Bulletin of the World Health Organization

Population-level impact of hormonal contraception on incidence of HIV infection and pregnancy in women in Durban, South Africa

Gita Ramjee a & Handan Wand b

a. HIV Prevention Research Unit, Medical Research Council, 123 Jan Hofmeyr Road, Westville, Durban, 3630, South Africa.
b. The Kirby Institute, University of New South Wales, Sydney, Australia.

Correspondence to Gita Ramjee (e-mail: Gita.Ramjee@mrc.ac.za).

(Submitted: 30 March 2012 – Revised version received: 28 May 2012 – Accepted: 05 July 2012 – Published online: 16 July 2012.)

Bulletin of the World Health Organization 2012;90:748-755. doi: 10.2471/BLT.12.105700


In sub-Saharan Africa, an estimated 22.5 million people live with human immunodeficiency virus type 1 (HIV-1) infection.1 In the province of KwaZulu-Natal, the prevalence of HIV-1 infection is estimated to be 39% among women who attend antenatal clinics.2 Poor socioeconomic conditions, sexual networking, multiple concurrent partnerships, low level of condom use and high rates of sexually-transmitted infections are factors that contribute to the high local prevalence of HIV-1 infection.2,3 Trials for the prevention of HIV-1 infection have been heavily focused on developing biomedical interventions under female control.46 Typically these trials recruit sexually active, non-pregnant women of childbearing age and, since the teratogenic effects of the products they test are seldom known, the women are required to use an effective birth control method. Furthermore, if a trial participant falls pregnant, she is usually taken off the product under study, a practice that reduces the trial’s statistical power and its ability to demonstrate the product’s efficacy.

Hormonal contraception has been reported to increase susceptibility to HIV-1 infection by promoting cervical ectopy, increasing vulnerability to sexually-transmitted infections and altering the flora of the genital tract and the structure of the vaginal epithelium.7 In South Africa, hormonal contraceptives, particularly injectables, are reported to be the most common contraceptive method because they are highly effective and safe.810 Contraceptives play a key role in efforts to prevent HIV-1 infection by reducing the number of unintended pregnancies and, hence, the number of potential cases of mother-to-child transmission of HIV-1.11

Several epidemiological studies have provided evidence, although inconsistent, that the use of hormonal contraceptives (injectables and pills) increases the risk of HIV-1 infection.1217 These findings were supported by a subsequent study by Heffron et al. that provided new evidence suggesting an increased risk of contracting and transmitting HIV-1 infection associated with the use of hormonal contraceptives, particularly injectables.18 Durban data from the MDP301 trial confirmed this.19 Morrison et al. combined the multisite data from the Carraguard® trial, including the Durban site data, and showed a moderately increased risk of HIV-1 seroconversion with the use of the injectable progestin depo-medroxyprogesterone acetate, but no effect with other types of hormonal contraceptives.20 Based on available data, in 2012 the World Health Organization released a technical statement clarifying that the data on hormonal contraceptives was inconclusive and that women should continue to use hormonal contraceptives to prevent unwanted pregnancy. Nonetheless, it advised women using progesterone-only contraception to use condoms and other methods for preventing HIV-1 infection.21

The data on hormonal contraceptives and an increased risk of HIV-1 infection makes it critically important to understand the dynamics of HIV-1 transmission, the relative risk and population attributable risk (PAR) associated with the use of hormonal contraceptives. Standard epidemiological methods that quantify risk relative to exposure through metrics such as relative risks are suitable for establishing a causal link between exposure to a given risk factor and a particular disease outcome. However, they provide no information about the population-level impact of removing the risk factor – hormonal contraceptives in our case – from the target population. Doing so yields the PAR, which quantifies the theoretical reduction in disease rates if the target population were not exposed to the risk factor at all.

After three decades of research, we have arrived at a good understanding of the behavioural and biological drivers of the HIV-1 epidemic.22,23 However, we still have a poor understanding of the impact of the use of family planning methods, particularly hormonal contraceptives, on the incidence of HIV-1 infection. Studies are urgently needed to determine the association between hormonal contraceptive use and the risk of HIV-1 infection and, most importantly, the population-level impact of hormonal contraceptives, not just on HIV-1 seroconversion but also on pregnancy rates.

