An updated systematic review of epidemiological evidence on hormonal contraceptive methods and HIV acquisition in women

Objective and design: Some studies suggest that specific hormonal contraceptive methods [particularly depot medroxyprogesterone acetate (DMPA)] may increase women's HIV acquisition risk. We updated a systematic review to incorporate recent epidemiological data. Methods: We searched for articles published between 15 January 2014 and 15 January 2016 and hand-searched reference lists. We identified longitudinal studies comparing users of a specific hormonal contraceptive method against either nonusers of hormonal contraception or users of another specific hormonal contraceptive method. We added newly identified studies to those in the previous review, assessed study quality, created forest plots to display results, and conducted a meta-analysis for data on DMPA versus non-use of hormonal contraception. Results: We identified 10 new reports of which five were considered ‘unlikely to inform the primary question’. We focus on the other five reports, along with nine from the previous review, which were considered ‘informative but with important limitations’. The preponderance of data for oral contraceptive pills, injectable norethisterone enanthate, and levonorgestrel implants do not suggest an association with HIV acquisition, though data for implants are limited. The new, higher quality studies on DMPA (or nondisaggregated injectables), which had mixed results in terms of statistical significance, had hazard ratios between 1.2 and 1.7, consistent with our meta-analytic estimate for all higher quality studies of hazard ratio 1.4. Conclusion: Although confounding in these observational data cannot be excluded, new information increases concerns about DMPA and HIV acquisition risk in women. If the association is causal, the magnitude of effect is likely hazard ratio 1.5 or less. Data for other hormonal contraceptive methods, including norethisterone enanthate, are largely reassuring.


Introduction
Empowering women and couples with the tools necessary to prevent unintended pregnancy and avoid sexually transmitted infections including HIV is critically important for individual and public health. Hormonal contraceptive methods are highly effective for prevention of unintended pregnancy and associated sequelae. However, some epidemiological studies suggest an association between use of specific hormonal contraceptive methods [particularly depot medroxyprogesterone acetate (DMPA)] and an increased risk of HIV acquisition in women; other studies have not reported this association [1]. This question is critically important for women's health, particularly in sub-Saharan Africa, where high rates of HIV coincide with high use of injectable contraception [2]. Many regions with high HIV prevalence also have high rates of unmet need for contraception, unintended pregnancy, and maternal mortality and morbidity, underlying the imperative for access to effective contraception [3,4].
Several biologically plausible mechanisms have been postulated to explain how various hormonal contraceptive methods could increase women's risk of HIV acquisition, including possible disruption of epithelial barriers, alterations in immune cell populations, or soluble inflammatory responses [5][6][7][8]. The effect of hormonal contraception on cervical immunity is influenced by the genital tract microenvironment and presence of infections [9]. Interpretation of current data on biologic and immunologic impacts from hormonal contraceptive use is hampered by studies that fail to account for different hormones, diverse dosages, and hormonal contraceptive delivery routes [7]. Women using particular hormonal contraceptive methods may also have other characteristics (e.g. different patterns of condom use), which could impact HIV acquisition risk.
A previous systematic review of epidemiological evidence assessed all relevant evidence published prior to 15 January 2014 [1]. The review was conducted independently of the WHO guidance development process and served as an input into WHO deliberations related to updating the medical eligibility criteria for contraceptive use (refer to Appendix A, http://links.lww.com/QAD/ A969 for current WHO guidance for hormonal contraceptive use among women at high risk of HIV) [10]. Given the public health importance of this topic, we updated our previous systematic review to incorporate newly published, pertinent epidemiological evidence.

Methods
We conducted this systematic review according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [11].
Inclusion/exclusion criteria We included published primary research reports on women who were HIV-negative at baseline in longitudinal studies (observational studies or randomized trials, or metaanalyses containing data not otherwise captured in our search strategy) that measured incident, laboratoryconfirmed HIV infection among women who used a specific method of hormonal contraception [injectables, oral contraceptives, implants, patches, rings, or levonorgestrel intrauterine devices (LNG-IUDs)] compared with incident HIV infections among women using a nonhormonal contraceptive method (e.g. condoms, nonhormonal IUD, sterilization, withdrawal, etc.) or no contraceptive method (henceforth, 'hormonal contraceptive versus non-use of hormonal contraception' comparisons). Some studies compared hormonal contraceptive users against a heterogeneous group including other hormonal contraceptive users, nonhormonal method users, and nonusers of contraception. We identified and included such studies, but considered the composition of the comparison group when assessing study quality.
We also included studies comparing incident HIV infection among HIV-negative women using a specific method of hormonal contraception against HIV-negative women using another specific method of hormonal contraception (henceforth, 'head-to-head' analyses) in which the comparison group did not contain nonhormonal method users or nonusers of contraception.
We excluded studies that did not report a risk estimate for the relationship between hormonal contraceptive use and HIV acquisition, cross-sectional studies, studies assessing only emergency contraception, conference abstracts, or other unpublished reports.

