Handling missing data in randomised trials Seminar

Abstract
Missing data are a near-ubiquitous complication in medical research. While recognising that the ideal solution is to avoid missing data, I will focus on the analysis of randomised trials with missing data. I will discuss two features specific to randomised trials: easy handling of missing values of covariates, and accommodating the intention-to-treat principle. Because all analyses with missing outcomes involve untestable assumptions, I will describe recent work on performing sensitivity analysis, including varying the value of unidentified parameters and estimating missing-not-at-random mechanisms using repeated attempts. I hope that discussion will explore differences in approach between biostatistics and sample surveys.