITICAL HIV positivity rate among tested (measurement time-point: ≤6 months) 13 fk randomised trials fl serious fm not serious fn not serious not serious none 412/12577 (3.3%) fo 344/8833 (3.9%) fo RR 0.91 (0.70 to 1.20) 4 fewer per 1,000 (from 12 fewer to 8 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (measurement time-point: >6 months) 4 fp randomised trials fq serious fr not serious fs not serious not serious none 32/1247 (2.6%) ft 13/856 (1.5%) ft RR 1.60 (0.85 to 2.98) 9 more per 1,000 (from 2 fewer to 30 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (online and mail distribution) Web Annex B 10 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 4 fu randomised trials fv very serious fw not serious fx not serious not serious none 27/1270 (2.1%) fy 12/931 (1.3%) fy RR 1.45 (0.75 to 2.78) 6 more per 1,000 (from 3 fewer to 23 more) ⨁⨁◯◯ LOW CRITICAL HIV positivity rate among tested (facility-based distribution) 4 fz randomised trials ga serious gb not serious gc not serious not serious none 97/828 (11.7%) gd 52/502 (10.4%) gd RR 1.02 (0.57 to 1.85) 2 more per 1,000 (from 45 fewer to 88 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (secondary distribution - women to male partners) 3 ge randomised trials gf serious gg not serious gh not serious not serious none 65/4155 (1.6%) gi 8/734 (1.1%) gi RR 0.58 (0.18 to 1.87) 5 fewer per 1,000 (from 9 fewer to 9 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (secondary distribution - HIV-positive clients to partners) 2 gj randomised trials gk very serious gl not serious gm not serious not serious none 74/360 (20.6%) gn 13/99 (13.1%) gn RR 1.42 (0.74 to 2.71) 55 more per 1,000 (from 34 fewer to 225 more) ⨁⨁◯◯ LOW CRITICAL HIV positivity rate among tested (secondary distribution - peers) 2 go randomised trials gp serious gq not serious gr not serious not serious none 88/567 (15.5%) gs 50/281 (17.8%) gs RR 0.85 (0.55 to 1.32) 27 fewer per 1,000 (from 80 fewer to 57 more) ⨁⨁⨁◯ MODERATE CRITICAL HIV positivity rate among tested (community- or home-based distribution) 1 randomised trials gt serious gu serious gv not serious not serious none 56/5353 (1.0%) gw 214/6728 (3.2%) gw RR 0.64 (0.18 to 2.22) 11 fewer per 1,000 (from 26 fewer to 39 more) ⨁⨁◯◯ LOW CRITICAL HIV positivity rate among tested (HIVST at facilities) Web Annex B 11 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 4 gx randomised trials gy very serious gz not serious ha not serious serious hb none 37/1291 (2.9%) hc 8/414 (1.9%) hc RR 1.18 (0.29 to 4.74) 3 more per 1,000 (from 14 fewer to 72 more) ⨁◯◯◯ VERY LOW CRITICAL Linkage to additional or confirmatory HIV testing (HIVST arm only) 7 hd randomised trials he serious hf serious hg not serious not serious none A pooled estimated from 7 studies showed 65% (52% - 78%) of those with a reactive HIVST result (n=497) had confirmatory HIV testing (n=332). ⨁⨁◯◯ LOW CRITICAL Linkage to ART initiation or HIV care among randomized (Overall) 12 hh randomised trials hi serious hj not serious hk not serious not serious none 217/12440 (1.7%) hl 79/8013 (1.0%) hl RR 1.07 (0.67 to 1.72) 1 more per 1,000 (from 3 fewer to 7 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (Overall) 11 hm randomised trials hn serious ho not serious hp not serious not serious none 217/403 (53.8%) hq 79/149 (53.0%) hq RR 0.94 (0.81 to 1.09) 32 fewer per 1,000 (from 101 fewer to 48 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (general population) 6 hr randomised trials hs serious ht not serious hu not serious not serious none 166/217 (76.5%) hv 37/47 (78.7%) hv RR 0.95 (0.81 to 1.09) 39 fewer per 1,000 (from 150 fewer to 71 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (key populations) 5 hw randomised trials hx serious hy not serious hz not serious ia not serious none 51/186 (27.4%) ib 42/102 (41.2%) ib RR 0.83 (0.58 to 1.18) 70 fewer per 1,000 (from 173 fewer to 74 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (men who have sex with men) Web Annex B 12 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 3 randomised trials ic serious id not serious ie not serious not serious none 8/8 (100.0%) if 2/2 (100.0%) if RR 1.14 (0.67 to 1.97) 140 more per 1,000 (from 330 fewer to 970 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (female sex workers) 2 ig randomised trials ih serious ii not serious ij not serious not serious none 43/178 (24.2%) ik 40/100 (40.0%) ik RR 0.66 (0.41 to 1.04) 136 fewer per 1,000 (from 236 fewer to 16 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (men) 8 il randomised trials im serious in not serious io not serious not serious none 83/115 (72.2%) ip 15/17 (88.2%) ip RR 0.88 (0.65 to 1.20) 106 fewer per 1,000 (from 309 fewer to 176 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (women) 4 iq randomised trials ir serious is not serious it not serious not serious none 52/189 (27.5%) iu 41/100 (41.0%) iu RR 0.72 (0.47 to 1.11) 115 fewer per 1,000 (from 217 fewer to 45 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (young people 15-24 years) 2 iv randomised trials iw very serious ix not serious iy not serious not serious none 4/10 (40.0%) iz 1/1 (100.0%) iz RR 0.59 (0.21 to 1.65) 410 fewer per 1,000 (from 790 fewer to 650 more) ⨁⨁◯◯ LOW CRITICAL Linkage to ART initiation or HIV care among HIV positive (measurement time-point: ≤6 months) 8 ja randomised trials jb serious jc not serious jd not serious not serious none 177/353 (50.1%) je 64/127 (50.4%) je RR 0.88 (0.71 to 1.09) 60 fewer per 1,000 (from 146 fewer to 45 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (measurement time-point: >6 months) Web Annex B 13 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 3 jf randomised trials jg serious jh not serious ji not serious not serious none 40/50 (80.0%) jj 15/22 (68.2%) jj RR 0.98 (0.79 to 1.22) 14 fewer per 1,000 (from 143 fewer to 150 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (no linkage support) 9 jk randomised trials jl serious jm not serious jn not serious not serious none 101/274 (36.9%) jo 59/122 (48.4%) jo RR 0.82 (0.62 to 1.07) 87 fewer per 1,000 (from 184 fewer to 34 more) ⨁⨁⨁◯ MODERATE CRITICAL Linkage to ART initiation or HIV care among HIV positive (phone reminder or follow up) 1 jp randomised trials jq serious jr serious js not serious not serious none 2/3 (66.7%) jt 1/1 (100.0%) jt RR 0.83 (0.28 to 2.51) 170 fewer per 1,000 (from 720 fewer to 1,000 more) ⨁⨁◯◯ LOW CRITICAL Linkage to ART initiation or HIV care among HIV positive (home visit or in-person referral) 1 ju randomised trials jv serious jw serious jx not serious not serious none 33/43 (76.7%) jy 13/20 (65.0%) jy RR 0.96 (0.76 to 1.21) 26 fewer per 1,000 (from 156 fewer to 136 more) ⨁⨁◯◯ LOW CRITICAL Linkage to ART initiation or HIV care among HIV positive (virtual real-time support) 1 randomised trials jz serious ka serious kb not serious serious kc none 1/1 (100.0%) kd 0/0 RR 1.50 (0.18 to 12.46) 2 fewer per 1,000 (from 12 fewer to 0 fewer) ⨁◯◯◯ VERY LOW CRITICAL Linkage to ART initiation or HIV care among HIV positive (financial incentive) 2 ke randomised trials kf serious kg not serious kh not serious not serious none 80/82 (97.6%) ki 6/6 (100.0%) ki RR 1.09 (0.87 to 1.38) 90 more per 1,000 (from 130 fewer to 380 more) ⨁⨁⨁◯ MODERATE CRITICAL Misuse of HIVST kits - coercion (HIVST arm only) Web Annex B 14 Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST SOC Relative (95% CI) Absolute (95% CI) 2 kj randomised trials kk serious kl not serious km not serious kn serious ko none One trial reported 4 instances of coercion or forced to test among 13267 participants. The other trial reported 0 instances of coercion to test or disclose results among 1063 participants. ⨁⨁◯◯ LOW CRITICAL Adverse events among those randomized 6 kp randomised trials kq serious kr not serious ks not serious kt serious ku none 12/5502 (0.2%) kv 1/3124 (0.0%) kv RR 1.89 (0.54 to 6.54) 0 fewer per 1,000 (from 0 fewer to 2 more) ⨁⨁◯◯ LOW CRITICAL Condomless anal sex with male partners among those randomized 4 kw randomised trials kx serious ky not serious kz not serious la not serious none 416/1544 (26.9%) lb 528/1890 (27.9%) lb RR 1.09 (0.95 to 1.24) 25 more per 1,000 (from 14 fewer to 67 more) ⨁⨁⨁◯ MODERATE IMPORTANT Condomless anal sex with male partners among those who completed follow-up 4 lc randomised trials ld serious le not serious lf not serious lg not serious none 416/1482 (28.1%) lh 528/1747 (30.2%) lh RR 1.02 (0.91 to 1.15) 6 more per 1,000 (from 27 fewer to 45 more) ⨁⨁⨁◯ MODERATE IMPORTANT CI: Confidence interval; RR: Risk ratio Explanations a. Meta-analysis: Choko, 2019a and Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined in these trials; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting); Choko, 2019b had two study groups (ANC women and ART clients) and these were presented separately. