Dapivirine ring HIV-1 prevention effectiveness for women engaged in vaginal and anal intercourse: insights from mathematical modeling

Abstract

Background:

The dapivirine vaginal ring reduces risk of HIV-1 acquisition in acts of vaginal intercourse (VI), and though it does not offer HIV-1 protection in acts of anal intercourse (AI), it may provide some overall risk reduction for women for whom the majority of sex acts are vaginal. We estimated the protective effect of the ring among women with high ring adherence engaged in both vaginal and anal intercourse.

Methods:

We developed a microsimulation model using data from the MTN-020/ASPIRE trial. Among women who reported any AI, we estimated the proportion of all sex acts that were AI. Model scenarios varied this proportion among women engaged in both VI and AI from 5–30%, including the trial-observed median proportion of 6.3% of all acts being AI. In primary analyses, dapivirine ring efficacy was model-calibrated at 70% for vaginal exposures and assumed to be 0% for anal exposures.

Results:

Among highly-adherent women for whom 6.3% of sex acts were AI, the ring reduced HIV-1 risk by 53% (IQR: 44, 60), with a decline to 26% (IQR: 16, 36) among women for whom 30% of acts were AI. Ring effectiveness was less than 40% among women for whom AI accounted for greater than 16% of all sex acts, although this represented less than 5% of all women in the ASPIRE trial.

Conclusions:

For the vast majority of women, including those who engage in AI, because most HIV-1 risk occurs in acts of vaginal sex, the dapivirine vaginal ring can provide important HIV-1 protection.

Introduction

Women in sub-Saharan Africa are at high risk of HIV-1 infection.¹ Prevention products that are both discreet and female-initiated are needed to address the unique challenges that women face in accessing and using biomedical HIV-1 prevention products.^2,3

The MTN-020/ASPIRE trial of a dapivirine vaginal ring for HIV-1 prevention estimated 27% HIV-1 risk reduction in modified intent-to-treat analyses.⁴ This value provides a population-level estimate of protection among women with varying levels of adherence, whereas the causal parameter of interest to prospective ring users and their providers is product efficacy when used, with analyses conditioned on high adherence estimating ring efficacy of 65%−75%.⁵ The ring does not provide protection in acts of anal intercourse (AI) due to minimal dissemination of drug from the vagina to the rectum,^6,7 though it may still provide some overall risk reduction by reducing transmission in acts of vaginal intercourse (VI). Prior analyses suggested that any AI in the trial would have a minimal impact on overall estimates of efficacy, yet estimated that the ring reduced HIV-1 risk among women reporting any AI by only 18% (95% CI: −57, 57).⁸ However, like the primary modified intent-to-treat analysis, this estimate was an average of ring effectiveness among a population of women with varying levels of adherence. As the ring moves toward public availability,^9–12 estimates of overall risk reduction with high adherence are needed for individual women and their providers to inform HIV-1 prevention method selection for women who engage in any AI.¹³ We therefore developed a microsimulation model of the ASPIRE trial to estimate the protective effect of the ring among women engaging in both vaginal and anal intercourse.

Methods

We developed a novel microsimulation model to reproduce the population, procedures, and outcomes of the ASPIRE trial, including 2,614 women confirmed HIV-1-negative at enrollment.⁴ Full model details are included in the technical appendix in Supplemental Digital Content (SDC) 1. In summary, a probabilistic model generates ring adherence, HIV-1 infection, and relationship formation, continuation, and termination in monthly timesteps, according to parameters determined from observed trial data conditional on age, sexual behaviors (number of partners and frequency of sex acts), baseline sexually transmitted infections and bacterial vaginosis, and study enrollment and exit times. HIV-1 status of male partners was largely unknown in the trial. The model therefore assigns prevalent and incident male partner HIV-1 status proportional to male age- and country-specific prevalence and incidence, respectively, with the probability of positive HIV-1 status modified by select factors observed in ASPIRE to be associated with HIV-1 acquisition, including participant age and reported condom use in the prior week. We used a sequential Monte Carlo approximate Bayesian computation calibration procedure¹⁴ to fit uncertainty parameters (including the per-act risk of receptive anal relative to receptive vaginal intercourse [RR_AI], among others) to trial-observed age-specific placebo arm incidence throughout the trial and overall placebo arm incidence in the final half of the trial.

