The meaning of adherence when behavioral risk patterns vary: obscured use- and method-effectiveness in HIV-prevention trials

Abstract

Background: Recently promising trials of innovative biomedical approaches to prevent HIV transmission have been reported. Participants' non-adherence to the prevention methods complicates the analyses and interpretation of trial results. The influence of variable sexual behaviors within and between participants of trials further obscures matters. Current methodological and statistical approaches in HIV-prevention studies, as well as ongoing debates on contradictory trial results, may fail to accurately address these topics.

Methodology/principal findings: Through developing a cumulative probability model of infection within HIV prevention trials, we demonstrate how adherence and sexual behavior patterns impact the overall estimate of effectiveness, the effectiveness of prevention methods as a function of adherence, and conclusions about methods' true effectiveness. Applying the model to summary-level data from the CAPRISA trial, we observe markedly different values for the true method effectiveness of the microbicide, and show that if the gel would have been tested among women with slightly different sexual behavior patterns, conclusions might well have been that the gel is not effective.

Conclusions/significance: Relative risk and adherence analyses in HIV prevention trials overlook the complex interplay between adherence and sexual behavior patterns. Consequently, they may not provide accurate estimates of use- and method-effectiveness. Moreover, trial conclusions are contingent upon the predominant sexual behavior pattern of participants and cannot be directly generalized to other contexts. We recommend researchers to (re)examine their data and use the cumulative probability model to estimate the true method effectiveness, which might contribute to resolving current questions about contradictory trial results. Moreover, we suggest taking into account the issues raised in the design of future trials and in population models estimating the impact of large-scale dissemination of prevention methods. Comprehension of the topics described will help readers to better interpret (apparently contradictory) trial outcomes.

Figure 1. Cumulative risk of infection as a function of the number of unprotected contacts () with an HIV-infected partner for women in the control and treatment group for different levels of adherence to the intervention gel (RR_100% and RR_50% indicate the relative risk of infection for 100% and 50% adherence to the gel).

Figure 2. Relative infection risk as a function of adherence for high-risk contacts under the cumulative probability model for three different behavior patterns

(  =  number of partners,  =  number of contacts per partner with condom use,  =  number of contacts per partner without condom use), assuming a true per-contact relative risk of and 100% adherence for low-risk contacts.

Figure 3. Relative infection risk as a function of adherence for high-risk contacts under the cumulative probability model for three different behavior patterns ( =  number of partners,  =  number of contacts per partner with condom use,  =  number of contacts per partner without condom use), assuming a true per-contact relative risk of and 50% adherence for low-risk contacts.