Hormonal contraceptive use and women's risk of HIV acquisition: a meta-analysis of observational studies - Editorial/Comment
Download the PDF here
Download the PDF here
Download the PDF here
3 Articles Below: Editorial, Comment & the original article:
Hormonal contraceptive use and women's risk of HIV acquisition: a meta-analysis of observational studies
"Our analysis also provides an insight into potential sources of heterogeneity in results. Studies in women in the general population, representing most of the studies in our analysis, provide estimates of the average population effect of hormonal contraception on women's risk of HIV acquisition. By contrast, studies in women at high risk, of which there were two in our analysis, provide estimates of the effect of hormonal contraception conditioned on a high likelihood of HIV exposure. For the millions of users of hormonal contraception worldwide, most of who are not in serodiscordant or other high-risk partnerships, this distinction is crucial. Although the increased risks for users of depot medroxyprogesterone acetate and users of combined oral contraception or oral contraceptive pills reported in the two studies with commercial sex workers39 and women in serodiscordant partnerships16 might warrant consideration of changing contraceptive guidelines for these populations, it would be too early to do so based on the results of just two studies. Furthermore, it is important that the results of these two studies are not generalised to women in the general population, who according to the results of our study had a smaller increase in risk that might only warrant a policy change in specific local contexts."
from Comment: "The investigators argue that their findings confirm those from previous research of an increased risk and suggest that their pooled estimates can be used immediately to guide decision making about the use of depot medroxyprogesterone acetate."
End of the debate on hormonal contraception and HIV risk?
Lancet Infectious Disease Feb 2015
Contraceptives are life-saving interventions because they prevent maternal and neonatal mortality from unwanted pregnancies. Worldwide, birth control (all forms) is estimated to reduce maternal deaths by 44% (272 000 deaths averted in 2008), and could prevent an additional 29% of such deaths if the full demand for contraception was met. Since their introduction in 1960, hormonal contraceptive methods have revolutionised the lives of many millions of women of child-bearing age, allowing them to take control of their own fertility. However, unlike barrier methods such as condoms, hormonal contraceptives do not protect against sexually transmitted infections such as HIV. Indeed, whether hormonal contraceptives might actually increase the risk of HIV acquisition has been debated since the early 1990s. Concern focuses particularly on the injectable drugs depot medroxyprogesterone acetate (DMPA) and norethisterone enanthate (NET-EN).
In an Article in this issue, Lauren Ralph and colleagues attempt to resolve the debate with a meta-analysis of observational studies of hormonal contraceptive use and women's risk of HIV acquisition. The researchers identified 12 relevant studies with a total of almost 40 000 women. Compared with non-hormonal or no contraception, DMPA use was associated with a 40% increased risk of HIV acquisition (hazard ratio 1·40, 95% CI 1·16-1·69). The risk was lower (1·31, 1·10-1·57) when only the general population was considered, after exclusion of women in high-risk groups such as sex workers and those in serodiscordant partnerships. No increased risk of HIV acquisition was found in women using oral contraceptive pills (1·00, 0·86-1·16) or NET-EN (1·10, 0·88-1·37).
Do Ralph and colleagues' findings close the debate on whether hormonal contraceptives increase the risk of HIV acquisition? Many experts will argue that a definitive answer is impossible when the available data come from observational studies. Despite the best efforts of investigators to minimise confounding, the results of observational studies are prone to bias introduced by unmeasured variables that differ between groups. For example, women using DMPA might have more frequent sexual intercourse or be less likely to use condoms than women using other contraceptive methods, both of which would increase the risk of HIV infection irrespective of any specific effect of DMPA.
Because of such uncertainty in the data, the debate around whether hormonal contraception increases risk of HIV acquisition has become polarised and heated, as pointed out by Christopher Colvin and Abigail Harrison in the Comment that accompanies the Article by Ralph and colleagues. A randomised controlled trial, usually regarded as providing the highest level evidence, appears to be the best way of getting a definitive answer. Indeed, the ECHO consortium is planning a trial that will randomise 7800 women in sub-Saharan Africa to receive contraception by either DMPA, progestin implant, or non-hormonal copper intrauterine device.
