HIV-infected women represent slightly more than half of all people living with HIV worldwide, the majority of whom have childbearing potential and may need adequate contraception [1,2]. Effective contraception is an essential component of family planning and prevention of mother-to-child transmission. An increasing number of HIV-infected women are receiving antiretroviral therapy (ART) , and combined oral contraceptive (COC) pills are the most common method of contraception worldwide , raising the potential for drug–drug interactions and potentially, reduced contraceptive effectiveness. A key question for the WHO, country policy makers, and providers is whether hormonal contraceptives remain effective and safe in women using ART. Pharmacokinetic interactions refer to the action of one drug on the absorption, metabolism, binding, or excretion of another drug. Drugs that induce (or inhibit) liver enzymes such as cytochrome P450 (CYP450) 3A polypeptide 4 (CYP3A4) or uridine diphosphate glucuronyltransferase (UDPG-T) are known to affect the pharmacokinetics of COCs . Drugs that induce these enzymes increase the metabolism of ethinyl estradiol and progestins in oral contraceptives, resulting in lower serum levels of the hormones. Contraceptive effectiveness of COCs could be compromised if other drugs affect the contraceptive steroid pharmacokinetics such that ovulation occurs, and/or changes in cervical mucus fail to prevent fertilization.
Antiretroviral (ARV) drugs that induce CYP3A4 or UDPG-T include protease inhibitors and nonnucleoside reverse transcriptase inhibitors (NNRTIs), such as nevirapine (NVP) or efavirenz [5–7]. In a previous pharmacokinetic study, Mildvan et al.  evaluated the effects of NVP on the pharmacokinetic of a single dose of a COC containing 35 μg of ethinyl estradiol and 1 mg of norethindrone in 10 HIV-infected women on combination ART (dual NRTIs and indinavir, nelfinavir, or saquinavir) with and without NVP. Compared with plasma concentrations observed prior to NVP administration, the addition of NVP (200 mg daily increased to 200 mg twice daily for 28 days) resulted in a significant (29%) reduction in ethinyl estradiol levels. This study concluded that COCs should not be the primary birth control method for women of childbearing potential who are treated with NVP-based ART. Another small study by Stuart et al. , however, found that ethinyl estradiol and levonorgestrel concentrations were not decreased in HIV-infected women taking NVP-based ART, and no women ovulated, but the sample size was insufficient to adequately assess differences in ovulation rates and pregnancy was not evaluated.
In this study, our overall goal was to see whether COCs remained effective in women taking ART, compared to those not taking ART. Because NVP-based regimens are commonly used worldwide, particularly in sub-Saharan Africa, we chose to study women using NVP-based ART. Our primary objective was to compare ovulation rates between two groups of women: those taking COCs concurrently with NVP-containing ART and those taking COCs alone. Our secondary objectives were to compare pregnancy rates between these groups and to evaluate the safety of the concurrent administration of COC and NVP-containing ART.
We conducted this nonrandomized comparative clinical trial at two sites in Uganda and South Africa between June 2009 and May 2011. The sites included the University of the Witwatersrand – Wits Reproductive Health and HIV Institute (WRHI) in Johannesburg, South Africa and Makerere University – Mulago Hospital in Kampala, Uganda. The FHI 360 Protection of Human Subjects Committee (PHSC), the University of the Witwatersrand Human Research Ethics Committee, and the Uganda National Council for Science and Technology – National HIV/AIDS Research Committee approved the study.
We enrolled HIV-positive women aged 18–35 years old who had regular menses, were sexually active, had no medical contraindications to COC use, were using or willing to use COCs for the duration of their study participation, and were not taking other liver-enzyme inducing medications. The ART group included 196 women who had been on a stable NVP-based ART regimen for at least 3 months, whereas the control group (non-ART), included 206 women not yet eligible for ART who had CD4+ cell counts at least 350 cells/μl. We treated all women with COCs containing 30 μg of ethinyl estradiol and 75 μg of norgestrel. To achieve a steady state of contraceptive hormone levels, all new COC users took COCs for at least one preevaluation cycle before we started measuring ovulation. Women were also provided with and counseled regarding the use of condoms for prevention of HIV transmission and sexually transmitted infection (STI) prevention.
