Understanding Adherence to and HIV-1 Protection from a Microbicidal Vaginal Ring

Topical vaginal microbicides and pre-exposure prophylaxis (PrEP) in the form of oral pills have been evaluated in efficacy trials for the prevention of HIV-1 infection over the past two decades with mixed results. In intention-to-treat analyses, some studies of PrEP agents, such as Truvada®, have shown positive efficacy results for reducing rates of HIV-1 infection in populations with high levels of adherence, while other studies even with the same agents and mostly among women, have not. However, in post-hoc analyses accounting for adherence to product use, efficacy estimates increased when adjusting for adherence in these same studies. The International Partnership for Microbicides developed a vaginal ring containing a PrEP agent called dapivirine to be replaced on a monthly basis, as a means to address the challenges women face adhering to PrEP agents requiring more frequent use. The goal of this dissertation was to further our understanding of the adherence to and HIV-1 protection from the dapivirine vaginal ring when used in a clinical trial setting among African women at high risk for HIV-1 acquisition. To this end, we utilized data from the MTN-020/ASPIRE study in which 2,629 African women participated in a randomized, placebo-controlled phase III trial of the dapivirine vaginal ring for the effectiveness of HIV-1 prevention conducted by the Microbicides Trials Network (MTN). The specific aims of this dissertation were 1) to assess the feasibility and effectiveness of real-time adherence monitoring strategies utilized within the MTN-020/ASPIRE study; 2) to evaluate the efficacy of a vaginal ring containing dapivirine compared to a placebo vaginal ring accounting for varying levels of adherence to product use; and 3) to identify demographic, sexually transmitted disease, and risk behavior correlates to adherence to inform future implementation of the dapivirine vaginal ring. We addressed Aim 1 by first describing the methods for implementation of the real-time adherence monitoring strategy. Using generalized estimating equations, we then assessed its effectiveness by evaluating whether percentages of blood plasma samples with dapivirine >95 pg/mL in active arm participants increased over calendar time. For the second aim, we conducted three analyses utilizing methods that adjust for adherence to ring use in different ways, specifically, per-protocol, principal stratification and marginal structural models. Four cut points for adherence were defined as follows: >95 pg/mL and >200 pg/mL of blood plasma, and ≥0.9 mg and >4.0 mg of drug released from the rings over a 28 day period of expected ring use. The per-protocol analysis aimed to emulate accrual under perfect protocol adherence, censoring participants at the last negative HIV-1 test before detection of pregnancy or the first occurrence of three or more non-compliant events in a 12 month period. In the principal stratification analysis, we first constructed logistic regression models to predict adherence defined as ≥0.9 mg and >4.0 mg of drug released from the rings at 6 months, and 12 months for sensitivity analyses, using baseline characteristics from only dapivirine arm participants’ data. Second, we applied this model to the entire cohort to obtain predicted probabilities of adherence, and last we constructed a Cox proportional hazards model with treatment arm, predicted probabilities of adherence and the interaction between the two variables. The marginal structural models analysis was based on the potential outcomes framework for causal inference and utilized inverse-probability-of-censoring weights (IPCWs) and an inverse-probability-of-treatment weight (IPTW) to create a pseudo-population free of confounding and selection bias as a result of informative censoring. We constructed pooled logistic regression models with restricted cubic splines to represent time on study, to develop two IPCWs, the first for the outcome of first pregnancy, and the second for the outcome of informative loss to follow-up. An IPTW model was also developed based on whether the participant was adherent using the ≥0.9 mg cutoff of drug released from the ring. A final pooled logistic regression model was constructed with a re-weighted population using a final stabilized weight calculated by multiplying together the two IPCWs and the IPTW. The last aim of this dissertation involved the discovery of correlates to adherence as defined by drug levels in plasma and estimated drug release levels from the vaginal ring. A multivariable generalized estimating equation model was constructed for each of the four adherence outcomes. Among a total of 2,629 women enrolled in MTN-020/ASPIRE, median age was 26 years (IQR: 22-31), and median follow-up time was 1.6 years (IQR: 1.1-2.3). In Aim 1, we found an upward linear trend of adherence over calendar time, from 63% during Quarter 1 2013 to 84% during Quarter 1 2015 (p < 0.0001). In Aim 2, in per-protocol analyses, we found HIV-1 efficacy estimates of 30.8% (3.6%, 50.3%) (p = 0.03), whereas from principal stratification analyses among the highest adherers HIV-1 efficacy estimates using 6 month data with the ≥0.9 mg cutoff was 53.6% (16.5%, 74.3%) (p = 0.01). Marginal structural models produced HIV-1 efficacy estimates ranging from 48.8% (21.8%, 66.4%) (p = 0.0019) to 56.5% (32.8%, 71.9%) (p = 0.0002). From the correlates of adherence analysis in Aim 3, we found that time on study, calendar time, primary partner knowledge that the participant was taking part in the study, and use of long-acting contraceptive methods were associated with ring adherence whereas younger age, ring worries, condom use during sex, episodes of menstrual bleeding and vaginal washing were associated with ring non-adherence. We found in Aim 1 that ongoing drug level testing as a marker of adherence in MTN-020/ASPIRE demonstrated the feasibility of real-time adherence monitoring while maintaining study blinding at the level of participants, sites, and study leadership. We found in Aim 2 that per-protocol analyses only modestly improved on intention-to-treat efficacy estimates, whereas estimates increased for strata of more highly adherent participants, and also when adjusting for time-varying confounding and selection bias due to informative censoring. Finally, in Aim 3 we found several important predictors of adherence that may prove to be useful for recruitment into future clinical studies and for dapivirine ring implementation efforts. Therefore, all hypotheses evaluated in this dissertation were shown to be true.