The very idea of the ECHO trial has faced substantial criticism. Critics question whether it is ethical to conduct a trial for which the primary outcome is harm in the form of HIV infection. Furthermore, they argue that because the available data show that DMPA increases the risk of HIV acquisition, the trial groups would not be in equipose at the point of randomisation. Trial participants could not be masked to contraceptive method, so another concern relates to whether women would be compliant with their allocated method. Whether the findings of the trial would lead to a change in practice has been questioned, because modelling analyses suggest that even if a doubling of risk was found, the benefits of DMPA in terms of maternal mortality would still outweigh the increased number of HIV infections. Writing in The Lancet last year, Gollub and Stein said: "A randomised trial appears to be a high-risk (for failure) venture for marginal gain." At present, the start of the ECHO trial has been delayed by problems in finding sufficient funding.
In sub-Saharan Africa, injectable hormonal contraceptives (particularly DMPA) are among the most widely used forms of birth control. Irrespective of whether a randomised trial eventually provides conclusive evidence on the risk of HIV acquisition related to this form of contraception, women will still need access to effective control over their fertility. Alternative-and more effective-methods, such as hormonal implants and intrauterine devices, already exist. Any moves to withdraw DMPA must go hand in hand with roll-out of alternative contraceptive methods.
Broadening the debate over HIV and hormonal contraception
Lancet Infectious Disease 2015
*Christopher J Colvin, Abigail Harrison
Division of Social and Behavioural Sciences, School of Public Health
and Family Medicine, Falmouth 3.46, Health Sciences Campus,
University of Cape Town, Observatory 7925, Cape Town, South
Africa (CJC); and Department of Behavioral and Social Sciences,
Brown University School of Public Health, Providence, RI, USA (AH)
The question of whether hormonal contraception, particularly depot medroxyprogesterone acetate, increases a woman's risk of acquiring HIV has been debated since an association was first noted in 1991.1 Subsequent data from observational studies, secondary analyses of trials, and systematic reviews2, 3, 4 largely support the view that depot medroxyprogesterone acetate makes a moderate contribution to HIV risk. Efforts to synthesise existing evidence, however, have shown significant heterogeneity and serious, uncontrolled risk of confounding.5, 6, 7
In The Lancet Infectious Diseases, Lauren Ralph and colleagues8 extend this debate by providing the first pooled estimate of the association between depot medroxyprogesterone acetate and HIV acquisition, drawn from a meta-analysis of the most up-to-date observational evidence. They conclude that depot medroxyprogesterone acetate use increases a woman's chance of acquiring HIV by 40% (pooled hazard ratio 1·40, 95% CI 1·16-1·69). In studies with women already at high risk of infection, much higher effects of depot medroxyprogesterone acetate on HIV risk were reported, although heterogeneity prevented calculation of a pooled estimate. Ralph and colleagues also did a priori secondary analyses to test and substantiate the robustness of their pooled estimates.
Ralph and colleagues' findings add an important element to the longstanding debates about the HIV and depot medroxyprogesterone acetate relation, and the growing calls for further evidence about the magnitude, mechanisms, and health effects of this link. The investigators argue that their findings confirm those from previous research of an increased risk and suggest that their pooled estimates can be used immediately to guide decision making about the use of depot medroxyprogesterone acetate.
Currently, the increasingly narrow and fierce debates over the HIV and depot medroxyprogesterone acetate link have focused on whether a large randomised controlled trial should be done to better understand this link.5 Like many scientific controversies, views have become hardened, personal, financial, or political agendas have been suggested, and there has even been intrigue in the form of leaked copies of articles under peer review.9 Both sides have raised important, compelling arguments, but their partisan character can weaken the quality of the debates and restrict the view of the complex relation between evidence, policy, and practice.
Trial proponents have argued that only a randomised trial can be used to answer the question about a link in a way that is both scientifically valid and credible to policy makers.10, 11 They argue that equipoise exists as to whether hormonal contraception affects HIV risk, and that a trial is feasible (or at least its feasibility is an open question). Their focus on the quality of the evidence from randomised controlled trials, however, tends to sidestep the many serious methodological challenges of trials like these, challenges recently faced in the VOICE and MIRA trials.12 Proponents have argued convincingly about the urgent need to avoid thousands of preventable HIV infections in women using depot medroxyprogesterone acetate. In doing so, however, they have tended to downplay other health risks of not using depot medroxyprogesterone acetate.
