For the analysis of plasma TFV concentrations, we used a case–cohort design,13 in which a ran- dom subcohort was selected from the active study- product groups (cohort) and was enriched with all remaining participants who had undergone HIV-1 seroconversion (cases). For every participant who underwent seroconversion, we sampled ap- proximately three participants who remained un- infected with HIV-1 throughout participation. Pre- dictors of TFV detection were assessed with the use of generalized estimating equations with a logistic link function and exchangeable working correlation matrix. We conducted multivariate sur-