Predicting effect of pre-exposure prophylaxis on HIV epidemics.

We agree with Angela Kashuba and colleagues (June 30, p 2409) that it is essential to develop new methods for measuring adherence to preexposure prophylaxis (PrEP) for HIV prevention in clinical trials. However, another important issue also needs to be addressed. Specifi cally, how can adherence data from clinical trials be interpreted and used to predict the success of PrEP in reducing transmission in the real world? We suggest that one approach is to use mathematical models to translate these data into population-level predictions. In each of the successful clinical trials of PrEP, several estimates of effi cacy have been calculated: an overall estimate of effi cacy calculated by use of data from all individuals irrespective of their level of adherence, and adherence-stratifi ed effi cacies for which each estimate was calculated for a specifi ed range of adherence. For example, in the IPrEx trial (in which adherence was defi ned on the basis of number of visits) the overall effi cacy was estimated to be 42% and adherence-stratifi ed effi cacies (based on an as-treated analysis) were estimated to be 68% (adherence ≥90%), 34% (50–90% adherence), and 16% (<50% adherence). Such data are necessary, but—even if adherence is measured accurately—are not suffi cient for predicting the success of PrEP in the real world. To make predictions, it is also necessary to know how population-level patterns of adherence to PrEP will aff ect success, where such patterns are defi ned in terms of the proportion of the population in each adherence category. The fi gure shows predictions from a mathematical model that translates adherence data from clinical trials of PrEP into populationlevel predictions. Contours show the reduction in transmission as a function of the proportion of the population that is highly (>90%) adherent to PrEP (Y-axis) and the average adherence in the rest of the population (X-axis). To generate these predictions, adherencestratifi ed effi cacy estimates, based on clinical trial data, were used. The fi gure illustrates that populationlevel adherence patterns (which cannot be measured in clinical trials) will be as important as effi cacy in determining success. Therefore even when it becomes possible to measure adherence accurately in clinical trials, these trials will not be able to generate all of the necessary data for predicting the success of PrEP in reducing HIV transmission in the real world.