Search strategy
We retained all articles included in the previous systematic review, unless superseded by a new published analysis based upon the same data. We searched PubMed and Embase (Appendix B, http://links.lww.com/QAD/ A969) for articles published in any language between 15 January 2014 and 15 January 2016, inclusive. We hand-searched reference lists of included studies. C.B.P. conducted the literature search and C.B.P., K.M.C., and P.C.H. screened titles, abstracts, and full-text manuscripts to determine inclusion using Covidence software [12].

Data extraction and quality assessment
We applied a study quality assessment framework used in our 2014 systematic review, with slight modifications for clarity [1]. Briefly, studies that did not include adjustment for condom use or which had unclear measurement of exposure to hormonal contraception (refer to Appendix C, http://links.lww.com/QAD/A969 for a full explanation of the quality assessment criteria) were considered 'unlikely to inform the primary question'. For comprehensiveness, we included all studies that met our inclusion Yes criteria, regardless of quality. However, we focused on studies with neither of the two quality concerns noted above; we considered these studies 'informative but with important limitations' (IBWILs) to acknowledge that all studies to date are vulnerable to residual or uncontrolled confounding. All authors participated in confirming the study quality assessment framework and in rating the quality of each study. We adapted previously used abstraction forms that were pilot tested by all coauthors. All coauthors abstracted data from each newly included study that was considered as IBWIL. We contacted study investigators if clarifications were needed. Most studies estimated hazards ratios using Cox proportional hazards models; some also included estimates from a marginal structural model (MSM) (for additional discussion, refer to [1,13]). A few estimated only incidence rate ratios (IRRs) ( Tables 1 and 2). For clarity of presentation, we display the IRR or Cox hazards ratio, unless the MSM model generated qualitatively different estimates, in which case both Cox and MSM estimates are shown.

Graphical summaries
As in 2014, we requested disaggregated estimates from authors of new studies classified as IBWIL and which included women from South Africa (where use of both DMPA and NET-EN is common) but which did not report separate estimates for each. Disaggregated estimates have reduced statistical power but greater epidemiological and clinical value, given the potential for different biological effects by contraceptive type or formulation.

Meta-analysis
Given concerns specific to DMPA, we performed a statistical meta-analysis for the effect of DMPA versus non-use of hormonal contraception on HIV acquisition (studies that did not disaggregate injectables were not included). For maximum comparability, we included the most fully adjusted Cox hazards ratio estimates from each study, except one that reported an adjusted IRR (IRRs can be interpreted similarly to hazards ratios under certain conditions [14]). We log-transformed reported adjusted point estimates and 95% confidence intervals (95% CIs) to calculate standard errors using a random effects model [15]. We assessed statistical heterogeneity using the I 2 statistic [16]. Analyses were performed using Stata (Version 13.1, College Station, Texas, USA).

