Table 2.
Comparison of newly included studies (published since 15 January 2014) considered ‘informative but with important limitations’, for systematic review update on use of various hormonal contraceptive methods among women at risk of HIV acquisition.
| Study, study population | Number seroconverted/number analyzed, number of seroconverters by exposure group, overall HIV incidence | Interval between visits, length of f/u, loss to f/u, and whether f/u was differential by HC status | Referent group (including overall proportion of condom use in population) | Handling of condom use | Results | Summary of strengths | Summary of weaknesses |
| Direct evidence from HC versus no-HC studies | |||||||
| Crook 2014 [21], S. Africa, Uganda, Tanzania, Zambia | 382/8663 seroconverted. 265 seroconversions among women using a method of HC at baseline (146 DMPA, 69 NET-EN, 50 OCs), 117 seroconversions in women using no HC at baseline. 4.7/100 person-years | 4 weekly-intersurvey interval. F/u continued for 52 weeks. Loss to f/u: 9% for DMPA, 10% for NET-EN, 11% for OCs, 10% for no HC; not differential by group | Percentage of non-HC group using each method at baseline: male or female condoms for family planning (50%), natural or traditional methods (4%), sterilization (1%), IUD (1%), no contraception (44%) | Controlled for condom use at last sex act (baseline and f/u every 4 weeks) | Time-updated covariate model including time-varying exposure: DMPA adjHR 1.45 (1.09–1.93); NET-EN adjHR: 1.20 (0.84–1.69); OC adjHR: 0.90 (0.63–1.26); IPW models; DMPA adjHR: 1.49 (1.06–2.08); NET-EN adjHR: 1.31 (0.86–1.99); OC adjHR: 1.00 (0.62–1.61); Additional models in Table 4 of manuscript | Large number of incident infections. Multisite study. Low loss to f/u. Thorough exploration of statistical methodology and sensitivity analyses, all suggesting similar results. Disaggregation of DMPA and NET-EN. Short intersurvey intervals (4 weeks). 9% of sample were serodiscordant couples. Findings consistent across sites | No information on partner's HIV status for most participants. Unable to separate POPs and COCs. Potential for residual/unmeasured confounding |
| Wall 2015 [27], Zambia | 252/1393 seroconverted; 99 seroconversions (linked and unlinked) in women using HC (49 OC, 41 injectable, 9 implant), 153 in women using no HC. 74 seroconversions (linked) in women using HC (35 OC, 33 injectable, 6 implant), 133 in women using no HC. 8.9/100 person-years | 3 month intersurvey interval (with a subset of participants followed monthly for HIV testing). Study took place over 17 years, median f/u 440 days (interquartile range 756). Loss to f/u unclear | Non-HC group comprised of individuals using condoms; copper IUD; hysterectomy, tubal ligation, or vasectomy; or no method. Proportions not described | Controlled for unprotected sex in last 3 months in analysis of linked infections. No control for condoms in analysis of linked and unlinked infections | Implants adjHR: 0.96 (0.29–3.14); DMPA adjHR: 1.34 (0.85–2.12); OCs adjHR: 1.39 (0.90–2.15); (Note: we display results from the analysis on incidence HIV infections genetically linked to the cohabitating male partner. Although this model had less statistical power, it was the most fully adjusted model, including controls for condom use as well as partner viral load.) | Analysis of serodiscordant couples and control for partner HIV characteristics. Large number of incident seroconversions. Included clinical characteristics of partners, such as viral load. Short intersurvey intervals (3 months). Examined presence of sperm on a vaginal swab wet prep. Long-term f/u. Conducted multiple sensitivity analyses to assess whether findings were robust to various assumptions | Various study quality components were poorly described, including: loss to f/u, number of couples with only one f/u visit, differences between exposure groups, how variables collected inconsistently over study duration were handled analytically; composition of the reference group; how information on contraceptive exposure was collected, and statistical power. Unable to separate POPs and COCs. Potential for residual/unmeasured confounding |
| Balkus 2016 [29], Malawi, S. Africa, Zambia, Zimbabwe (US site excluded) | 106/2830 seroconverted; 88 seroconversions in women using HC (72 injectable, 15 OCs). 19 seroconversion in women using no HC. 4.07/100 person-years | Pregnancy tests monthly, HIV and contraceptive info quarterly, HSV info at baseline and study exit; 12 month f/u. Loss to f/u unclear | % of non-HC group using each method at baseline: condoms (58%), sterilization (14%), no contraceptive method (28%) | Controlled for unprotected sex at last vaginal intercourse | Injectables adjHR: 1.17 (0.70–1.96); OCs adjHR: 0.76 (0.37–1.55) | Large sample size, multisite study, short intersurvey intervals (between monthly and 3 monthly) | Didn’t differentiate between injectables or OC type. Loss to f/u unclear. No control for study arm. Potential for residual/unmeasured confounding |