Critics of the proposed randomised controlled trial maintain that, with the existing evidence, there is no equipoise and the trial would be unethical. They argue that existing policy options and further modelling research could effectively mitigate the risk of depot medroxyprogesterone acetate and that the opportunity costs of the trial would be huge. Additionally, there are complex tradeoffs between effective HIV prevention and effective family planning. Also they point out that hard-fought progress in increasing access to contraception, especially in sub-Saharan Africa, will be threatened if governments and health-care workers abandon support for depot medroxyprogesterone acetate based on insufficient evidence. Critics, however, tend to gloss over problems with existing evidence, assume too easily that resources are finite and research funding is fungible, and can overestimate the ability of modelling studies to answer policy questions. Like proponents, they also often frame the stakes here in the language of women's lives and bodies, emphasising maternal mortality and other pregnancy-related risks.
We believe rhetorical arguments13, 14 about protecting the lives of women are not the best way to address this issue. Neither are generic arguments about the value and hierarchies of different forms of evidence. Ralph and colleagues' signature contribution is their nuanced discussion of what their research adds and what is possible with current and future evidence. They synthesise and evaluate the existing observational data and identify many ways in which further modelling, epidemiological studies, behavioural and clinical research, and basic biological science could work synergistically to increase our knowledge. Additionally, they address the complicated policy decision points and local contextual factors that always complicate the path from evidence to policy and practice. They describe an approach to evidence, policy, and practice rooted in an "ecology of evidence"15 as the foundation for thinking through the next steps. The current polarised environment around the proposed trial makes this more holistic approach all the more difficult, but necessary.
Hormonal contraceptive use and women's risk of HIV acquisition: a meta-analysis of observational studies
Lancet Infectious Disease
Lauren J Ralph, Sandra I McCoy, Karen Shiu, Nancy S Padian
The evidence from epidemiological research into whether use of hormonal contraception increases women's risk of HIV acquisition is inconsistent. We did a robust meta-analysis of existing data to provide summary estimates by hormonal contraceptive method which can be used to inform contraceptive guidelines, models, and future studies.
We updated a recent systematic review to identify and describe studies that met inclusion criteria. To ensure inclusion of more recent research, we searched PubMed for articles published after December, 2011, using the terms "hormonal contraception", "HIV/acquisition", "injectables", "progestin", and "oral contraceptive pills". We assessed statistical heterogeneity for these studies, and, when appropriate, combined point estimates by hormonal contraception formulation using random-effects models. We assessed publication bias and investigated heterogeneity through subgroup and stratified analyses according to study population and design features.
We identified 26 studies, 12 of which met inclusion criteria. There was evidence of an increase in HIV risk in the ten studies of depot medroxyprogesterone acetate (pooled hazard ratio [HR] 1·40, 95% CI 1·16-1·69). This risk was lower in the eight studies done in women in the general population (pooled HR 1·31, 95% CI 1·10-1·57). There was substantial between-study heterogeneity in secondary analyses of trials (n=7, I2 51·1%, 95% CI 0-79·3). Although individual study estimates suggested an increased risk, substantial heterogeneity between two studies done in women at high risk of HIV infection (I2 54%, 0-88·7) precluded pooling estimates. There was no evidence of an increased HIV risk in ten studies of oral contraceptive pills (pooled HR 1·00, 0·86-1·16) or five studies of norethisterone enanthate (pooled HR 1·10, 0·88-1·37).
Our findings show a moderate increased risk of HIV acquisition for all women using depot medroxyprogesterone acetate, with a smaller increase in risk for women in the general population. Whether the risks of HIV observed in our study would merit complete withdrawal of depot medroxyprogesterone acetate needs to be balanced against the known benefits of a highly effective contraceptive.
Despite more than two decades of scientific investigation, uncertainty remains about whether the use of hormonal contraception increases women's risk of HIV infection.1 Worldwide, 144 million women use hormonal contraception, including 41 million users of injectables and 103 million who take oral contraceptive pills.2 Use of these hormonal contraceptives prevents unintended pregnancies, reduces the rate of maternal and infant morbidity and mortality, and enables women to achieve other life goals.3 With high fertility and maternal mortality rates, particularly in settings of high HIV prevalence, women need to be able to avoid pregnancy without increasing their risk of HIV infection. An increased risk of HIV associated with hormonal contraceptive use might have important contraceptive counselling and other policy implications.