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. b. 14 individual randomized trials, 9 cluster randomized trials. c. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 14 trials and attrition bias in 3 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm; Patel, 2018: 36% LTFU overall, 44% in the intervention and 27% in the control arm). Three cluster randomized trials were subject to recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. Web Annex B 15 d. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.091; Chi² = 409.12, df = 25, p < 0.01; I² = 94%, 92% - 95%). Sub-group analyses by population, study design, measure time-point, and distribution method did not fully explain heterogeneity. Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. e. We did not downgrade for indirectness. f. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. g. Meta-analysis: Choko, 2019a had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined. Choko, 2019b had two study groups (ANC women and ART clients) and these were presented separately. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. h. 7 individual randomized trials and 6 cluster randomized trials i. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 10 trials and attrition bias in one trial (Patel, 2018: 36% LTFU overall, 44% in the intervention and 27% in the control arm). Three cluster randomized trials were subject to recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. 9 of 13 trials had more than three high risk or unclear risk of bias domains. j. There was a high statistical heterogeneity (Heterogeneity: Tau² = 0.133; Chi² = 213.31, df = 13, p < 0.01; I² = 94%, 91% - 96%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. k. We did not downgrade for indirectness but noted that all but one trial were conducted in Africa (6 in Malawi, 4 in Kenya, one in Zambia, one in South Africa, one in the US). This is expected most countries with generalized epidemics are in Africa. l. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. m. Meta-analysis: Wray, 2018 had two intervention arms which both included HIVST with similar HIVST kit distribution method, therefore both arms were combined. Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. n. 7 individual randomized trials and 3 cluster randomized trials o. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 5 trials and attrition bias in 2 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. p. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.040; Chi² = 108.69, df = 11, p < 0.01; I² = 90%, 846% - 93%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. q. We did not downgrade but noted that only men who have sex with men, transgender women and female sex workers were represented in included studies and results should be viewed with caution when applying to other key populations. r. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. s. Meta-analysis: Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore intervention arms were combined. in these trials. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. t. 6 individual randomized trials and one cluster randomized trial. Web Annex B 16 u. We downgraded once. due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 3 trials and attrition bias in 2 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. v. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.028; Chi² = 45.50, df = 6, p < 0.01; I² = 87%, 75% - 93%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. w. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. x. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. y. One individual randomized trial and 2 cluster randomized trials. z. We downdgraded once. This was due to potential for performance bias (lack of blinding) in all trials and detection bias (self-reported or non-validated outcomes) in 2 trials and unclear risk in one trial. One trial had unclear risk associated with random sequence generation. aa. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.052; Chi² = 39.90, df = 4, p < 0.01 ; I² = 90%, 79% - 95%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. ab. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ac. 7 trials were conducted among MSM, 4 were among male partners of ANC women, 2 were among male truck drivers, and for the remaining sub-analysis among men was included. Meta-analysis: Choko, 2019a and Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined in these trials; Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms and presented ANC data only. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. ad. 11 individual randomized trials and 6 cluster randomized trials. ae. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 11 trials and attrition bias in 2 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Three cluster randomized trials were subject to recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. af. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.084; Chi² = 240.11, df = 17, p < 0.01; I² = 93%, 91% - 95%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. ag. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ah. Three trials were conducted among female sex workers, one was among young women, and for two trials sub-analysis among women was included. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster- adjusted analysis was included in meta-analysis for cluster randomized trials. ai. 2 individual randomized trials and 4 cluster randomized trials. aj. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and detection bias (self-reported or non-validated outcomes) in 3 trials or unclear detection bias in 2 trials. One cluster randomized trial was subject to recruitment bias. Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. Web Annex B 17 ak. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.110; Chi² = 62.95, df = 7, p < 0.01; I² = 89%, 80% - 94%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. al. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. am. Age groups 15-24 years were included. One trial was conducted among young women, one among young MSM, and for three sub-analysis among young people was included. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. an. 2 individual randomized trials and 3 cluster randomized trials. ao. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 4 trials or unclear risk of detection bias in one trial, and attrition bias in one trial (Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. Each of the trials had more than three high risk or unclear risk of bias domains. ap. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.230; Chi² = 82.25, df = 4, p < 0.01; I² = 95%, 91% - 97%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. aq. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ar. Meta-analysis: Choko, 2019a had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting); Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but presented sub-groups separately. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. as. 10 individual randomized trials and 7 cluster randomized trials. at. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 11 trials or unclear risk of detection bias in 3 trials, and attrition bias in 2 trials (Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm; Patel, 2018: 36% LTFU overall, 44% in the intervention and 27% in the control arm). Three cluster randomized trials were subject to recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. au. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.131; Chi² = 351.02, df = 20, p < 0.01; I² = 95%, 93% - 96%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. av. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. aw. Meta-analysis: Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore intervention arms were combined. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. ax. 4 individual randomized trials and 2 cluster randomized trials. ay. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 4 trials or unclear risk of detection bias in one trial, and attrition bias in one trial (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm). Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. 4 of 6 trials had more than three high risk or unclear risk of bias domains. az. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.029; Chi² = 42.71, df = 5, p < 0.01; I² = 88%, 77% - 94%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. ba. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. Web Annex B 18 bb. Meta-analysis: Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore intervention arms were combined. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. bc. 4 individual randomized trials and one cluster randomized trial. bd. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 3 trials and attrition bias in 2 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. be. There was high moderate statistical heterogeneity (Heterogeneity: Tau² = 0.012; Chi² = 12.51, df = 4, p = 0.01; I² = 68%, 17% - 88%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. bf. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. bg. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, in this analysis comparison of one arm (facility-based HIVST distribution) with standard of care was included. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. bh. 4 individual randomized trials and 2 cluster randomized trials. bi. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 3 trials or unclear risk for detection bias in 2 trials, and attrition bias in one trial (Patel, 2018: 36% LTFU overall, 44% in the intervention and 27% in the control arm). There was unclear risk of bias for random sequence generation for Patel, 2018 and Pettifor, 2018. bj. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.068; Chi² = 60.88, df = 5, p < 0.01; I² = 92%, 85% - 96%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. bk. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. bl. Meta-analysis: Choko, 2019a had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore intervention arms were combined. Choko, 2019b had two study groups (ANC women and ART clients), only ANC women group was included in this analysis. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. bm. 2 individual randomized trials and 2 cluster randomized trials. bn. We downgraded once. This was due to potential for performance bias (lack of blinding) and detection bias (self-reported or non-validated outcomes) in all trials. One trial had unclear risk of attrition bias and two cluster randomized trials were subject to recruitment bias. bo. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.133; Chi² = 53.54, df = 3, p < 0.01; I² = 94%, 89% - 97%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. bp. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. bq. Meta-analysis: Choko, 2019b had two study groups (ANC women and ART clients), but only ART group was included in this analysis. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. br. One individual randomized trial and one cluster randomized trial. bs. We downgraded twice. This was due to potential for performance bias (lack of blinding) and detection bias (self-reported or non-validated outcomes) in both trials. Choko, 2019b cluster randomized trial was subject to recruitment bias. Each of the trials had more than three high risk or unclear risk of bias domains. Web Annex B 19 bt. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.211; Chi² = 5.45, df = 1, p = 0.02; I² = 82%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. bu. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. bv. Meta-analysis: Both trials (Chanda, 2017 and Ortblad, 2017) had more than one intervention arm which included HIVST but HIVST kit distribution method was different, only one arm which involved direct distribution of HIVST kits to peers compared to standard of care is included in this analysis. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. bw. 2 cluster randomized trials. bx. We downgraded once. This was due to potential for performance bias (lack of blinding) and detection bias (self-reported or non-validated outcomes) in both trials. by. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.018; Chi² = 5.02, df = 1, p = 0.03; I² = 80%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. bz. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ca. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. cb. 3 cluster randomized trials. cc. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials and detection bias (self-reported or non-validated outcomes) in three trials. Some other risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. All trials had more than three high risk or unclear risk of bias domains. cd. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.122; Chi² = 52.78, df = 3, p < 0.01 ; I² = 96%, 92% - 98%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. ce. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. cf. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. cg. 2 cluster randomized trials. ch. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in one trial or unclear detection bias in 2 trials. Three of 4 trials had unclear risk of bias for random sequence generation. Three of 4 trials had more than three high risk or unclear risk of bias domains. ci. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.274; Chi² = 29.10, df = 3, p < 0.01; I² = 90%, 76% - 95%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. cj. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ck. Meta-analysis: Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore intervention arms were combined. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. cl. 4 individual randomized trials. cm. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 3 or unclear detection bias in one trial and attrition bias in 2 trials (Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm; Patel, 2018: 36% LTFU overall, 44% in the intervention and 27% in the control arm). Each of the trials had more than three high risk or unclear risk of bias domains. Web Annex B 20 cn. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.090; Chi² = 18.22, df = 3, p < 0.01; I² = 84%, 58% - 94%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. co. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. cp. Meta-analysis: Choko, 2019a had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined. Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but presented sub-groups separately. Cluster-adjusted analysis was included in meta- analysis for cluster randomized trials. cq. 5 individual randomized trials and 3 cluster randomized trials. cr. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 6 trials and attrition bias in one trial (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm). Two cluster randomized trials were subject to recruitment bias. Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. cs. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.083; Chi² = 126.61, df = 8, p < 0.01; I² = 94%, 90% - 96%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. ct. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. cu. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. cv. 3 cluster randomized trials. cw. We downgraded once. This was due to potential for performance bias (lack of blinding) and detection bias (self-reported or non-validated outcomes) in all trials. One cluster randomized trial was subject to recruitment bias. One cluster randomized trial had unclear risk of bias for selection bias (allocation concealment), selective reporting, and baseline cluster imbalance. cx. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.054; Chi² = 39.74, df = 4, p < 0.01; I² = 90%, 79% - 95%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. cy. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. cz. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. da. One individual randomized trial and 3 cluster randomized trials. db. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 3 trials or unclear detection bias in one trial. Three of 4 trials had more than three high risk or unclear risk of bias domains. dc. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.121; Chi² = 85.06, df = 3, p < 0.01; I² = 96%, 94% - 98%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. dd. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. de. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. df. 3 individual randomized trials. Web Annex B 21 dg. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and unclear risk for detection bias (self-reported or non-validated outcomes) in 2 trials. Two trials had unclear risk for random sequence generation (selection bias). dh. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.123; Chi² = 10.49, df = 2, p < 0.01; I² = 81%, 40% - 94%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. di. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. dj. We downgraded once due to potential for performance bias due to lack of blinding. dk. Single trial, we downgraded once as inconsistency cannot be evaluated. dl. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. dm. Three trials were included in meta-analysis (Jamil, 2017; Katz, 2018; MacGowan, 2017). All trials were conducted among men who have sex with men in high income countries (Australian and the United States). dn. 3 individual randomized trials. do. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in one trial or unclear risk of detection bias in one trial, and attrition bias in one trial (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm). dp. There was high statistical heterogeneity (Heterogeneity: Tau² = 1.435; Chi² = 63.07, df = 2, p < 0.01 ; I² = 97%, 94% - 98%). Study effects from individual RCTs were consistently beneficial and no difference was observed in other critical outcomes. The GDG determined that downgrading for inconsistency was not necessary. We did not downgrade. dq. Meta-analysis: Choko, 2019a and Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined in these trials; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting); Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but presented sub-groups separately. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. dr. 11 individual randomized and 6 were cluster randomized trials. ds. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 6 trials or unclear risk of detection bias in 4 trials, and attrition bias in two trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Three cluster randomized trials were subject to potential recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. dt. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.117; Chi² = 24.26, df = 18, p = 0.15; I² = 26%, 0% - 57%). We did not downgrade but noted that outcome definition was not consistent across trials: 8 trials defined positivity as confirmed HIV diagnosis; the remaining trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. du. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. dv. Meta-analysis: Choko, 2019a and Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined in these trials; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting); Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but presented sub-groups separately. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. dw. 11 individual randomized and 6 were cluster randomized trials. Web Annex B 22 dx. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 6 trials or unclear risk of detection bias in 4 trials, and attrition bias in two trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Three cluster randomized trials were subject to potential recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. dy. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.048; Chi² = 22.14, df = 19, p = 0.28; I² = 14%, 0% - 49%). We did not downgrade but noted that outcome definition was not consistent across trials: 8 trials defined positivity as confirmed HIV diagnosis; the remaining trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. We also noted that one trial (Nichols, 2019) had significantly lower HIV positivity among those tested in the intervention arm. This trial focused on young people in a setting with high testing coverage. dz. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ea. RR for HIV positivity among randomized: 1.06 (0.76 - 1.48). eb. Meta-analysis: Choko, 2019a had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined; Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but presented sub-groups separately. Cluster-adjusted analysis was included in meta- analysis for cluster randomized trials. ec. 4 individual randomized trials and 4 cluster randomized trials. ed. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 2 trials and attrition bias in 2 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm. One trial had unclear risk of bias for allocation concealment (selection bias). ee. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.283; Chi² = 13.12, df = 8, p = 0.11; I² = 39%, 0% - 72%). We did not downgrade but noted that the outcome definition was not consistent across trials: 4 of 9 trials defined positivity as confirmed HIV positive diagnosis; the remaining 5 trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. ef. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. eg. Meta-analysis: Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined in these trials; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. eh. 7 individual randomized trials and 2 cluster randomized trials. ei. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 4 trials or one unclear risk for detection bias, and attrition bias in 2 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Two trials had unclear risk of selection bias (one for random sequence generation and one for allocation concealment). ej. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 8.93, df = 10, p = 0.54; I² = 0%, 0% - 55%). We did not downgrade but noted that outcome definition was not consistent across trials: 4 of 9 trials defined positivity as confirmed HIV positive diagnosis; the remaining 5 trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. ek. We did not downgrade but noted that only men who have sex with men, transgender women and female sex workers were represented in included studies and results should be viewed with caution when applying to other key populations. el. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. em. Meta-analysis: Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore intervention arms were combined. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. Web Annex B 23 en. 6 individual randomized trials. eo. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 1.62, df = 5, p = 0.90; I² = 0%, 0% - 22%). We did not downgrade but noted that outcome definition was not consistent across trials: 4 of 6 trials defined positivity as confirmed HIV positive diagnosis; the remaining 5 trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. ep. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. eq. Three 3 arm RCTs contributed to this analysis. Kelvin 2019 included two SoC arms, one enhanced and one standard. Enhanced SoC was excluded and standard SoC used in the comparison with the intervention arm. Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST: intervention arms were not pooled as differences in the intervention package appeared likely to impact positivity. er. 2 individual randomized and one cluster randomized trials. es. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 2 trials or unclear risk of detection bias in one trial. One trial had unclear risk of bias for random sequence generation (selection bias). et. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.008; Chi² = 4.24, df = 4, p = 0.35; I² = 6%, 0% - 80%). We did not downgrade. eu. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ev. 6 trials were conducted among MSM, 3 were among male partners of ANC women, 2 were among male truck drivers, and for the remaining sub-analysis among men was included. Meta-analysis: Choko, 2019a and Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined in these trials; Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but included ANC group only in this analysis. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. ew. 10 individual randomized and 3 cluster randomized trials. ex. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 4 trials or unclear risk of detection bias in 2 trials, and attrition bias in 2 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Three cluster randomized trials were subject to recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. ey. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 8.62, df = 12, p = 0.73; I² = 0%, 0% - 40%). We did not downgrade but noted that the outcome definition was not consistent across trials: 7 of 13 trials defined positivity as confirmed HIV positive diagnosis; the remaining 7 trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. ez. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. fa. 3 trials were conducted among female sex workers and for the remaining one sub-analysis among women was included. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. fb. One individual randomized trial and 3 cluster randomized trials. fc. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 3 trials or unclear risk of detection bias in one trial. One cluster randomized trial was subject to recruitment bias. Two trials had unclear risk of selection bias (one random sequence generation and one allocation concealment). fd. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 4.38, df = 5, p = 0.47; I² = 0%, 0% - 71%). We did not downgrade. fe. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ff. Two trials were conducted among young people and other two presented sub-analysis among young people. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. Web Annex B 24 fg. 2 individual randomized trials and two cluster randomized trial. fh. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in three trials or unclear risk of detection bias in one trial, and attrition bias in one trial (Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). One cluster randomized trial was subject to recruitment bias. All trials had unclear risk of selection bias. 3 of 4 trials had more than three high risk or unclear risk of bias domains. fi. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.074; Chi² = 3.45, df = 3, p = 0.33; I² = 13%, 0% - 87%). We did not downgrade. fj. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. fk. Meta-analysis: Choko, 2019a had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting); Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but presented sub-groups separately. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. fl. 7 individual randomized trials and 6 cluster randomized trials. fm. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 5 trials or unclear risk of detection bias in 4 trials, and attrition bias in 2 trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). Three cluster randomized trials were subject to recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. fn. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.056; Chi² = 18.08, df = 15, p = 0.26; I² = 17%, 0% - 54%). We did not downgrade but noted that outcome definition was not consistent across trials: 5 of 14 trials defined positivity as confirmed HIV positive diagnosis; the remaining 9 trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. fo. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. fp. Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore intervention arms were combined. Cluster-adjusted analysis was included in meta- analysis for cluster randomized trials. fq. 4 individual randomized trials. fr. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in one trial and attrition bias in one trial (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm. fs. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 1.20, df = 3, p = 0.75; I² = 0%, 0% - 62%). We did not downgrade but noted outcome definition was not consistent across trials: confirmed HIV positive diagnosis in three trials and self-reported positivity in one trial. ft. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. fu. Meta-analysis: Wray, 2018 had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore intervention arms were combined. fv. 4 individual randomized trials. fw. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 2 trials and attrition bias in two trials (MacGowan, 2017: 27.1% LTFU in the intervention arm and 28.5% in the control arm; Merchant, 2018: 38.4% LTFU overall, 26% in the intervention 50% in the control arm). 3 of 4 trials had more than three high risk or unclear risk of bias domains. fx. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 0.88, df = 3, p = 0.83; I² = 0%, 0% - 48%). We did not downgrade but noted that outcome definition was not consistent across trials: 2 of 4 trials defined positivity as confirmed HIV positive diagnosis; the remaining 2 trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. Web Annex B 25 fy. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. fz. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting) and only HIVST arm with facility-based HIVST distribution is included in this analysis. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. ga. 2 individual randomized and 2 cluster randomized trials. gb. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and detection bias (self-reported or non-validated outcomes) in 2 trials. gc. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.107; Chi² = 4.25, df = 3, p = 0.24; I² = 29%, 0% - 74%). We did not downgrade but noted that outcome definition was not consistent across trials: 2 of 4 trials defined positivity as confirmed HIV positive diagnosis; the remaining 2 trials reported reactive HIVST results as positive, relied on self-reported positivity or the definition was unclear. gd. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ge. Meta-analysis: Choko, 2019a had more than one intervention arm which included HIVST with similar HIVST kit distribution method, therefore all intervention arms were combined; Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but only included ANC group in this analysis. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. gf. One individual randomized and 2 cluster randomized trials. gg. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and unclear risk of attrition bias in one trial. Two cluster randomized trials were subject to recruitment bias and one unclear risk for loss of clusters and not reporting cluster-adjusted analysis (we adjusted for clustering in meta-analysis). gh. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 1.60, df = 2, p = 0.45; I² = 0%, 0% - 87%). We did not downgrade. gi. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. gj. Meta-analysis: Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but only included ART group in this analysis. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. gk. One individual randomized and one cluster randomized trial. gl. We downgraded twice. This was due to potential for performance bias (lack of blinding) in both trials and detection bias (self-reported or non-validated outcomes) in one trial. One trial unclear risk of selection bias (random sequence generation and allocation concealment). One cluster randomized trials were subject to recruitment bias. Each of the trials had more than three high risk or unclear risk of bias domains. gm. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.061; Chi² = 1.36, df = 1, p = 0.24; I² = 26%). We did not downgrade. gn. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. go. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined and only HIVST arm with direct distribution to peers was included in this analysis (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. gp. 2 cluster randomized trials. gq. We downgraded once. This was due to potential for performance bias (lack of blinding) and detection bias (self-reported or non-validated outcomes) in both trials. gr. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 0.27, df = 1, p = 0.61; I² = 0%). We did not downgrade. Web Annex B 26 gs. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. gt. 1 cluster randomized trials. gu. We downgraded once. This was due to potential for performance bias (lack of blinding) and unclear risk of detection bias (self-reported or non-validated outcomes). gv. Single trial - we downgraded once. gw. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. gx. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. gy. 3 individual randomized and one cluster randomized trial. gz. We downgraded twice. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 2 trials or unclear risk of detection bias in 2 trials. Each of the trials had more than three high risk or unclear risk of bias domains. ha. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.667; Chi² = 4.39, df = 3, p = 0.22; I² = 32%, 0% - 76%). We did not downgrade. hb. We downgraded once because confidence intervals around pooled effect estimate are wide, and likely due to small number of events. hc. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. hd. This outcome is only relevant for intervention (HIVST) arm and reported in 9 trials. Results from 7 trials were pooled and reported as pooled percentage. The remaining two trials did not report data usable for pooled analysis. he. Three individual randomized trials and 6 cluster randomized trials. hf. We downgraded once. This was due to potential for performance bias (lack of blinding) in 8 trials, detection bias (self-reported or non-validated outcomes) in 3 trials and attrition bias in one trials. Two cluster randomized trials were subject to recruitment bias and one loss of clusters. Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. hg. There was high statistical heterogeneity (Heterogeneity: I² = 77%, p < 0.01). The measurement time-point varied (range: 2 weeks - 5 months) and point estimate from individual studies also varied (25% - 76%). We downgraded once. hh. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with different linkage interventions - we combined arms with similar linkage interventions; Choko, 2019b had two intervention arms which included HIVST and different linkage intervention - we did not combine arms; Shahmanesh, 2019 had two different interventions - we did not combine arms; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST with different HIVST kit distribution method but no linkage intervention, we combined arms (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. hi. 5 individual randomized and 7 cluster randomized trials. hj. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 5 trials or unclear risk of detection bias in one trial, and unclear attrition bias in one trial. Three cluster randomized trials were subject to recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. hk. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 13.74, df = 15, p = 0.56; I² = 0%, 0% - 47%). Sub-group analysis showed heterogeneity was explained by population type. We did not downgrade but noted that outcome definition was not consistent across trials: 7 trials defined linkage as ART initiation whereas other 5 defined it as linkage to any HIV care. hl. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. Web Annex B 27 hm. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with different linkage interventions - we combined arms with similar linkage interventions; Choko, 2019b had two intervention arms which included HIVST and different linkage intervention - we did not combine arms; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST with different HIVST kit distribution method but no linkage intervention, we combined arms (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. hn. 5 individual randomized and 6 cluster randomized trials. ho. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in 5 trials or unclear risk of detection bias in one trial, and unclear attrition bias in one trial. Three cluster randomized trials were subject to recruitment bias. Several risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. hp. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.011; Chi² = 15.71, df = 13, p = 0.29; I² = 15%, 0% - 53%). Sub-group analysis showed heterogeneity was explained by population type. We did not downgrade but noted that outcome definition was not consistent across trials: 7 trials defined linkage as ART initiation whereas other 5 defined it as linkage to any HIV care. hq. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. hr. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with different linkage interventions - we combined arms with similar linkage interventions; Choko, 2019b had two intervention arms which included HIVST and different linkage intervention - we did not combine arms (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. hs. 2 individual randomized trials and 4 cluster randomized trials ht. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and detection bias (self-reported or non-validated outcomes) in one trial or unclear risk of detection bias in one trial. Three cluster randomized trials were subject to recruitment bias. Three trials had unclear risk of selection bias (one random sequence generation and 3 allocation concealment) and some other risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. hu. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.018; Chi² = 11.29, df = 8, p = 0.19; I² = 29%, 0% - 67%). We did not downgrade but noted that the outcome definition was not consistent across trials: some trials defined linkage as ART initiation whereas others defined it as linkage to any HIV care. hv. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. hw. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. hx. 3 individual randomized trials and two cluster randomized trials. hy. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and detection bias (self-reported or non-validated outcomes) in 2 trials. hz. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 3.24, df = 5, p = 0.52; I² = 0%, 0% - 74%). We did not downgrade but noted that the outcome definition was not consistent across trials: some trials defined linkage as ART initiation whereas others defined it as linkage to any HIV care. ia. We did not downgrade but noted that trials were conducted among men who have sex with men, transgender women and female sex workers, so results should be viewed with caution when applying to other key populations. ib. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ic. 3 individual randomized trials. id. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials. Web Annex B 28 ie. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 0.27, df = 2, p = 0.87; I² = 0%, 0% - 24%). We did not downgrade. if. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ig. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. ih. 2 cluster randomized trials. ii. We downgraded once. This was due to potential for performance bias (lack of blinding) and detection bias (self-reported or non-validated outcomes) in both trials. ij. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 0.63, df = 1, p = 0.43; I² = 0%). We did not downgrade. ik. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. il. 3 studies were conducted among MSM, 3 among male partners of women, and for 2 sub-group analysis for men was included. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with different linkage interventions - we combined all intervention arms for this analysis; Choko, 2019b had two intervention arms which included HIVST and had two sub-groups (ANC women and ART clients) - we combined HIVST arms but only included ANC group in this analysis. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trial. im. 5 individual randomized trials and 3 cluster randomized trials. in. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, and detection bias (self-reported or non-validated outcomes) in one trial or unclear risk of detection bias in one trial. Three cluster randomized trials were subject to recruitment bias. Some other risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. Each of the trials had more than three high risk or unclear risk of bias domains. io. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.045; Chi² = 9.30, df = 7, p = 0.23; I² = 25%, 0% - 66%). We did not downgrade. ip. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. iq. 2 trials were conducted among female sex workers and for 2 trials sub-group analysis among women was included. Meta-analysis: Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. ir. One individual randomized and two cluster randomized trials. is. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and detection bias (self-reported or non-validated outcomes) in 2 trials or unclear risk of detection bias in one trial. Two trials had unclear risk of selection bias (one random sequence generation, two allocation concealment) and some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. it. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 2.48, df = 3, p = 0.48; I² = 0%, 0% - 81%). We did not downgrade. iu. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. iv. For both trials sub-analysis among young people was included in the analysis. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. iw. One individual randomized and once cluster randomized trial. ix. We downgraded twice. This was due to potential for performance bias (lack of blinding) in both trials and unclear risk of selection bias (one random sequence generation and two allocation concealment). One cluster randomized trial was subject to recruitment bias. Each of the trials had more than three high risk or unclear risk of bias domains. Web Annex B 29 iy. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 0.26, df = 1, p = 0.61; I² = 0%). We did not downgrade. iz. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ja. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with different linkage interventions - we combined arms with similar linkage interventions; Choko, 2019b had two intervention arms which included HIVST and different linkage intervention - we did not combine arms; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. jb. 3 individual randomized and 5 cluster randomized trials. jc. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, and detection bias (self-reported or non-validated outcomes) in 4 trials or unclear risk of detection bias in one trial. Three cluster randomized trials were subject to recruitment bias. Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. jd. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.034; Chi² = 17.74, df = 10, p = 0.14; I² = 32%, 0% - 67%). We did not downgrade. je. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. jf. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. jg. 2 individual randomized trials and one cluster randomized trial. jh. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and unclear risk of allocation concealment (selection bias) in one trial. ji. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 0.46, df = 2, p = 0.79; I² = 0%, 0% - 55%). We did not downgrade. jj. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. jk. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with different linkage interventions - we only included the arm with no linkage intervention in this analysis; Choko, 2019b had two intervention arms which included HIVST and different linkage intervention - we only included the arm with no linkage intervention in this analysis; Chanda, 2017 and Ortblad, 2017 had more than one intervention arm which included HIVST but HIVST kit distribution method was different, therefore intervention arms were not combined (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. jl. 4 individual randomized trials and 5 cluster randomized trials. jm. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials and detection bias (self-reported or non-validated outcomes) in 3 trials or unclear risk of detection bias in one trial. Three cluster randomized trials were subject to recruitment bias. Two trials were unclear risk of selection bias (one random sequence generation and two allocation concealment) and some other risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. jn. There was moderate statistical heterogeneity (Heterogeneity: Tau² = 0.058; Chi² = 13.03, df = 8, p = 0.11; I² = 39%, 0% - 72%). Heterogeneity can be explained by difference in study populations. We did not downgrade. jo. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. jp. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with different linkage interventions - we only included the arm with phone reminder for linkage. Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. jq. One cluster randomized trial. jr. We downgraded once. This was due to potential for performance bias (lack of blinding) and high risk of cluster recruitment bias. Web Annex B 30 js. Single trial, we downgraded once as inconsistency cannot be evaluated. jt. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ju. Meta-analysis: Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. jv. 1 cluster randomized trials. jw. We downgraded once. This was due to potential for performance bias (lack of blinding) in both trials and unclear risk of allocation concealment (selection bias) in one trial. jx. Single trial - we downgraded once as inconsistency cannot be evaluated. jy. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. jz. One individual randomized trial. ka. We downgraded once. This was due to potential for performance bias (lack of blinding). kb. Single trial, we downgraded once as inconsistency cannot be evaluated. kc. Confidence intervals around point estimate are wide. We downgraded once. kd. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. ke. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with different linkage interventions - we combined arms with financial linkage interventions and included in the analysis; Choko, 2019b had two intervention arms which included HIVST and different linkage intervention - we did not combine arms and included the arm with financial linkage intervention in this analysis (control arm sample was adjusted to prevent double counting). Cluster-adjusted analysis was included in meta-analysis for cluster randomized trials. kf. 2 cluster randomized trials. kg. We downgraded once. This was due to potential for performance bias (lack of blinding) in both trials and unclear risk of detection bias (self-reported or non-validated outcomes) in one trial. Both cluster randomized trials were subject to recruitment bias. One trial had unclear risk of bias for attrition or loss of clusters and unadjusted cluster analysis (we included cluster adjusted estimates). kh. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 0.08, df = 1, p = 0.77; I² = 0%). We did not downgrade. ki. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. kj. Two trials reported this outcome which is only relevant to the intervention (HIVST) arm. kk. Two cluster randomized trials. kl. We downgraded once. This was due to potential for performance bias (lack of blinding) and detection bias (self-reported or non-validated outcomes) in both trials. One cluster randomized trials was subject to recruitment bias. Some risk of bias domains were unclear risk due to lack of information from unpublished reports or conference abstracts. km. The outcomes were not pooled therefore inconsistency cannot be evaluated. Similar outcomes were reported from both included trials. kn. We did not downgrade but noted that one trial involved facility-based HIVST and one trial involved home-based HIVST distribution by community workers. Both trials were among general population. The results should be viewed with caution when applying to other populations. ko. The results are based on very few events so caution is needed when interpreting. We downgraded once. Web Annex B 31 kp. Two trials were conducted involved HIVST distribution by women to their male partners (Choko, 2019a; Masters, 2016), one involved HIVST distribution by HIV-positive clients to their partners (Dovel, 2019), one involved HIVST distribution by both women to their male partners and HIV-positive clients to their partners (Choko, 2019b), and two involved distribution of HIVST kits or coupons by female sex workers to their peers (Chanda, 2017, Ortblad, 2017). kq. 2 individual randomized and 4 cluster randomized trials. kr. We downgraded once. This was due to potential for performance bias (lack of blinding) and detection bias (self-reported or non-validated outcomes) in all trials. One cluster randomized trials was subject to recruitment bias and one trial had unclear risk of selection bias (random sequence generation and allocation concealment). ks. There was low statistical heterogeneity (Heterogeneity: Tau² = 0; Chi² = 2.96, df = 5, p = 0.71; I² = 0%, 0% - 57%). We did not downgrade but noted that type of adverse events varied across trials (intimate partner violence: Chanda, 2017, Choko, 2019a, Masters, 2016, Ortblad, 2017; temporary relationship breakdown: Choko, 2019b; verbal abuse: Dovel, 2019). kt. All trials involved secondary distribution to partners (in Mulubwa, 2019 a subset of all participants distributed to their partners) or peers so right population of interest for this outcome. ku. Confidence intervals around pooled effect size were large, likely due to very few events. We downgraded once. kv. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. kw. All trials were among men who have sex with men. kx. 3 individual randomized and one cluster randomized trial. ky. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in all trials, and unclear risk of selection bias (allocation concealment) in one trial. kz. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.002; Chi² = 3.39, df = 3, p = 0.34; I² = 11%, 0% - 86%). We did not downgrade but noted that outcome definitions varied across trials (Jamil, 2017: condomless anal intercourse with casual male partner(s) during 12 months follow-up; Katz, 2018: non-concordant condomless anal intercourse with male partner(s) in the past 3 months measured at 15 months; Tang, 2018: condomless anal sex with male partner(s) in the past 3 months measured at 6 months (mid-point during follow-up); Wang, 2017: condomless anal intercourse with male partner(s) in the past 3 months measured at 6 months). la. All trials were among men who have sex with men so results should be viewed with caution when applying to other populations. lb. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. lc. All trials were among men who have sex with men. ld. 3 individual randomized and one cluster randomized trial. le. We downgraded once. This was due to potential for performance bias (lack of blinding) in all trials, detection bias (self-reported or non-validated outcomes) in all trials, and unclear risk of selection bias (allocation concealment) in one trial. lf. There was low statistical heterogeneity (Heterogeneity: Tau² = 0.0003; Chi² = 3.06, df = 3, p = 0.38; I² = 2%, 0% - 85%). We did not downgrade but noted that outcome definitions varied across trials (Jamil, 2017: condomless anal intercourse with casual male partner(s) during 12 months follow-up; Katz, 2018: non-concordant condomless anal intercourse with male partner(s) in the past 3 months measured at 15 months; Tang, 2018: condomless anal sex with male partner(s) in the past 3 months measured at 6 months (mid-point during follow-up); Wang, 2017: condomless anal intercourse with male partner(s) in the past 3 months measured at 6 months). lg. All trials were among men who have sex with men so results should be viewed with caution when applying to other populations. lh. Original sample size, does not represent effective cluster adjusted sample size in the meta-analysis. Web Annex B 32 Web Annex B 33 Table B2. GRADE table for HIVST + linkage intervention compared to HIVST only for linkage to HIV CARE or ART Author(s): Muhammad S. Jamil Question: HIVST + linkage intervention compared to HIVST only for linkage to HIV care or ART Certainty assessment № of patients Effect Certainty Importance № of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations HIVST + linkage intervention HIVST only Relative (95% CI) Absolute (95% CI) Linkage to HIV care or ART initiation 5 a randomised trials b serious c serious d not serious not serious none 308/632 (48.7%) 117/375 (31.2%) RR 1.38 (1.02 to 1.87) 119 more per 1,000 (from 6 more to 271 more) ⨁⨁◯◯ LOW CRITICAL CI: Confidence interval; RR: Risk ratio Explanations a. Meta-analysis: Choko, 2019a had multiple intervention arms which included HIVST with financial incentive for linkage - we combined these arms (HIVST only arm sample was adjusted to prevent double counting); Choko, 2019b: HIVST + financial incentive for linkage vs. HIVST only; MacPherson, 2014: home-based HIVST + optional home ART initiation vs. HIVST + standard linkage; Nichols, 2019: HIVST arm had two linkage interventions (HIVST + self-referral and HIVST + in-person referral/escort); we compared HIVST + in-person referral/escort vs. HIVST + self-referral; Sibanda, 2019 had two intervention arms (HIVST + fixed provider financial incentive and HIVST + fixed provider incentive and conditional incentive per linkage); we compared HIVST + fixed and conditional incentive vs. HIVST + fixed incentive. Cluster-adjusted analysis was included in meta-analysis cluster randomized trials. b. 5 cluster randomized trials. c. We downgraded once. This was due to potential for performance bias (lack of blinding) in 4 trials or unclear risk of performance bias in one trial, high risk of detection bias (self-reported or non-validated outcomes) in one trial. Four cluster randomized trials were subject to recruitment bias (high or unclear risk). Some risk of bias domains were unclear due to limited information from conference abstracts. d. There was high statistical heterogeneity (Heterogeneity: Tau² = 0.097; Chi² = 21.75, df = 5, p <0.01; I² = 77%, 49% - 90%). We downgraded once and noted that study design, type of linkage intervention and comparison group varied across studies. Web Annex B 34