| Morrison 2015 IPD meta-analysis [26,38] (and subanalysis of seven databases); East and Southern Africa | 1830 incident seroconversions in data from full IPD meta-analysis; DMPA: 5.1/100 woman-years; NET-EN: 4.8/100 woman-years; COCs 3.4/100 woman-years; No HC 3.9/100 woman-years | Ranged from monthly to every 6 months in full IPD meta-analysis | No HC (condoms, sterilization, nonhormonal IUD, diaphragm, no modern method) | Controlled for condom use, (parameterization unspecified) | Full IPD meta-analysis [26]: COC adjHR: 1.07 (0.91–1.25); DMPA adjHR: 1.52 (1.27–1.82); NET-EN adjHR: 1.27 (0.99–1.61); Subanalysis of seven previously unpublished studies (two-stage random effects model) [38]; COC adjHR: 0.79 (0.38–1.64); DMPA adjHR: 1.69 (1.02–2.78); NET-EN adjHR: 1.58 (0.66–3.79) | IPD meta-analysis included both published and previously unpublished data. Represents the largest analysis to date of this subject, and offered a consistent approach to coding and multivariable analysis across datasets. Multisite (by nature of inclusion of studies from various settings). Extremely high statistical power permitted ability to conduct several key subgroup analyses. Numerous sensitivity analyses which generally supported overall findings (except study quality and region) | For most studies, no information on partner HIV status; variable length of intersurvey interval. Quality ranking of studies (for higher versus lower risk of bias) is necessarily subjective (and discrepant with our study quality criteria). Potential for unmeasured/residual confounding |
| Indirect evidence from head-to-head studies | |||||||
| Noguchi 2015 [30], South Africa | 207 seroconversions in 2733.7 person-years, for an incidence of 7.57/ 100 woman-years. 152/1763 person-years of DMPA (incidence: 8.62/ 100 woman-years) and 55/970.8 woman-years of NET-EN (incidence: 5.67/ 100 woman-years) | Monthly | NET-EN users | Condom use at last sex, assessed monthly | DMPA versus NET-EN adjHR = 1.41 (1.06–1.89) | Large prospective study, careful documentation of exposure to injectables, use of ACASI, adjustment for variety of time-varying covariates, monthly intersurvey intervals, head-to-head comparison may be less likely confounded by behavioral differences, multiple sensitivity analyses generally supported overall findings. Low loss to f/u | No information on partners’ HIV status. Loss to f/u differential by comparison arm. Head-to-head comparisons cannot assess whether DMPA increases risk of HIV acquisition relevant to no hormonal contraception; underlying risk of comparison group is uncertain. Potential for residual/unmeasured confounding |
| Morrison 2015 [26] IPD | 1830 incident seroconversions in data from full IPD meta-analysis; DMPA: 5.1/100 woman-years; NET-EN: 4.8/100 woman-years; COCs 3.4/100 woman-years; No HC 3.9/100 woman-years | Ranged from monthly to every 6 months in full IPD meta-analysis | Head-to-head comparisons included: DMPA versus COC; DMPA versus NET-EN; NET-EN versus COC | Controlled for condom use, (parameterization unspecified) | DMPA versus COC adjHR: 1.43 (1.23–1.67) (17 studies included); DMPA versus NET-EN adjHR: 1.32 (1.08–1.61) (8 studies included); NET-EN versus COC adjHR: 1.30 (0.99–1.71) (8 studies included) | IPD meta-analysis included both published and previously unpublished data. Represents the largest analysis to date of this subject, and offered a consistent approach to coding and multivariable analysis across datasets. Multisite (by nature of inclusion of studies from various settings). Extremely high statistical power permitted ability to conduct several key subgroup analyses. Numerous sensitivity analyses which generally supported overall findings (except study quality and region). Head-to-head comparisons may be less likely confounded by behavioral differences | For most studies, no information on partner HIV status; variable length of intersurvey interval. Quality ranking of studies (for higher versus lower risk of bias) is necessarily subjective (and discrepant with our study quality criteria). Head-to-head comparisons cannot assess whether various HC methods increase risk of HIV acquisition relevant to no hormonal contraception; underlying risk of various comparison groups is uncertain. Potential for unmeasured/residual confounding |
Note: Please refer to 2014 systematic review for detail on previously included studies [1]. adjHR, adjusted hazard ratio; COCs, combined oral contraceptive pills; DMPA, depot medroxyprogesteone acetate; f/u, follow up; f/u, follow-up; HC, hormonal contraception; HIV, human immunodeficiency virus; IPD, individual participant data; IPW, inverse probability weighted; IUD, intrauterine device; NET-EN, norethisterone enanthate; OCs, oral contraceptive pills; POPs, progestin-only pills.