Description of included studies
Twenty-two studies were included in our previous review [1]. For this review, we screened 312 new references, assessed 14 full-text reports, and excluded four: two did not report on the association of interest [17,18] and two meta-analyses contained published data already captured by our search strategy (including them would have resulted in double-counting of data, instead they are mentioned in our discussion) (Fig. 1) [19,20].
We included 10 new reports [21][22][23][24][25][26][27][28][29][30]; one [21] superseded a previously included study [31]. A large, individual participant data (IPD) meta-analysis [26] used raw data from 18 datasets, including seven not previously utilized to investigate the association of interest [17,[32][33][34][35][36][37]. To incorporate the previously unpublished information (while avoiding double-counting from previously published studies), we requested a subanalysis restricted to data from these seven studies in a hormonal contraceptive versus non-use of hormonal contraception comparison [38]. The IPD meta-analysis also included a head-tohead comparison that none of our included component studies had assessed; here we used results from the original article [26].  [21,26,27,29]. Table 2 provides details on new IBWIL studies; information on previously included IBWIL studies is available elsewhere [1]. The four new studies included a large IPD metaanalysis that assessed oral contraceptives, DMPA, and NET-EN across a range of datasets [26], an analysis from an 18-year cohort study of Zambian serodiscordant couples to assess oral contraceptives, DMPA, and implants [27], and two analyses from large microbicide trials, one assessing unspecified injectables [29] and the other assessing oral contraceptives, DMPA, and NET-EN [21]. Below, we summarize results from all 12 hormonal Reports excluded based on full text review (n = 4) Did not report on association between hormonal contraceptives and HIV acquisition (n = 2) • • Meta-analysis, did not provide previously unpublished data (n = 2) New reports included (n = 10) One new study (Crook 2014) replaces a study (Wand 2012) included in previous systematic review due to being based on fuller dataset of same trial contraceptive versus non-use of hormonal contraception studies considered IBWIL. Readers should consult the relevant tables and figures for additional detail (such as 95% CIs); descriptions below provide a succinct synthesis of the overall evidence base. We discuss studies according to whether results were significant at P less than 0.05, but acknowledge that, considered alone, P values are an imperfect indicator of significance [63].

Head-to-head studies considered informative but with important limitations
No head-to-head comparison studies were available in the previous review [1]. Both newly included head-tohead studies were considered IBWIL (Tables 1 and 2) [26,30]. Both reported a statistically significant increased risk of HIV for DMPA use (adjHR: 1.32 and 1.41) versus NET-EN use [26,30]. The IPD meta-analysis also compared each injectable against COCs, reporting significantly increased risk for DMPA versus COCs  (Fig. 6) [26].

Meta-analysis
Ten estimates, from nine published studies with DMPAspecific estimates versus non-use of hormonal contraception [21,27,39,42,47,51,53,55,56] and a subanalysis of previously unpublished information from an IPD metaanalysis [26], were included in our meta-analysis of the effect of DMPA on HIV acquisition (Appendix D, Fig. 5   Error bars show 95% confidence intervals. Studies arranged in order of decreasing magnitude of risk estimate, except if a single study disaggregated depot medroxyprogesterone acetate and norethisterone enanthate, in which the case both estimates are adjacent (as indicated by a box around the study identifiers). Graph does not display estimates from marginal structural models, except where use of such models resulted in different conclusion regarding statistical significance; in such cases, estimates from both models are displayed on a single line (also identified by bracket signs). Note: displays all data on injectables (depot medroxyprogesterone acetate, norethisterone enanthate, or unspecified). adjIRR, adjusted incidence risk ratio; adjHR, adjusted hazard ratio. Ã Analysis showed significant findings at P ¼ 0.05 (marker also displayed in red). y Estimate for Cox model taken from slightly updated analysis which controlled for total number of unprotected sex acts.^Unpublished estimates from a subanalysis of Morrison et al. [26] meta-analysis, restricted to pooled analysis using databases not previously used to publish estimates on hormonal contraceptive methods and HIV acquisition risk. ¥ Different statistical models adjusted for slightly different confounders. § Unpublished estimates disaggregated by injectable type.
older women [52]; eight studies reported no effect modification by age [21,27,29,39,42,47,51,56]; most studies reported no effect modification by herpes simplex virus type 2 (HSV-2) status [21,29,39,42,51], whereas two reported effect modification in opposite directions [One observed higher HIV risk with DMPA in HSV-2 seronegative women (Morrison et al. [52]) and the other observed higher HIV risk with DMPA (versus NET-EN) in HSV-2 seropositive women (Noguchi et al. [30]).]. Two studies reported no effect of modification by study site [21,30], one reported greater risk for oral contraceptives and DMPA in a Ugandan site versus a Zimbabwean site [53]. A study in serodiscordant couples reported no effect modification for genital ulceration, inflammation, viral load of HIV-positive partner at baseline, or fertility intentions [27]. Another study reported no effect modification by reported condom use at baseline, participant behavioral risk, or prevalent chlamydia or gonorrhea [53].
Within the IPD meta-analysis, assessment for effect modification was conducted with information from all 18 studies (some of which were also included in our review). No evidence of interaction was reported with any method for age (15-24 versus >25 years), HSV-2 status at baseline, or HIV incidence in population (low versus high) [26].   Fig. 4. Use of depot medroxyprogesterone acetate (or unspecified injectable) versus non-use of hormonal contraception and HIV acquisition, among 12 studies considered informative but with important limitations. Error bars show 95% confidence intervals. Studies arranged in order of decreasing magnitude of risk estimate. Graph does not display estimates from marginal structural models, except where use of such models resulted in different conclusion regarding statistical significance; in such cases, estimates from both models are displayed on a single line (also identified by bracket signs). adjIRR, adjusted incidence risk ratio.
Note: restricts to all data on depot medroxyprogesterone acetate or unspecified injectables, as estimates of unspecified injectables are likely comprised largely of DMPA. adjHR, adjusted hazard ratio. Ã Analysis showed significant findings at P ¼ 0.05 (marker also displayed in red). y Estimate for Cox model taken from slightly updated analysis which controlled for total number of unprotected sex acts.^Unpublished estimates from a subanalysis of Morrison et al. [26] meta-analysis, restricted to pooled analysis using databases not previously used to publish estimates on hormonal contraceptive methods and HIV acquisition risk. ¥ Different statistical models adjusted for slightly different confounders. § Unpublished estimates disaggregated by injectable type.
0.47-0.96). Table 3 details how the IPD meta-analysis investigators defined lower risk of bias in comparison with our quality criteria.

