Abstract:
|
In many preclinical HIV prevention studies, efficacy is typically calculated as one minus the risk ratio (RR). The standard estimation methods used in these studies ignores (1) the time-to-event nature of the data; or (2) possible heterogeneity in susceptibility across animals; or (3) possible carry-over effect of prior challenges on the risk of infection during current challenge. In this study, we compared five different methods of measuring efficacy: 1) estimating the RR from a single, final binary outcome per subject; 2) a logistic model for sequences of challenges, assuming homogeneous risk across subjects and challenges; 3) GEE or a random effects logistic model that accounts for between-subject heterogeneity but not time heterogeneity; 4) a conditional logistic regression; and 5) a frailty Cox proportional hazards model to accommodate both between-subject and time heterogeneity. Our simulation results indicated that the model-based methods were robust compared to methods (1) and (2). Especially, the conditional logistic regression and the HR approach generated smaller biases and standard errors compared to GEE method.
|