All Times EDT
Keywords: dropout, estimand, missing at random, quantile regression, rare disease, sensitivity analysis, small samples, trimmed mean
Despite best efforts at maximizing patient retention in randomized clinical trials, some dropouts are inevitable. In this talk, we compare different statistical approaches for tackling the resulting missing data problem, with a focus on small trials (40-60 patients) encountered in the rare disease setting. All approaches link to a common estimand associated with the clinical question of interest that is intended to be addressed using longitudinal data over a fixed duration for each trial participant. Parametric as well as non-parametric approaches are included in the simulation-based comparison, and a recommendation is made on the basis of type 1 error, power, bias and confidence interval coverage properties. In addition, attention is drawn to concerns with the commonly used mixed-effects model for repeated measures (MMRM) with the embedded missing at random (MAR) assumption for dropouts.