Women reported their number of vaginal and anal sex acts at enrollment and month three in ASPIRE.⁸ From these reports, we estimated the proportion of total acts that were AI among women who reported engaging in both VI and AI and used the median proportion in primary analyses. We estimated the overall protective effect of the ring as the average per-exposure ring effect, defined as one minus the ratio of the number of HIV-1 infections out of the number of HIV-1-exposed acts in the dapivirine arm relative to the placebo arm. This quantity can be considered a weighted average of the per-vaginal exposure ring efficacy and the per-anal exposure ring efficacy (0%), weighted by the relative number of vaginal and anal sex acts and the HIV-1 risk for each act type. Among women who engage in only VI, the average per-exposure ring effect and per-vaginal exposure efficacy are equivalent. We evaluated the protective effect of the ring among women with high adherence across variation in the proportion of acts that are AI, ranging from 5 to 30% in 5% increments, as well as the trial-observed median and interquartile range. Furthermore, small imbalances in randomization arm at enrollment in ASPIRE⁴ may bias the estimated protective effect of the ring. Our primary analyses of the protective effect of the ring therefore re-randomized participants, with re-randomization stratified on site, enrollment bacterial vaginosis, and average monthly number of sex acts.

To identify values of per-vaginal exposure efficacy among highly-adherent women that were consistent with ASPIRE’s population-level estimate of 27% HIV-1 risk reduction,⁴ we conducted the primary modified intent-to-treat analysis in model scenarios that varied per-vaginal exposure efficacy from 60%−80%. Models included trial-observed small imbalances in randomization arm at enrollment and observed levels of non-adherence and AI. Additionally, because existing metrics for ring adherence may not correctly classify women’s use of the ring at the time of infection and AI may be both under- or overreported due to stigma or misinterpretation of survey items,¹⁵ we varied prevalence of non-adherence and AI in 5% increments from 0% to 50% and 0% to 30%, respectively. The model distributes non-adherence and AI proportional to participant characteristics associated with each to reflect differences observed within the trial.^4,8 We estimated the geometric mean hazard ratio across 250 simulations and used locally weighted regression with a quadratic polynomial to fit a smooth surface to these modeled scenarios to identify scenarios within 1% of the primary modified intent-to-treat hazard ratio of 0.73.⁴ Per-vaginal exposure ring efficacy values of 65%−75% were consistent with an HR of 0.73, while values of 60% and 80% were not (Supplement Figure 2; SDC 2). Each model scenario was therefore evaluated with per-vaginal exposure efficacy of 65%, 70%, and 75%.

Primary analyses used a single set of best-fitting parameters obtained in model calibration. We additionally evaluated the sensitivity of results to uncertainty in the per-act HIV-1 acquisition risk of AI relative to VI by repeating analyses with the top-fitting ten parameter sets that represented the range of per-act relative risk values (from 5.4 to 15.8) included in the model calibration prior distribution, compared to the per-act relative risk of 6.9 used in primary analyses. We simulated each model scenario 250 times and report the median and interquartile range of the protective effect of the ring. All analyses were completed in R.¹⁶

All trial participants provided written informed consent, and ethical review boards at each site approved the study protocol.⁴

Results

Among 2,562 (98.0%) women who responded to AI items at enrollment and/or month three, 468 (18.3%) women reported engaging in any AI, with a median of 6.3% of their acts being AI (IQR: 2.7, 16.7). Among those who reported any AI, AI accounted for more than 30% of all acts for 12.6% of women and for less than 5% of all acts among 42.1% of women. In simulations with 70% per-vaginal exposure efficacy and stratified re-randomization, the protective effect of the ring among highly-adherent women engaged in AI for 6.3% of total acts was 53% (IQR: 44, 60), a 24% reduction in effectiveness relative to women with high adherence engaging in only VI (Figure 1). The protective effect of the ring declined as the proportion of acts that were AI increased, from 70% among women with only vaginal exposures, to 57% (IQR: 48, 64) among women for whom 5% of acts were AI, to 26% effectiveness (IQR: 16, 36) among women for whom 30% of acts were AI (Figure 1). In a sensitivity analysis in which the per-act relative HIV-1 acquisition risk of AI was 15.8-fold higher than VI, the protective effect of the ring was somewhat lower than in our primary analysis, at 43% (IQR: 32, 52) among women for whom 6.3% of acts were AI, and declined to 16% (IQR: 3, 26) among women engaged in AI for 30% of her acts (Supplement Figure 1; SDC 2). Across all scenarios, stratified re-randomization increased estimated ring effectiveness by an absolute difference of 12% (IQR: 11, 14) (Figure 1).

Figure 1.

Overall ring-attributable HIV-1 risk reduction among women engaged in both receptive vaginal and anal intercourse estimated from simulations varying per-vaginal exposure efficacy and the proportion of a woman’s acts that are anal from among her total acts (RR_AI = 6.9 in all simulations). The horizontal dashed line indicates the per-vaginal exposure efficacy. The median and interquartile range of the average per-exposure effect are shown.

