Worldwide, a substantial unmet need for safe, effective contraception remains, particularly in areas with a high HIV prevalence [1]. For HIV-infected women, contraception is important for their own health, for preventing mother-to-child HIV transmission, and to control their fertility. For women using antiretroviral therapy (ART), concerns have been raised that some progestin-based contraceptive methods may be less effective when used with certain ART agents. Specifically, it has been suggested that the non-nucleoside reverse-transcriptase inhibitors (NNRTIs), nevirapine (NVP) and efavirenz (EFV), may accelerate the progestin metabolism through the cytochrome p450 pathway [2]. Studies of interactions between ART and contraceptive methods have been limited by short follow-up times, HIV-uninfected populations, and/or surrogate endpoints [3]. We sought to understand the effect of ART on contraceptive effectiveness, measured by the clinical endpoint of pregnancy, using prospectively collected data from over 5,000 women with chronic HIV infection from East and Southern Africa.

Among the 5,153 women included in this analysis, 54% were under 30 years old at baseline and 51% had CD4 counts ≥500 cells/mm³ at enrollment (Table 1). A total of 9,266 person-years were accrued and median follow-up time was 1.8 years (IQR 1.2-2.3). During follow-up, 24% of women became pregnant and 31% ever used ART, including 23% who used NVP and 5% who used EFV. Also during follow-up, 9% of women ever used implants, 40% ever used injectable contraception, and 14% ever used OCs; for women who used implants at baseline, 82% of subsequent visits had continued use, while for injectables continuation was 70% and OCs 58%.

The overall pregnancy rate was 14.8 per 100 person-years. Women not using contraception but using ART had a lower pregnancy rate compared to those not using contraception and not using ART (13.2 vs 22.5 per 100 person-years; aHR 0.62, 95% CI 0.51-0.75, p<0.001.)In a sensitivity analysis that excluded the first six months after each pregnancy, this association between ART use and pregnancy was attenuated (aHR 0.80, 95% CI 0.63-0.99, p=0.03), potentially indicating that this difference was due to a post-partum effect for women who had initiated ART during pregnancy.

Implants were highly effective in reducing incident pregnancy, with rates less than 1.5 per 100 person-years among those both on and not on ART (Table 2); pregnancy incidences among implant users were >90% lower when compared to women not using contraception. There was no statistical difference between the effects of implants on pregnancy prevention for women using versus not using ART (interaction p=0.73). Both injectables and OCs also significantly reduced pregnancy risk compared to women not using contraception, although to a lesser extent than implants. For women using injectables, the reduction in pregnancy incidence was approximately 80%, and for OCs the relative reduction was approximately 60%; neither of these effects differed significantly for women on ART versus not on ART.

Nevirapine was the most common ART agent reported, with 1000 person-years of follow-up. The results among women using NVP were very similar to the overall analysis. Injectables reduced pregnancy by approximately 80% and OCs reduced pregnancy by approximately 60%; for both methods, the pregnancy prevention effects were not different when using versus not using NVP (interaction term p=0.80 for injectables and p=0.95 for OCs). There were no pregnancies among women using NVP and implants.

Approximately 200 person-years accrued from women using EFV. The pregnancy prevention point estimates for women using EFV were lower for all types of progestin-based contraception: implants reduced pregnancy incidence by approximately 60%, injectables by 70%, and OCs by only 15%. None of these reductions in incidence were statistically significant, in part likely because of the small sample size; there were also no statistically significant interactions for EFV use with any contraceptive method compared to women not using ART. For women using OCs with EFV, the point estimate for the pregnancy rate was double that of those using NVP. For women using implant, a single pregnancy occurred during 16.7 person-years of concurrent use with EFV, at eleven months post-insertion of the implant.

In this analysis of a large, prospective database from African women living with HIV, we found that hormonal contraceptives were effective in reducing pregnancy, including for women using ART. Implants were the most effective at reducing pregnancy incidence, by >90%, and we found no evidence that ART use in general diminished implant effectiveness. Injectable and oral methods were also effective, and their effectiveness did not differ significantly by ART use. All hormonal methods had point estimates suggesting lesser effectiveness among EFV users, although the sample size for analyses limited to EFV users was small.

Our analysis is consistent with the existing literature. Most pharmacokinetic and observational studies of NVP have found no difference in progesterone levels nor contraceptive effectiveness with progestin-based contraception [8–15]. The literature regarding EFV is more mixed. Studies have found no effect of EFV on progesterone levels when using DMPA [8,10,16], but significantly lower progesterone levels among implant users [17–19,15], along with contraceptive failure rates up to 15% among levonorgestrel implant users [20–22]. Evidence regarding EFV and OCs is very limited, suggesting lower progesterone levels but not necessarily increased ovulation on EFV [13,23,24]. The existing evidence has led at least one national health department in Africa to recommend against implant use for women on EFV [25]. However, a recent retrospective analysis found that implants were more effective at pregnancy prevention than injectables for women using NVP or EFV [26]. Likewise, in the present analysis contraceptive failures were seen on all methods, although they were rare for women using implants.

Among women not using contraception, we found ART use was associated with lower pregnancy incidence; other studies have found that ART use was associated with either an increase or no difference in pregnancy [27,28]. However this difference would not affect the significance of the interaction terms presented, since the primary comparison was pregnancy incidence for women using versus not using contraception, separately for those on and not on ART. In a sensitivity analysis that excluded six-months after each pregnancy, the association between ART use and pregnancy was attenuated, suggesting potential confounding with PMTCT and reduced fertility in the post-partum period. Furthermore, the estimated hazard ratios and significance of the interaction terms were not changed. In addition, there may have been residual confounding due to behavioral and health differences between women who initiate ART versus those not yet starting ART.

Limitations of this study include that ascertainment of ART use, contraceptive use, and important potential confounders, such as sexual behavior, were based on self-report. We are unable to further distinguish specific progestin-based contraceptive methods, although DMPA was the predominant form of injectable, levonorgestrel was the most common implant, and combined pills were the most common OCs available at the study sites. Duration of ART and hormonal contraception use were not examined and may be of interest. This study benefits from a large sample size from 2004 to 2013, with over 9000 person-years of follow-up; however time on EFV was limited. Detailed information was collected to date the beginning of each pregnancy.

This longitudinal study of incident pregnancies suggests that hormonal contraceptive methods remain effective for HIV-infected women who are using ART, including NNRTIs. Implants, which have the lowest adherence requirements out of these three methods, showed the greatest reduction on pregnancy rates, including for women concurrently using ART. Pregnancy incidence was highest for women using OCs, followed by injectables, with failure rates considerably higher than in ideal use settings, possibly relating to inconsistent or incorrect use. As national policies evaluate the potential pharmacokinetic interactions between ART and hormonal contraception, prospective studies such as this, which comparatively evaluate the real-world effectiveness of contraceptive methods, are essential.