After reviewing available epidemiological evidence, an expert panel convened by WHO in 2012 recommended keeping hormonal contraception as a category 1 method with no restrictions for use. However, the panel also recommended that women using progestin-only injectable contraception, including depot medroxyprogesterone acetate, be "strongly advised to also always use condoms".4 Despite this guidance, some countries in sub-Saharan Africa are considering withdrawing depot medroxyprogesterone acetate from their family planning programmes, even though the results of modelling studies suggest that the effects of such a decision on unintended births and maternal and infant morbidity and mortality rates would be substantial in most settings.5, 6, 7 Thus, the decision to withdraw hormonal contraception will depend not only on whether there is an actual association, but importantly also on whether the increased HIV risk outweighs the tremendous benefits of highly effective contraception.
Because of the public health urgency about whether use of hormonal contraception is associated with increased risk of HIV infection, it is crucial to take full advantage of existing observational evidence. The authors of systematic reviews concluded that existing evidence suggests an increased risk of HIV associated with use of progestin-only injectable contraception, possibly only in women at high risk, but did not quantitatively summarise results because of perceived heterogeneity in study designs and populations.8, 9, 10 However, up to now, heterogeneity has never been quantitatively assessed, and even a moderate amount should not preclude moving forward with meta-analyses of observational data, especially when data from randomised control trials are not available to address an urgent public health issue requiring policy decisions.11, 12 Furthermore, as research in this topic has intensified in recent years, the methodological approaches to addressing this issue have increased in rigour and similarity, making it an opportune time to do a meta-analysis.
Here, we build on the findings of one recent review8 to quantitatively summarise observational evidence, providing a series of pooled estimates of the effect of the use of hormonal contraception on HIV risk by type of method. We focus our analyses on studies of sufficient quality and comparability, and assess the heterogeneity through a series of a-priori secondary analyses.
Search strategy and selection criteria
We did this meta-analysis in accordance with PRISMA guidelines.13 All our statistical analyses were guided by Egger and colleagues.14
We used the 2012 WHO technical review4 to initially identify studies. WHO used an unpublished version of the systematic review that was later published by Polis and Curtis,8 which was subsequently updated and published in October, 2014.10 To ensure inclusion of recent research, we searched PubMed for articles published in English after Dec 1, 2011, using the terms "hormonal contraception", "HIV/acquisition", "injectables", "progestin", and "oral contraceptive pills". Additionally, we identified relevant abstracts presented at the 2011-14 International AIDS Society and Conference on Retroviruses and Opportunistic Infections and followed up with authors to ascertain if their analyses had been published. We also reviewed lists of studies with experts in the specialty.
Two investigators (LJR and KS) reviewed the full text of articles identified to ascertain if the studies met the following inclusion criteria: assessed hormonal contraceptive use as an exposure, including at least one of the following categories: depot medroxyprogesterone acetate, norethisterone enanthate, combined oral contraceptives, or progestin-only pills; used a prospective design and excluded HIV-positive women at baseline, ensuring exposure assessment preceded detection of an incident HIV infection; used an analytic approach that minimised confounding and selection bias by adjustment for, at least, age and condom use, and having minimum loss to follow-up (defined as <30%); published in a peer-reviewed journal by May, 2014; and data gathering took place in a low-income or middle-income country as defined by the World Bank.
Data extraction and coding
Two reviewers (LR, KS, or SM) independently extracted data from studies using a custom, piloted spreadsheet. One investigator (LR) compared extractions to ensure intercoder reliability, and when discrepancies arose a third investigator who was not involved in the original extraction arbitrated.
With the array of hormonal contraceptive methods available, studies often differed in their classification of the types of contraception and many presented several effect estimates. We focused our data extraction on estimates disaggregated by hormone formulation (eg, depot medroxyprogesterone acetate, norethisterone enanthate, combined oral contraceptives, or progestin-only pills). When only the type of method (eg, injectable or pill) was specified, we reviewed the article to identify whether a specific formulation (eg, depot medroxyprogesterone acetate vs norethisterone enanthate) predominated. We coded how comparison groups were constructed, noting whether women using condoms (alone or in addition to hormonal contraception), other types of hormonal contraception, or no contraception were included.
