Online Program

Missing Data in Longitudinal Clinical Trials

*Edward Vonesh, Northwestern University 

Keywords: Missing at random, Non-ignorable dropout, Multiple imputation, Inverse probability of weighting, Pattern mixture models, Shared parameter models

Missing data due to dropout is a frequently occurring problem in longitudinal clinical trials involving repeated measurements over time. This occurs most often in prospective randomized controlled trials where observations are planned at specified times during the course of follow-up. When there are no missing values present, all of the standard statistical methods used for analyzing repeated measurements data such as maximum likelihood (ML) and generalized estimating equations (GEE) allows one to draw valid inference provided the complete data modeling assumptions are met. However, when we have incomplete data due to dropout, we run into the problem of never being able to verify these assumptions for the unobserved data. This short course will examine different missing data mechanisms that lead to incomplete data as well as various methods one can use to analyze longitudinal data when missing values are assumed to be either ignorable or non-ignorable. Particular emphasis will be placed on those methods which are easily handled using readily available software in SAS. The material will be illustrated using examples from several different clinical trials.