The effect of heterogeneity on HIV prevention trials

Abstract

Background Randomized controlled trials (RCTs) for the prevention of HIV heterosexual acquisition are usually conducted among adult African populations with high heterogeneity in individual risk of infection.

Purpose The objectives were to (a) review how this heterogeneity has been considered when designing and interpreting such RCTs, (b) evaluate its effect on the findings and the statistical power of these trials, and (c) assess the potential advantages of using the crossover design with single failure-time endpoint.

Methods Individual-level HIV prevention RCTs conducted in Africa and published in the period 1998–2008 were reviewed. Using Monte Carlo simulations and statistical calculations, we assessed the effect of heterogeneity on the findings and the statistical power of HIV prevention RCTs.

Results All reviewed RCTs used the parallel design. The heterogeneity in individual risk of infection within study sites was not used for stratification nor generally considered in the design and interpretation of RCTs. Simulations showed that in the context of high HIV incidence, high heterogeneity can lead to a substantial underestimation of the impact of an intervention and reduced statistical power. Calculations demonstrated that the crossover design allowed for similar or better estimation and statistical power. The crossover design has the ethical advantage of sharing the potential benefits and risks of the intervention between participants.

Limitations Only trials with two treatment arms and two follow-up periods were modeled. The baseline risk of infection of each participant was assumed to be constant over time and HIV status was assessed at the end of each follow-up period.

Conclusions The heterogeneity in individual risk of HIV infection is an underestimated problem which should be taken into account when designing and interpreting RCTs that test prevention methods of HIV heterosexual acquisition in adult African populations with high HIV incidence. When the effects of tested interventions are rapidly reversible, the use of the crossover design should be considered.