Randomized controlled trials 6: On contamination and estimating the actual treatment effect

Abstract

The intention-to-treat analysis is the gold standard for evaluating efficacy in a randomized controlled trial. However, when non-adherence to randomized treatments is high, the actual treatment effect may be underestimated. The impact of drop-out from the intervention group or drop-in to the control group may be controlled by trial design, increasing the sample size, effective study execution, and a prespecified analytical plan to take contamination into account. These analyses may include censoring at time of co-interventions associated with stopping treatment, lag censoring which allows an additional period after discontinuation of study treatment to account for residual treatment effects, inverse probability of censoring weights (IPCW), accelerated failure time models, and contamination adjusted intent-to-treat analysis. These methods are particularly useful in assessing the “prescribed efficacy” of the study treatment, which can aid clinical decision-making.