Our primary endpoint was ovulation. We considered a single progesterone value of 10 nmol/l during each cycle during the first two cycles (8 weeks) of the study as presumptive evidence of ovulation. We also did a sensitivity analysis using 20 nmol/l as the definition of probable ovulation. We followed each woman for six treatment cycles after COC steady state was achieved. We assessed for presumptive ovulation in the first two treatment cycles (8 weeks) by measuring blood progesterone levels weekly. We tested for pregnancy every 4 weeks during all six treatment cycles (24 weeks). In addition to pregnancy testing every 4 weeks, we asked participants about adverse events, concomitant medications, pill continuation, bleeding patterns, and changes in ART.
Secondary outcomes were pregnancy, as detected by monthly urine pregnancy testing, and adverse events. We estimated that, if ovulation rates were 3% in each group per cycle with intrasubject correlation of 0.20, a sample size of 185 women per group, with ovulation assessed over two treatment cycles would give us 80% power to conclude that the ovulation rate in the ART group was no more than 4% larger than in the non-ART group.
Our primary analysis population was a subset of the enrolled population excluding participants who either never returned for follow-up or did not have at least one cycle of COC use (non-ART group) or one cycle of concurrent COC and ART use (ART group). Participants were censored from analysis when they reported discontinuing COCs. Differences in participant characteristics were summarized and compared by treatment groups using Cochran–Mantel–Haenszel and Exact tests for categorical variables and the Wilcoxon–Mann–Whitney test for continuous variables. The primary analysis was based on a comparison of overall ovulation rates over two cycles between treatment arms. To account for the heterogeneity in the data due to multiple test cycles contributed by each participant, we used stratified logistic regression with generalized estimating equations (GEEs). We used the same modeling techniques to estimate the odds ratio (OR) of ovulation between groups while adjusting for potential cofounders, selecting our final model using a backward selection procedure. Potential interaction effects between treatment group and covariates were explored as part of the analysis process. We estimated and compared the probability of pregnancy over time using Kaplan–Meier estimates and log-rank tests. We used a Cox regression model to adjust for potential confounders for the pregnancy analysis. All statistical analyses used SAS (version 9.3; SAS Institute, Cary, North Carolina, USA).
We screened 562 women for this study, of whom 402 met eligibility criteria and were enrolled in this study: 196 in the ART group and 206 in the non-ART group. We excluded from analyses 24 women in the ART group and 28 women in the non-ART group who either never returned for follow-up or used COCs for less than one cycle. Thus, the primary analysis population included 350 participants (172 in the ART group and 178 in the control group). Of these participants, 328 (93.7%) completed the first 8 weeks and 316 (90.3%) completed the study: 159 (92.4%) in the ART group and 157 (88.2%) in the non-ART group. Thirteen participants (3.7%) were lost to follow-up and 21 (6%) discontinued the study early: one participant in the non-ART group started ART, 12 participants discontinued COCs, and eight discontinued for medical reasons.
Enrolled participants (Table 1) had a median age of 29 years and median CD4+ cell count of 486; all were black, and most were unmarried, nonsmokers, and of low educational status (between 7 and 11 years of school). Women in the ART group were older and had lower CD4+ cell counts. The mean age for the ART group was 30 and it was 27 for non-ART group (P = <0.001). The median number of prior pregnancies was two for both groups, though more women in the ART group had more than three prior births; approximately, half of women in both groups reported a prior unintended pregnancy. The most common contraceptive method used at baseline was male condoms (77.6% in the ART group and 65.5% in the non-ART group). Similar proportions of women were using COCs at enrollment (20.9% in the ART group and 26.2% in the non-ART group). Upon study entry, 48% of participants in the ART group reported ever having used COCs, compared with 53.4% in the non-ART group (Table 1). More women in the ART group had previous histories of various medical problems (data not shown).