Translating these results to AI behaviors reported by the full cohort of women in ASPIRE, ring effectiveness was greater than 50% for more than 90% of all women (Figure 2). Among the 4.6% of women for whom AI accounts for greater than 16.7% of all sex acts, the protective effect of the ring was less than 40% (Figure 2), compared to 70% per vaginal exposure. Results were similar with per-vaginal exposure efficacy of 65% and 75% (Supplement Table 1; SDC 2).

Figure 2.

Among all women in the ASPIRE trial, the distribution of (1) engaging in only receptive vaginal intercourse or any receptive anal intercourse (inner circle) and (2) overall ring-attributable HIV-1 risk reduction given the reported proportion of all acts that were receptive anal intercourse (outer circle).

Discussion

In a mathematical model replicating the participants, procedures, and outcomes of the ASPIRE trial, simulations demonstrate that the dapivirine vaginal ring, when consistently used, confers substantial overall risk reduction to most women who engage in any AI, despite not conferring protection in acts of AI. Among the 18% of women who reported any AI, at the median frequency of AI of one in every 16 acts, we estimated that the ring reduced overall HIV-1 risk by more than 50% among women with high adherence. While this risk reduction diminished to 26% among women for whom one in every three acts is AI, that population of women was relatively small: 12.6% of women engaging in both VI and AI, translating to approximately 2.5% of the full cohort of participants in the ASPIRE trial. This supports the utility of the ring for the vast majority of women who cannot or choose not to use systemic PrEP,¹⁷ but nevertheless emphasizes the need for counselling approaches that help women identify the prevention option best-aligned with their context and sexual lifestyle, as well as for discreet methods that offer protection in both vaginal and rectal HIV-1 exposures.^18,19 Furthermore, in our model calibration, per-vaginal exposure efficacy values of 65% to 75% were consistent with the modified intent-to-treat estimate of 27% protection obtained in the primary ASPIRE analysis.⁴ These results support prior analyses conditioning on high adherence and emphasize that the ring offers high levels of HIV-1 risk reduction to women who may prefer an alternative prevention method to condoms and oral pre-exposure prophylaxis.

There is substantial uncertainty in the per-act HIV-1 acquisition risk of AI relative to VI, with a range of 5 to 16 based on per-act estimates of AI and VI obtained from different populations, time periods, settings, and study designs.^20–22 Our analysis benefited from use of a calibration approach that fit this value to observed data, yielding a per-act relative risk of AI of 6.9. However, model calibration was relatively insensitive to this parameter, perhaps due to competing risks for infection among this cohort of women at high risk of HIV-1 infection, and the resulting posterior credible interval is wide (90% CrI: 5.6, 15.5). Nonetheless, in sensitivity analyses assuming a per-act relative risk of AI of 5.4 to 15.8, results were qualitatively similar to the primary analysis.

In conclusion, among the majority of women who engaged in both VI and AI, AI is sufficiently infrequent that the ring provides substantial overall HIV-1 protection, supporting use of the ring for most women at risk of sexual acquisition of HIV-1.

MTN-020/ASPIRE Study Team

Study Team Leadership:

Jared Baeten, University of Washington (Protocol Chair); Thesla Palanee-Phillips, Wits Reproductive Health and HIV Institute (Protocol Co-chair); Elizabeth Brown, Fred Hutchinson Cancer Research Center (Protocol Statistician); Lydia Soto-Torres, US National Institute of Allergy and Infectious Diseases (Medical Officer); Katie Schwartz, FHI 360 (Clinical Research Manager)

Study sites and site Investigators of Record

Malawi, Blantyre site (Johns Hopkins University, Queen Elizabeth Hospital): Bonus Makanani;

Malawi, Lilongwe site (University of North Carolina, Chapel Hill): Francis Martinson

South Africa, Cape Town site (University of Cape Town): Linda-Gail Bekker;

South Africa, Durban – Botha’s Hill, Chatsworth, Isipingo, Tongaat, Umkomaas, Verulam sites (South African Medical Research Council): Vaneshree Govender,

Samantha Siva, Zakir Gaffoor, Logashvari Naidoo, Arendevi Pather, and Nitesha Jeenarain;

South Africa, Durban, eThekwini site (Center for the AIDS Programme for Research in South Africa): Gonasagrie Nair

South Africa, Johannesburg site (Wits RHI): Thesla Palanee-Phillips

Uganda, Kampala site (John Hopkins University, Makerere University): Flavia Matovu

Zimbabwe, Chitungwiza, Seke South and Zengeza sites (University of Zimbabwe, University of California San Francisco): Nyaradzo Mgodi

Zimbabwe, Harare, Spilhaus site (University of Zimbabwe, University of California San Francisco): Felix Mhlanga