The primary aim of this study was to examine the population-level impact of hormonal contraceptives (injectables and pills) with respect to HIV-1 seroconversion and the incidence of pregnancy during follow-up in two combined cohorts of HIV-1-negative, non-pregnant women who participated in two biomedical trials conducted in Durban, South Africa, in 2004–2009 to test the effectiveness of two vaginal microbicides in preventing HIV-1 infection: the MDP 301 trial, which tested PRO 2000, and the Carraguard® trial, which tested the potential microbicide, Carraguard. ®5,6


Participants and design

A total of 3704 study participants were included in our analysis: 1456 from the Carraguard® trial and 2248 from the MDP301 trial.5,6 The two Durban cohorts were broadly similar in terms of sociodemographic and behavioural characteristics and prevalence of HIV-1 infection and pregnancy rates, except that the proportion of women who reported use of a condom by their partners in their most recent sexual act was significantly higher in the MDP301 trial.

The MDP301 trial followed women every 4 weeks for 52 weeks, and women’s HIV-1 status was assessed at weeks 12, 24, 40 and 52. Serostatus was confirmed by means of parallel HIV-1 rapid tests, and discordant/positive test results were confirmed by enzyme-linked immunosorbent assay.5 The Carraguard® trial had visits at screening, enrolment, at months 1 and 3, and every 3 months thereafter. Rapid HIV-1 blood testing was conducted at all visits except at enrolment. Serostatus was confirmed with parallel HIV-1 rapid tests, and positive/discordant tests were confirmed by third-generation enzyme immunoassay or polymerase chain reaction (PCR) for the detection of HIV-1 ribonucleic acid (RNA).6

In Durban, the MDP301 and the Carraguard® trial were both approved by the University of KwaZulu-Natal Biomedical Research Ethics Committee and the South African Medical Research Council.

Risk factors

We assessed commonly-reported risk factors for HIV-1 seroconversion, such as being less than 25 years old, being unmarried or not living with a regular sex partner, having multiple sex partners, having frequent sex, being unemployed and having a sexually-transmitted infection.24,25 To this end we used the following data, as collected across the two trials: age, marital status, patterns of contraceptive use, diagnosis of a sexually-transmitted infection, condom use, educational level, employment status, number of sex partners and average number of sex acts per week. Data that were not common to both trials, such as being abused by partner or forced to have sex (which were collected in the Carraguard® trial but not in the MDP301 trial), were not included. Women were categorized according to their self-reported type of contraceptive use: hormonal methods (injectables or pills) and other methods (male condoms, hysterectomy, female sterilization and traditional methods such as abstinence and rhythm).

Age at baseline was based on self-reported date of birth and verified by examining each woman’s local identification card. Women were categorized by self-reported level of education (lower than high school versus high school or higher), employment status (employed with a regular income versus unemployed and having an irregular income or none) and religion (Christian versus others). Cohabitational status (married or living with a sexual partner versus not married or not living with a sexual partner), multiple sexual partners (two or more) in the three most recent months, average weekly number of sexual acts (three or more versus less than three) and diagnosis of a sexually-transmitted infection (at baseline) were used as binary variables. Since more than 90% of the women reported having a regular partner, having a regular sex partner was not among the risk factors considered in the analyses.

In the current study, hormonal contraception was treated as a theoretically modifiable risk factor, whereas age was assumed to be non-modifiable. The remaining variables (cohabitational status, number of sexual partners, average weekly number of sexual acts and presence or absence of a sexually-transmitted infection) were assumed to be background risk factors (or potential confounders) and hence non-modifiable and unchanged. All decisions regarding risk factor classification were made before conducting the analyses.

Statistical analyses

Demographic characteristics and sexual behaviour, including cohabitational status, number of sexual partners, average weekly number of sexual acts during the 2 weeks immediately preceding the study visit and diagnosis of a sexually-transmitted infection, were compared across contraceptive groups using χ2 analyses. Kaplan–Meier survival analyses were carried out to estimate time to infection among women who became infected with HIV-1 and time to pregnancy among women who became pregnant during trial follow-up, with stratification by contraceptive groups (i.e. hormonal injectables or pills versus other contraceptives). Calculation of the time to seroconversion in each of the two trials is described in detail elsewhere.5,6 Briefly, the seroconversion date was imputed as the midpoint between the date at which the participant first tested positive for HIV-1 and the date of the previous negative test. Time to seroconversion was then calculated as the difference between the seroconversion date and the date of enrolment in the trial.5,6

In unadjusted analyses, univariate Cox proportional hazard regression models were used to determine the association between baseline and follow-up use of contraceptives and risk of HIV-1 infection and pregnancy. We calculated hazard ratios (HRs) and 95% confidence intervals (CIs) using Cox regression models. Results with a P < 0.05 were considered significant. We also adjusted for age, number of sexual partners, condom use by partner during the most recent sexual act (at trial baseline and follow-up) and pregnancy status after baseline. To accommodate missing data, we applied the last-observation-carried-forward convention, a method used to impute unknown values from existing data, to all time-varying covariates (e.g. condom use), except for pregnancy status.