In the ART group, 43 of 168 (26%) had presumptive evidence of ovulation (progesterone >10 nmol/l) in cycle 1, 30 of 163 (18%) in cycle 2, and 18 of 163 (11%) in both cycles. In the non-ART group, 26 of 168 (15%) ovulated in cycle 1, 31 of 165 (19%) in cycle 2, and 20 of 165 (12%) in both cycles. Overall, 69 out of 336 participants (21%) ovulated in cycle 1, 61 of 328 (19%) in cycle 2, and 38 of 328 (12%) ovulated in both cycles. We found statistically significant differences in ovulation rates between the ART and non-ART groups in the first cycle, but not the second cycle: the difference in proportions for cycle 1 was 0.11 (95% CI 0.02–0.20), and for cycle 2, it was −0.01 (95% CI −0.09 to 0.07). In a sensitivity analysis (ovulation defined as progesterone >20 mmol/l), 37 of 168 (22%) of the ART group ovulated in cycle 1, 28 of 163 (17%) in cycle 2, and 14 of 163 (9%) in both cycles. In the non-ART group, 25 of 168 (15%) ovulated in cycle 1, 29 of 165 (18%) in cycle 2, and 19 of 165 (12%) in both cycles. Ovulation rates differed by site; in Uganda, 28% of women ovulated in both cycles 1 and 2, and in South Africa 14% of women ovulated in cycle 1 and only 10% in cycle 2.
Before adjustment, we found no significant association between treatment groups in overall ovulation rates (unadjusted OR for ART group, compared with non-ART group was 1.36, 95% CI 0.85–2.18). When we controlled for potential confounders, ART use still had no significant effect on risk of ovulation (OR 1.47, 95% confidence interval, CI 0.85–2.55). In multivariable analysis (Table 2), the only factors associated with ovulation included using COCs before enrollment (OR 0.38, 95% CI 0.20–0.70), age between 29 and 32 years compared to age 33 and above (OR 0.51, 95% CI 0.27–0.98), and being at the Johannesburg, South Africa site compared to the Kampala, Uganda site (OR 0.28; 95% CI 0.16–0.46). No evidence of an interaction was found between treatment group and either site or using COCs before enrollment. Missed pills, CD4+ cell count, marital status, education, BMI, smoking, parity, previous unplanned pregnancy, time since last pregnancy, new/worsening health problems during the study, other medication use, starting pill packs late, condom use, and intermenstrual bleeding were not associated with risk of ovulation.
Pregnancy probabilities at 6 months were 4.8% (95% CI 1.2–8.4%) in the ART group and 5.0% in the non-ART group (95% CI 1.2–8.9%). The log-rank test for time to pregnancy was not statistically significant between the two groups (P = 0.98), and overall pregnancy incidence did not differ significantly between groups: nine pregnancies in each group for a rate of 10.0 per 100 women-years (95% CI 4.6–19.1) in the ART group and 10.1 per 100 women-years (95% CI 4.6–12.1) in the non-ART group. In multivariable analysis (Table 3), only two factors were significantly associated with risk of pregnancy: participants’ report of missing three or more COC pills in a row (hazard ratio 13.06, 95% CI 2.51–67.95), and participant's report of always using condoms (hazard ratio 0.12, 95% CI 0.02–0.76). All other factors, including ART use, CD4+ cell count, age, site, COC use before enrollment, marital status, education, BMI, smoking, parity, previous unplanned pregnancy, time since last pregnancy, new/worsening health problems during the study, other medication use, starting pill packs late, and intermenstrual bleeding were not associated with risk of pregnancy. We found no evidence of an interaction between missed pills, site, or condom use and treatment group.
The most common adverse events reported were respiratory tract infections, infections/infestations (such as malaria), and gastrointestinal disorders, with no differences between the two groups. No differences were also evident in possible COC associated side-effect (Table 4). Five serious adverse events (SAEs) were reported, all in the non-ART group. Three were unrelated to COC use: malaria, cellulitis, and fracture, and two were possibly related: menorrhagia and dizziness (leading to pill and study discontinuation).