We extracted effect estimates and 95% CIs for each model presented in the paper. We noted the confounders adjusted for in multivariate models and the analytic strategy used (eg, Cox or inverse probability of treatment weighted marginal structural model). In one instance, we also extracted a depot medroxyprogesterone acetate specific estimate and its 95% CI from a letter15 submitted in response to an original manuscript.16
We extracted information about features that might affect internal or external validity (and overall study quality) or explain heterogeneity, including study retention rates, intersurvey intervals, the risk profile of study participants, and the study design. For participants' risk profile, we distinguished women at high risk or key populations (eg, commercial sex workers, injection drug users, or women in serodiscordant partnerships) from women in the general population. Also, we extracted details about the demographic characteristics of participants, recruitment sites, study durations, and exclusion criteria.
Effect estimates and their 95% CIs were log transformed and the SE of each estimate was calculated. Funnel plots were generated to assess publication bias.
We selected one effect estimate per hormonal contraception formulation for each study to include in the primary pooled analyses. When analyses of the same study population were published in different articles and all articles met inclusion criteria, we selected only the most comprehensive or recent paper to include in pooled analyses (for details, see appendix). Although some investigators of the studies did not explicitly describe the oral contraceptive pill as combined or progestin only, use of progestin-only pills is less common in sub-Saharan Africa than in the rest of the world and is typically restricted to post-partum, breastfeeding women. Thus, we assumed that categories of the oral contraceptive pill would be comprised predominantly of users of the combined oral contraception and combined studies that presented estimates for combined oral contraception specifically or oral contraceptive pills generally in our analysis to produce pooled-effect estimates that represent the relation between combined oral contraception and HIV. Four studies17, 18, 19, 20 presented separate estimates for combined oral contraception and progestin-only pills and we used the estimate for combined oral contraception in pooled analyses. When several effect estimates were available, we selected the estimate from the most fully adjusted multivariate model. Although in four studies16, 17, 18, 21 estimates were derived with inverse probability of treatment-weighted marginal structural models, we did not include these estimates in our primary pooled analyses because they were different from traditional regression approaches and the two should not be compared or combined. Specifically, traditional Cox models are used to estimate the average effect of treatment on an individual, whereas marginal structural models provide the average effect of treatment on the population.22 However, we did separate analyses that combined only those estimates generated using inverse probability of treatment-weighted marginal structural model.
Evidence for statistical heterogeneity between studies was assessed for each hormonal contraception formulation (depot medroxyprogesterone acetate, norethisterone enanthate, combined oral contraceptives, or progestin-only pill) using the I2 statistic and its 95% CI; an I2 of greater than 50% indicated sufficient heterogeneity to contraindicate a pooled estimate.23 When the I2 was less than 50%, pooled-effect estimates were calculated with DerSimonian and Laird random-effects models.24
We assessed the robustness of findings and investigated heterogeneity through a series of a-priori secondary analyses. First, we did an influence analysis to identify whether any study alone disproportionately affected the results. Second, we stratified meta-analyses according to the risk profile of the study population (high risk vs general population), and the original study design (prospective cohort vs randomised trial). Third, because of concerns that having a reference group that is composed largely of condom users might artificially inflate the risk of HIV acquisition for users of hormonal contraception,25 we investigated whether our results were sensitive to the exclusion of condom users from the comparison group. Last, we investigated whether the results were qualitatively different when studies with intersurvey intervals longer than the duration of the contraceptive methods under study (1-3 months) were excluded. All analyses were done in Stata (version 12.0).
We reported effect estimates as hazard ratios (HRs) because these were reported in all but one26 of the studies in our pooled analyses. The investigators of that study reported incidence rate ratios, which are comparable to the HR.27, 28
Role of the funding source
There was no funder for this analysis. The corresponding author had full access to all the data in the study (previous publications) and had final responsibility for the decision to submit for publication.