Discussion
Interpretation of overall results As in our 2014 review, current data do not suggest an increased risk of HIV acquisition among women using oral contraceptives [1]. Extremely limited data do not suggest a statistically significant increased risk of HIV acquisition among users of levonorgestrel implants; no data are available regarding etonogestrel implants. In 2014, one of five studies that was considered IBWIL suggested an increased risk of HIVacquisition with NET-EN injectables [31]. In this updated review, that study was replaced by a larger, more sophisticated analysis of the same dataset [21], and increased HIV risk was no longer observed. Thus, currently available data for injectable NET-EN use do not suggest an association with HIV acquisition in women.
Although concerns around confounding in observational data remain relevant, newly available evidence regarding injectable DMPA use increases concern about a potential causal association with HIV acquisition. Twelve studies considered IBWIL assessed DMPA or nondisaggregated injectables compared with non-use of hormonal contraception; four or five (depending on the statistical model considered) reported statistically significant increased risks of HIVacquisition, ranging from adjHR 1.45 to 2.04 in Cox models (or 2.19 in MSM models). Among four newly included studies, two reported statistically significant increased risk (adjHR: 1.45 and 1.69), including one very large study [21] and a subanalysis of a large IPD meta-analysis [26]. A smaller study among serodiscordant couples [27] reported a nonsignificant estimate of adjHR 1.34, and data from a microbicide trial also had a nonsignificant estimate of adjHR 1.17 but did not disaggregate between injectables [29]. Head-to-head comparisons were newly available and may be less confounded by unmeasured or residual behavioral differences than comparisons from hormonal contraceptive versus non-use of hormonal contraception studies, particularly if groups compared in head-to-head studies use different types of the same delivery method (i.e., injectable DMPA versus injectable NET-EN) [30]. A head-to-head analysis of VOICE data reported a 41% increased risk of HIV acquisition in DMPA versus NET-EN users [30]. In the IPD meta-analysis [26], DMPA use was associated with a significantly increased risk of HIV acquisition of 30-40% when compared with either NET-EN or COC use. Comparing NET-EN against COC users suggested higher risk with NET-EN, though this was not statistically significant at P less than 0.05 (P ¼ 0.055). Although residual confounding cannot be ruled out in any observational study, several recently published studies suggesting an increased risk of HIV acquisition among DMPA users had few limitations apart from being observational ( Table 2).
Recent analyses contradict the hypothesis that differential over-reporting of condom use by hormonal contraceptive users explains observed associations between hormonal contraceptive use and HIV infection in some studies [64]. However, the possibility remains that certain confounders are specific to DMPA users. In South Africa (where both DMPA and NET-EN injectables are used), studies suggest that women perceive DMPA and NET-EN differently, and providers may preferentially prescribe different injectable types to specific subpopulations, which could result in confounding specific to injectable type [30,65,66]. Although beyond the scope of this review, emerging evidence related to DMPA and HSV-2 acquisition must also be considered [67,68].
Taken together, the new evidence points toward heightened concerns that the association between DMPA use and HIV acquisition may not be fully explained by confounding or other methodological problems. In contrast, additional reassuring evidence of no significant association for other hormonal contraceptive methods (oral contraceptives, NET-EN, and implants) is newly available. If the association between DMPA and HIV acquisition risk is causal, meta-analyses, including our own, suggest a likely increase in risk of hazards ratio 1.5 or less.
The quality of epidemiological evidence on this issue has improved over time. Several newly published studies used recent analytic recommendations [13] or other innovative analytic techniques. For example, Crook et al. [21] conducted a particularly thorough exploration of statistical methodology and incorporated multiple sensitivity analyses to assess the robustness of their findings, Morrison et al. [26] contributed substantial new data in a carefully conducted IPD meta-analysis, and Noguchi et al. [30] examined an alternative comparison group (NET-EN users).
The methodological contribution of three newly published meta-analyses varied. In addition to the IPD meta-analysis included in our review [26], two meta-  Fig. 6. Hormonal contraceptive methods and HIV acquisition in head-to-head studies, among two studies considered informative but with important limitations. Error bars show 95% confidence intervals. Within each comparison group, studies are arranged in order of decreasing magnitude of risk estimate. adjHR, adjusted hazard ratio. Ã Analysis showed significant findings at P ¼ 0.05 (marker also displayed in red). Table 3. Inclusion and quality rating of publications and databases across systematic reviews and meta-analyses assessing the association of injectables (versus nonuse of hormonal contraception) with risk of HIV acquisition in women.

