Abstract
One question of particular importance in phase III HIV vaccine trials is the choice of efficacy measure (EM) to validly and precisely estimate the true vaccinal efficacy. Traditional EMs, based on hazard rate ratio (HRR) or cumulative incidence ratio (CIR) are time-sensitive to mode of vaccine action and population heterogeneities. Through Monte-Carlo simulation, the performance of HRR and CIR based EMs are examined across different trial designs and vaccine and population characteristics. A new EM based on log-spline hazard regression (HARE) is proposed. Given that vaccinal properties (mode of action, time-lag, waning) are unknown a priori, appropriate selection of EM is problematic, and HRR and CIR can be unreliable to estimate the true maximum efficacy of candidate products. Non-random sexual mixing can exacerbate biases in HRR and CIR. HARE can offer valid estimation across different modes of vaccine action and in presence of frailty effects, contrary to its traditional counterparts. Our simulation studies highlight the weaknesses of widely used EMs while offering guidelines for trial design and suggesting new avenues for statistical analysis.