Self-reported COC adherence was high and similar in both the ART group and non-ART group (Table 5). Self-reported condom use also did not differ between groups during the study. At week 24, 104 of 147 (70.8%) participants in the ART group reported always using condoms, compared with 95 of 139 (68.4%) of the non-ART group (Table 4). Self-reported pill adherence was higher among the participants who were COC users before enrollment than those who were not, and among women in Kampala compared with women in Johannesburg (data not shown). At all visits, more participants who were new COC users reported either completely stopping birth control pills, missing three or more pills in a row, or starting pill packs late than participants who were already using COCs at enrollment.
We also evaluated reported intermenstrual bleeding monthly, and found no difference between groups. For example, at week 24, only one of 147 (0.7%) woman in the ART group reported intermenstrual bleeding compared with three of 139 (2.2%) in the non-ART group (P = 0.36).
We found that concurrent use of NVP-containing ART did not affect overall COC effectiveness in women taking this common ART regimen compared with women not taking ART, a finding that should be reassuring for programs providing contraception to HIV-infected women taking NVP-based ART regimens.
Current guidelines state that women taking ARVs who use COCs should either consistently use a reliable method of barrier contraception to both prevent HIV transmission and compensate for any possible reduction of the effectiveness of COCs, or increase the dose [2,9]. Such recommendations create potential issues for women, in particular the impracticality of using a backup, presumably barrier and thus male-initiated, method of contraception such as condoms. Furthermore, increased doses of estrogen or progestin may also increase the risk of adverse events related to COCs, such as thromboembolic risk . Though our confidence limits were wide, our findings are generally reassuring. Our findings are also supported by recent small pharmacokinetic studies showing no pharmacokinetic interactions between NVP-based ART and COCs [8,10].
Contrary to other clinical trials [11–13], we found high rates of ovulation in COC users, regardless of ART use. In typical use, women miss a lot of pills, even if they do not report doing so . Although we used a serum progesterone level of more than 10 nmol/l to detect ovulation, and we did not have ultrasound or other hormonal correlates of ovulation, our findings were robust even when a definition of level more than 20 nmol/l was used, so we believe that most of these were true ovulations. Regardless, ovulation alone does not always lead to pregnancy, and even imperfect use of COCs provides contraceptive protection. Though our pregnancy rates were relatively high, they were similar to other typical use pregnancy rates in the region . These data support the need for expanding the contraceptive method mix for women with HIV who do not desire pregnancy, particularly with long-acting forgettable more effective methods.
We found that participants already using COCs before enrollment were more likely to use them effectively. These participants were approximately 60% less likely than new COC users to ovulate. This might be explained by the fact that women who were COC users before enrollment were more adherent to their COC pill regimens. It is possible that poor compliers drop out over time, and women who continue COC use have persisted despite initial side-effect issues. Another interesting finding is that participants in Kampala were almost four times more likely than those in Johannesburg to ovulate, but we found no interactions between site and treatment group, and self-reported adherence was actually higher in Kampala than in Johannesburg. Finally, the finding that participants aged 29–32 were less likely to ovulate than older or younger women could also reflect adherence issues or may be by chance. Although no women entered the study desiring pregnancy, we did not ask women about risk perceptions regarding pregnancy.
We also found that participants who reported missing three or more COC pills in a row were 13 times more likely to get pregnant. Although study participants often overreport adherence, this particular question appeared to be a sensitive marker of contraceptive pill adherence over time. Similarly, those reporting always condom use were at approximately a 10th as likely to get pregnant. Unlike with risk of ovulation, site and age were not associated with risk of pregnancy once missed pills and condom use were controlled for. Report of missing three or more pills in a row was not associated with risk of ovulation, though few women reported missing pills in the first 8 weeks of the study. Overall, our findings suggest that reproductive health programs should provide increased support to facilitate adherence and continuation of COC use among HIV-infected women who do not desire fertility and choose to use COCs, particularly for new COC users. Similarly, regular condom use, both for additional pregnancy and STI prevention, should be encouraged.