Current systematic review and metaanalysis
Ralph et al. [19] meta-analysis Brind et al. [20] meta-analysis (longitudinal data) Morrison et al.    analyses [19,20] utilized data already included as primary studies in our systematic reviews (thus, adding no information beyond that already included in this review). Although all three meta-analyses reported summary estimates for DMPA similar to our own (hazards ratio 1.4-1.5), one of the excluded meta-analyses contained no assessment of study quality and included several studies with serious methodological limitations [20], which raises particular concern in the context of meta-analysis of observational data (Table 3) [69]. Both excluded metaanalyses [19,20] double-counted [70] data by inclusion of both Wand and Ramjee [31] and Crook et al. [21]. We generated a meta-analytic estimate for DMPA, but recommend that such results be interpreted with caution, given the potential for spurious precision in meta-analyses of observational data [71]. The I 2 value for our metaanalysis suggested minimal statistical heterogeneity, but qualitative differences between study populations and methods remain an important consideration [16]. That said, estimates from all four meta-analyses are similar, despite inclusion of slightly different component studies [26].

Limitations
Previous reviews have addressed key methodological considerations about this body of literature, including potential for confounding, frequency, and accuracy of variable measurement, considerations related to 'direct' and 'total' effects, potential for publication bias, and limitations of individual studies, such as failure of some studies to disaggregate by specific hormonal content or formulation (e.g., most studies assessing oral contraceptives failed to disaggregate estimates by COCs or POPs) [1,72]. Our study quality framework is necessarily subjective, and we encourage continued discussion on how best to evaluate study quality in this body of evidence.

Conclusion
There remain no data on use of contraceptive patches, rings, or hormonal IUDs and HIV acquisition in women. For implants, very limited data pertaining to levonorgestrel implants do not suggest increased risk, but more information is needed. In comparison, a larger amount of data are available for oral contraceptives and are generally reassuring. A growing number of studies have assessed injectable NET-EN, and although still limited, data are generally reassuring. For injectable DMPA, although some new, high-quality studies do not report a statistically significant increased risk of HIV acquisition, other new data, including studies directly comparing DMPA and NET-EN, tend to strengthen concerns about DMPA. If the association between DMPA and HIV acquisition risk is causal, data suggest a likely increase in risk of hazards ratio 1.5 or less. Several new studies have used recently proposed recommendations for analysis or other innovative methodological approaches [13], although as with all observational data, the possibility of uncontrolled or residual confounding remains. The growing, generally reassuring evidence about other hormonal contraceptive methods, including other injectables like NET-EN, stands in contrast to the DMPA-specific findings. An important next step is for WHO to determine whether these concerns warrant a reconsideration of global guidance for DMPA. Modeling studies can be useful in understanding net health impacts of various policy responses in different epidemiological contexts, including the risk of HIV, maternal mortality and morbidity, and access to alternative contraception and HIV prevention methods [2,[73][74][75][76].