We identified 26 articles,16, 17, 18, 19, 20, 21, 26, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47 12 of which met our inclusion criteria (figure 1; appendix).16, 17, 18, 19, 20, 21, 26, 39, 40, 44, 45, 47 Two articles were analyses of the same population; however, because different analytic approaches were used (Cox regression20 vs inverse probability of treatment weighted marginal structural model21), both were included but in separate pooled analyses to prevent double counting.
All studies in the final sample were done in sub-Saharan Africa. Three, all prospective cohort studies, were designed specifically to assess the hormonal contraception-HIV relation.20, 39, 40 Seven studies were secondary analyses on cohorts enrolled in randomised trials of various HIV16, 17, 18, 44, 45, 47 and one cervical cancer19 prevention interventions. Two study populations consisted of women at high risk of HIV, commercial sex workers39 or women in serodiscordant partnerships.16 The remainder were composed of women in the general population, typically recruited at family planning or other health centres. The median age of participants ranged from 25 years to 40 years. With the exception of two studies in which women were surveyed every 6 months19 or 10 months,26 women in the other studies were surveyed at least every 3 months. With the exception of one study in which a subset of women were followed up for 6 months,19 women were followed up for at least 1 year in the other studies. The median follow-up ranged from 12 months to 31·2 months. Because of the heterogeneity in the presentation of estimates of loss to follow-up, we did not quantitatively summarise this metric. However, generally, study retention was high, with a minimum of six of 12 studies having retention rates of more than 85% (Table 1, Table 2).
Funnel plots to assess injectable contraception and oral contraceptive pills were symmetrical, suggesting no publication bias for studies (appendix).
Ten reports were investigations of the association between depot medroxyprogesterone acetate and HIV. In pooled analyses, use of depot medroxyprogesterone acetate was associated with an increased risk of HIV acquisition compared with use of non-hormonal or no methods (pooled hazard ratio [HR] 1·40, 95% CI 1·16-1·69; figure 2). An influence analysis showed that no single study was altering the results (data not shown). The pooled-effect estimate for the three studies in which the inverse probability of treatment weighted marginal structural model was used (pooled HR 1·41, 1·15-1·72; table 3) was similar to the overall estimate.
In subgroup analyses, the pooled HR for the three prospective cohort studies was 1·44 (95% CI 1·04-2·01; table 3). A high level of between-study heterogeneity (I2 51·1%, 95% CI 0-79·3) in the seven secondary analyses of cohorts from randomised controlled trials precluded the calculation of a pooled estimate for this subgroup (table 3).
The eight studies of women in the general population had a lower heterogeneity (I2 27·3%, 95% CI 0-67·3) than did the primary analysis (42·5%, 0-72·5; table 3). The pooled estimate suggested a moderate increase in the risk of HIV acquisition (pooled HR 1·31, 95% CI 1·10-1·57; table 3). Estimates were higher in each of the two studies with women at high risk (1·73, 1·28-2·34, in commercial sex workers39 and 3·93, 1·37-11·2, in women in serodiscordant partnerships).16 However, the high heterogeneity (I2 54·0%, 95% CI 0-88·7%; table 3) between these two studies contraindicated the pooled estimates.
In an analysis of the nine studies in which the reference group comprised women whose partners were using condoms (in addition to other methods or no method), the pooled effect estimate did not change much from the primary analysis (pooled HR 1·44, 95% CI 1·20-1·73; table 3). An analysis restricted to the eight studies in which the intersurvey interval did not exceed 3 months showed a pooled-effect estimate that was slightly larger than our primary analysis (1·48, 1·24-1·77; table 3).
The estimates of the relation between combined oral contraceptives or oral contraceptive pills with HIV were reported in ten studies. Risk of HIV acquisition was not increased in users of combined oral contraception or oral contraceptive pills compared with those using non-hormonal or no methods (pooled HR 1·00, 95% CI 0·86-1·16; table 4) and the results of our influence analysis showed that no study was affecting these results (data not shown). There was minimum evidence of between-study heterogeneity (I2 0%, 95% CI 0-48·6; table 4). The pooled estimate for the five studies in which inverse probability of treatment weighted marginal structural model was used was similar to the primary pooled result (pooled HR 1·03, 95% CI 0·81-1·32; table 4). A subgroup analysis of the two studies in high risk women showed an increased risk of HIV acquisition in users of combined oral contraceptives or oral contraceptive pills (pooled HR 1·49, 95% CI 1·04-2·13; table 4). There was only one study that assessed progestin-only pills as an exposure;18 thus, we were not able to pool effect estimates for this specific exposure.
