Design of epidemiological studies measuring genital and plasma HIV-1 outcomes: lessons from a randomised controlled trial

Objective  HIV-1 genital viral loads have not been extensively used as markers of HIV transmissibility. We set out to determine whether the variability of genital HIV-1 RNA over time necessitates design adjustments in studies measuring genital shedding to account for this variability.
Methods  We used data from a completed trial of HSV suppressive therapy to estimate the correlation of plasma and genital HIV-1 RNA quantities sampled at different times. These correlation estimates were used to estimate the relative sample sizes needed to detect an impact on HIV-1 genital and plasma quantities assuming a variable number of pre- and post-randomisation repeated measurements. The treatment effect on quantities of genital and plasma HIV-1 RNA were analysed using random effects linear regression.
Results  Post-randomisation plasma HIV-1 RNA was highly correlated within-women, while genital HIV-1 RNA was less strongly correlated. Related to this, the sample size required to detect a treatment effect on genital HIV-1 RNA decreased with increasing numbers of post-randomisation measurements up to 6–7 measurements, but varied less for plasma HIV-1 RNA. In contrast, repeated pre-randomisation measurements of plasma HIV-1 RNA increased study power more than genital HIV-1 RNA because of the high correlation of plasma HIV-1 RNA measurements between the pre- and post-randomisation samples. Re-analysis of the trial data illustrated the increased precision of the treatment effect on genital HIV-1 with increasing post-randomisation measurements.
Conclusions  Designs allowing for repeated post-randomisation measures should be used to increase the precision in estimates of genital HIV-1 RNA. Repeated post-randomisation measurements of plasma HIV-1 RNA are of limited benefit.