Population attributable risk

We used a novel method to estimate the PARs,26,27 which reflect the proportion of HIV-1 seroconversions that would not have occurred had all women been using a low-risk family planning method (i.e. any non-hormonal method), under the assumption that the observed associations were causal. When calculating PARs, we used Cox regression models to analyse how much individual risk factors and their various combinations affected the rate of HIV-1 seroconversion. The prevalence of different combinations of variables for all risk factors were estimated as multinomial probabilities using the empirical fraction of person–time of follow-up in the cohort. This approach to calculating PARs is intended to determine the joint impact of several theoretically modifiable risk factors on HIV-1 transmission while keeping non-modifiable and/or background risk factors unchanged. The result is known as the partial PAR, which is based on the estimated relative risk associated with each risk factor of interest and the estimated population prevalence of each factor. Analyses were performed using SAS statistical software version 9.2 (SAS Inc., Cary, United States of America).


At enrolment in the study, approximately 78% of the women reported using at least one type of contraceptive (Table 1). Hormonal contraceptives (injectables and pills) were the most commonly reported method, followed by condoms and other types of contraceptives (36.28%, 26.53% and 14.87%, respectively). Hormonal contraceptives were most often used by women that were younger and unmarried. Reported use of a condom during the most recent sexual act was highest (89.11%) among women who indicated that condoms were their preferred contraceptive method.

HIV-1 seroconversion

Table 2 gives crude incidence rates and unadjusted and adjusted HRs for HIV-1 seroconversion in association with the use of hormonal contraceptives at enrolment and during trial follow-up. A total of 272 HIV-1 seroconversions were observed in the cohort as a whole. Compared with women who reported using other contraceptives, those who were using hormonal contraceptives at study enrolment had a significantly higher risk of HIV-1 seroconversion. This association was slightly attenuated when the analysis was adjusted for age, consistent condom use in the most recent sexual act, average number of sexual acts per week during the two weeks immediately preceding the study visit, and incidence of pregnancy. When pattern of contraceptive use was included as a time-dependent covariate (i.e. updated at each visit), use of hormonal contraceptives was significantly associated with an increased risk of HIV-1 infection in both unadjusted and adjusted analyses.

Pregnancy incidence

The incidence of pregnancy was 3.7 per 100 woman–years and 10.1 per 100 woman–years for women who reported using hormonal contraceptives and other methods, respectively (Table 2). The use of hormonal contraceptives at enrolment and during the study was associated with a significantly decreased risk of pregnancy. These strong protective associations were sustained but slightly attenuated in adjusted analyses.

Partial population attributable risk

The partial PARs and their 95% CIs are presented in Table 3. In analyses without adjustments (crude PAR), the use of hormonal contraceptive pills or injectables at baseline and during study follow-up accounted for approximately 20% of the HIV-1 seroconversions. When adjustments were made for other risk factors, such as multiple sex partners, no condom use by partner during the most recent sexual act and younger age (< 25 years), the partial PAR was 12%. Although the use of hormonal contraceptives at enrolment was reasonably high (36.28%), relatively small effect sizes were responsible for this small percentage. Meanwhile, if women who reported using non-hormonal contraceptive methods (at baseline and during study follow-up) had used hormonal contraceptives instead, 72% (adjusted) of all pregnancies could have been avoided. High rates of use of other types of contraceptives (64%) having a strong protective effect accounted for this large impact on pregnancy incidence. Only a small percentage of women reported having changed the type of contraceptive during the study (data not shown).


The use of hormonal contraceptives, which are highly effective, is known to directly benefit women by reducing the physical, emotional and social consequences of unintended pregnancies and by indirectly reducing cases of mother-to-child transmission of HIV-1. This was shown by a study in sub-Saharan Africa in which the use of hormonal contraceptives reduced the number of neonates becoming positive for HIV-1 by an annual 22%.11 Our data concurs with those from previous studies in which the use of hormonal contraceptives was found to be associated with an increased risk of HIV-1 infection.1215 The relatively small effect size in our study was statistically significant in both the adjusted and unadjusted analyses because of the study’s high statistical power (> 90%). However, the population-level impact of the use of hormonal contraceptives on HIV-1 infection rates was determined to be only 12% in adjusted analyses. Alternatively, using non-hormonal contraceptives or no contraception at all accounted for 72% of the incident pregnancies during the trials. In short, removing all use of hormonal contraceptives from the target population would result in 12% fewer HIV-1 seroconversions but 72% more pregnancies. These findings provide an idea of the risk-benefit and cost-effectiveness ratios – both beyond the scope of this study – associated with the use of hormonal contraceptives among women targeted by HIV-1 prevention efforts. These assumptions can be more accurately assessed using mathematical modelling. Our study established that hormonal contraceptives were most commonly used by young, unmarried women, and this may have contributed to their increased risk of HIV-1 seroconversion in our setting.28