Estimates of the relation between norethisterone enanthate and HIV did not show an increased risk of HIV acquisition (pooled HR 1·10, 95% CI 0·88-1·37) and showed minimum heterogeneity (I2 0%, 95% CI 0-74·6; table 4). Similar estimates were obtained for the two studies with inverse probability of treatment-weighted marginal structural model (pooled HR 1·08, 95% CI 0·77-1·52; table 4). The results of the influence analysis were not significant (data not shown) and subgroup analyses were not possible because of the small number of studies.
The results of our meta-analysis showed that in observational studies with similarly and precisely defined exposures, adjustment for key confounders, minimum selection bias, and sound analytical approaches, there was a small but increased risk of HIV acquisition associated with use of depot medroxyprogesterone acetate. Consistent with the results of an earlier meta-analysis of oral contraceptive pills,48 risk was not increased for users of oral contraceptive pills or combined oral contraceptives in the general population. Furthermore, there was no increased risk in users of norethisterone enanthate; however, the few studies in this analysis preclude any definitive statements about an association of norethisterone enanthate with HIV.
The results from this analysis, particularly for depot medroxyprogesterone acetate, should be used as an input parameter in ongoing modelling studies to quantify the compromises associated with removing injectables from the contraceptive method mix. For example, Butler and colleagues6 used both a hypothetical (relative risk [RR] 1·2) and a single study (odds ratio 2·19)16 estimate to predict changes in the numbers of HIV and maternal deaths after reductions in use of injectable hormonal contraception. Their findings suggest that, except in southern Africa where both HIV incidence and use of injectable contraception are high, the effect of removing hormonal contraception on the number of maternal and HIV-related deaths is sensitive to the chosen effect estimate. In view of these results, it is possible that an increased risk of magnitude found in our study (HR 1·4; figure 2), particularly for women in the general population, would not merit complete withdrawal of depot medroxyprogesterone acetate because maternal mortality would still exceed HIV-related deaths in most settings, particularly if women did not have immediate access to and uptake of alternate, effective contraceptive options in the absence of depot medroxyprogesterone acetate, one of the assumptions in Butler and colleagues' models.6 Moving forward, we encourage Butler6 and others5, 7 to apply our estimates and more fully explore regional or geographic and subpopulation differences so that context-specific contraceptive policy can be developed.
Our analysis also provides an insight into potential sources of heterogeneity in results. Studies in women in the general population, representing most of the studies in our analysis, provide estimates of the average population effect of hormonal contraception on women's risk of HIV acquisition. By contrast, studies in women at high risk, of which there were two in our analysis, provide estimates of the effect of hormonal contraception conditioned on a high likelihood of HIV exposure. For the millions of users of hormonal contraception worldwide, most of who are not in serodiscordant or other high-risk partnerships, this distinction is crucial. Although the increased risks for users of depot medroxyprogesterone acetate and users of combined oral contraception or oral contraceptive pills reported in the two studies with commercial sex workers39 and women in serodiscordant partnerships16 might warrant consideration of changing contraceptive guidelines for these populations, it would be too early to do so based on the results of just two studies. Furthermore, it is important that the results of these two studies are not generalised to women in the general population, who according to the results of our study had a smaller increase in risk that might only warrant a policy change in specific local contexts.