We found no differences in self-reported contraceptive pill adherence between the ART group and the non-ART group, as well as no difference in self-reported condom use between groups. Although self-report is often inaccurate, because these factors were associated with pregnancy, we believe them to be at least partially accurate, especially over time. Finally, no evidence of differences in adverse events between COC users taking ART and those taking COCs alone is encouraging, given that NVP is commonly included in the first-line ART regimen recommended by the WHO and used in many countries.
Our study was the first large study to explore clinical interactions between NVP-containing ART and COCs, and was thus able to detect differences with reasonable precision. Our findings can be used to inform future clinical investigations evaluating ART and COC drug interactions. Though our study did not measure contraceptive drug levels, ovulation and pregnancy are more meaningful clinical endpoints. To achieve a steady pharmacokinetic state of contraceptive hormones, all participants had to take COCs for at least one cycle before we started measuring ovulation.
One limitation is that because we did not include a group of women taking ART but not COCs, we could not look at the effect of COCs on ART effectiveness. Another limitation is that our trial was not randomized, because it would have been unethical to randomize HIV-infected women who needed ART to ART treatment or no treatment. We considered several study designs before beginning this study, and decided that a nonrandomized clinical trial was the most ethical and feasible. In attempts to make health status (in terms of risk of ovulation and pregnancy) more comparable between groups, we only enrolled participants who had regular menses and who had been on a stable ART regimen for at least 3 months (for the ART group). We also evaluated and controlled for several potential confounders such as age, CD4+ cell count, and BMI.
Another limitation of our study is that ovulation rates were higher than expected. On the basis of previous studies [11–13], we estimated that only approximately 3% of women would ovulate in the control group. However, ovulation rates were much higher overall. Thus, the precision of our estimates is lower than we expected due to the higher than expected ovulation rates, as reflected in the presented confidence intervals.
Because different ART regimens have differing effects on drug metabolism, our findings cannot be generalized to non-NVP-containing ARV, such as efavirenz-based regimens. In a recent study, Landolt et al.  found that coadministration of COCs with an efavirenz-based ART regimen in HIV-infected women was associated with possible risk of ovulation and lower efavirenz levels. Further studies including pharmacokinetic studies evaluating both progestin and ART levels, and adequately powered studies of clinical outcomes such as ovulation and pregnancy, are urgently needed for different ART regimens, such as protease inhibitors, and newer forms of hormonal contraceptives, such as implants. With the advent of preexposure prophylaxis, the issue of ART and COC interactions is also potentially relevant for HIV-negative women at higher risk for HIV acquisition. Understanding clinical outcomes of hormonal contraceptive use in women taking ART should become a top priority in countries with high HIV prevalence seeking to limit unwanted pregnancy among HIV-infected women.
Women living with HIV including those on ART, who desire to avoid pregnancy, require effective contraception. These women can be reassured that NVP-based regimens do not impact effectiveness of COCs. Women who choose to use COCs need adequate counseling and support, particularly when starting the method. Because of issues of adherence with client-dependent methods such as COCs, efforts to expand the contraceptive method mix to include long-acting methods such as intrauterine devices and implants for women with HIV who do not desire pregnancy should also be a priority.
The authors gratefully acknowledge the study participants, the funder, and the study teams in Johannesburg, South Africa, and Kampala, Uganda. All contributing authors drafted the article or provided critical insight during revisions; and provided final approval for the version to be published. K.N. conceived of the study, secured funding, designed the study, and interpreted the data; S.D-M. and F.M. contributed to study design and acquisition of data; K.D. and A.L. contributed to study design and data interpretation; C.K. performed the statistical analysis and contributed to study design; L.M., S.N.,V.E., and B.M. contributed to acquisition of data.
This study was registered with clinicaltrial.gov:ClinicalTrials.gov Identifier: NCT00829114.
Conflicts of interest
The authors of this paper declare no conflict of interest. The study was funded by the U.S. Agency for International Development (USAID) under the terms of Cooperative Agreement GHO-A-00-09-00016-00, the Preventive Technologies Agreement. The contents are the responsibility of FHI 360 and do not necessarily reflect the views of USAID or the United States Government.