Heffron et al. reported a significant association between use of hormonal contraceptives and HIV-1 seroconversion with a reasonably large effect size (adjusted HR: 1.98; 95% CI: 1.06–3.68).18 The study reported that almost 15% of women were using hormonal contraceptives at enrolment and 21% at follow-up. Given this data, determining the population-level impact is critical. Based on our results, the population level impact would be estimated to be in a range of 10–20%, suggesting that if the exposure to hormonal contraceptives was removed from the study population, 10––20.0% of the HIV-1 seroconversions could have been prevented. Heffron et al. also reported pregnancy rates of 5.2% and 16.0% among women who did and who did not report using hormonal contraceptives, respectively.18 Since the adjusted hazard ratio was not presented, the percentage of pregnancies attributed to the use of other types of contraceptives in adjusted analyses could not be estimated. However, a large absolute difference between rates of HIV-1 seroconversion and pregnancy rates may indicate that hormonal injectable contraceptives have a large population-level impact in terms of preventing pregnancies in the target population.

Our study was not designed to assess the association between contraceptive use and the risk of HIV-1 infection. Therefore, the results presented here may not be generalizable. Additionally, since the questions on behavioural and sociodemographic characteristics differed between trials, we were unable to include in the analysis some key risk factors (such as young age at sexual debut). Similarly, since a common protocol was not used across both trials, we cannot be sure that all categorizations were the same. Also, contraceptive methods were self-reported and therefore subject to misclassification due to recall bias. In the cohorts described, some women were given contraceptives at the research sites, whereas others received them from their local family planning clinics. However, the significantly lower pregnancy rate among women who reported using hormonal contraceptives provided strong biological evidence of good adherence to the method. Our results cannot be generalized because of the selected population targeted in the HIV-1 prevention trials. Importantly, our findings, which are based on behavioural data, do not constitute biological evidence that hormonal contraception per se facilitates disease transmission. What they suggest, rather, is that the partners of women on hormonal contraceptives are less likely to use condoms than the partners of women who do not use contraceptives of this type (the assumption being that women with a steady partner have sex with no one else).

The impact of using hormonal contraceptives, particularly injectables, on HIV-1 transmission is a public health question that remains unanswered, yet understanding such impact is vital.1214,2931 Conflicting study outcomes have triggered doubts regarding the role of injectable contraceptives and their use in developing countries.1218 It is critical to determine to what degree hormonal contraceptives increase the risk of HIV-1 infection. Although studies designed to address this question may be complex and difficult to implement, such difficulty must be weighed against their potential impact on HIV-1 acquisition.

In our study setting, condom use among the partners of women on hormonal contraceptives was low, a finding that underscores the importance of establishing comprehensive programmes that integrate both family planning and HIV-1 prevention for women and their partners. Our data points to the urgent need to develop methods for simultaneously protecting women against HIV-1 infection and pregnancy. Furthermore, there is an urgent need to expand women’s contraceptive choices in areas of high HIV-1 endemicity, by providing access to safe and effective long-acting, non-hormonal contraceptives, such as intrauterine devices and low-dose hormonal implants. To our knowledge, the current study is the first to have explored the population-level impact of the use of hormonal contraceptives in a large combined cohort of women in Durban. Our results confirm an urgent need to adopt innovative health-care strategies and to educate women and health-care workers on the importance of dual counselling on pregnancy and the prevention of HIV-1 infection.

Thirty years of risk factor research has provided us with a good understanding of how HIV-1 is acquired. The current challenge is to interpret risk factor analyses in terms of prioritizing prevention strategies. The findings of our study and of previous studies suggest that the use of hormonal contraceptives is a speculative risk factor for HIV-1 seroconversion. However, its population-level impact in terms of the risk of HIV-1 infection is lower than its population-level benefit in preventing unintended pregnancies and, indirectly, HIV-1 infection among neonates. Integration of family planning and HIV-1 counselling and testing programmes may be the key to promoting a multi-pronged approach to a healthy, HIV-1-free reproductive life for women.


We thank the participants of the Carraguard® and MDP301 trials. Our thanks also go to the sponsors of these trials: The United Kingdom Department for International Development and the Medical Research Council (grant number G0100137) for MDP301 and the US Agency for International Development (HRN-A-00–99–00010 and GPO-A-00–04–00019) and the Bill & Melinda Gates Foundation (grant number 18057) for the Carraguard® study. We also thank Brodie Daniels and Nathlee Abbai for their valuable assistance.

Competing interests:

None declared.