A priori, we established a strict set of inclusion criteria for our meta-analysis. Although this left us with fewer studies and reduced power in our planned secondary analyses or analysis of heterogeneity through meta-regression, only comparable estimates were combined. Contrary to the perception that these studies are too diverse for meta-analysis, we did not find heterogeneity that would preclude pooling estimates in most analyses. One notable exception is that although the studies done as secondary analyses of randomised trials contribute to the primary pooled analyses, we were unable to present a separate pooled estimate specific to this subset of studies for depot medroxyprogesterone acetate. The heterogeneity statistic for this group (I2 51·1%; table 3) is borderline between moderate and substantial according to current Cochrane guidance.49 Compared with the prospective cohort studies, each of which was designed specifically to examine the relation between hormonal contraceptive use and women's risk of HIV, the randomised trials had divergent research objectives. This might partly explain the higher level of heterogeneity noted in this subgroup. Hence, a very conservative application of our findings would be to use the pooled HR and 95% CI from only the prospective cohort studies. However, the randomised trials have many strengths, notably their large sample sizes, assessment of contraceptive method use and switching, and efforts to ensure high retention. In the absence of another prospective cohort study or data from the proposed randomised trial of the association of hormonal contraception and HIV,50 the results of which will not be available for several years, other HIV prevention trials are the primary source of data to study this association in the near future.51
Our study findings have several limitations. First, meta-analyses of observational studies, like observational studies themselves, are inherently prone to bias and cannot be used to address whether the association between hormonal contraception and HIV is causal.14 There has been extensive discussion about whether studies so far have sufficiently addressed the potential confounding effects of misreported condom use,25, 52 particularly because many study populations were drawn from HIV prevention trials in which condom use is strongly encouraged and women might feel pressure to report socially desirable behaviours.53, 54 However, the results of recent modelling studies suggest that the practical effects of misreporting condom use might be overstated. For example, Smith and colleagues55 show that only a substantial amount of condom use under-reporting by non-hormonal contraceptive users, an unlikely scenario, could explain the increased effect estimate in the study by Heffron and colleagues16 (HR 2·19 for all injectable contraception and 3·93 for depot medroxyprogesterone acetate specifically). Furthermore, our own work with biomarkers of unprotected sex has shown that misreporting of condom use is not significantly different between women using hormonal contraception and those using other methods, and therefore might not bias effect estimates to the extent suggested.56 Even a randomised controlled trial is not likely to overcome many of the measurement challenges inherent to studying whether an association exists between HIV and use of hormonal contraception.5, 57 Likewise, the limitations of the original studies are also limitations of our analysis. For example, none of the studies prospectively assessed acute HIV infection, which would strengthen our confidence in the timing of exposure to hormonal contraception and subsequent acquisition of HIV in women.
A second limitation is that, despite our efforts to ensure systematic inclusion of all studies of the association between hormonal contraception and HIV and publication bias with funnel plots, like all meta-analyses, our results might be biased if only studies with significant results have been published. However, here, publication bias is less likely because over the past two decades, a null finding was equally compelling in terms of advancing the debate. However, if studies with positive and significant effects of hormonal contraception on women's risk of HIV acquisition were more likely to be published, that would imply that our findings are an overestimate of the true association.
Although our study findings are similar to what was previously presented qualitatively in two systematic reviews8, 9, 10 (ie, there is evidence of a moderate increase in the risk of HIV infection for users of injectable hormonal contraception, specifically in women at high risk), this study is the first in which existing evidence is quantitatively summarised, particularly for depot medroxyprogesterone acetate, and a series of weighted, pooled estimates of effect and their variances are reported by type of hormonal contraceptive method used, for all studies published until May 31, 2014. Since we approached data extraction and definitions of study quality independently from the other reviews, our study provides another perspective of methodological rigour of the existing evidence.
In view of the concerns about the observational evidence gathered so far, efforts are underway to fund a randomised trial to investigate the relation between hormonal contraception and HIV. Arguably, the increase in risk noted for users of depot medroxyprogesterone acetate in our study, who would comprise one of the intervention groups, might violate the principle of equipoise required for a trial.58 Another important concern is whether, with the methodological challenges inherent in studying the association between hormonal contraception and HIV,57 the randomised trial will provide evidence superior to current evidence, especially when also considering the personal and financial investments required for a trial.1 Our pooled estimates can inform contraceptive policy now, without waiting several years for trial data. Additionally, our findings emphasise an immediate need to refocus secondary analyses on commercial sex workers and women in serodiscordant partnerships because evidence for these women at high risk is inadequate but suggests an increased risk of HIV infection. Meanwhile, basic science research must continue so that the biological mechanisms underlying the association found here can be more precisely defined.59 Imperative for public health is the continued need to promote a wide array of existing methods and develop and promote long-term options for reversible contraceptives for women worldwide.