Overview

This course will introduce participants to the key concepts and methods relevant for analysing clinical trials when some data are missing. We will describe missing data assumptions and Rubin's framework for classifying them, based on missing completely at random, missing at random (MAR), and missing not at random, and what these imply when missingness is due to dropout or the occurrence of intercurrent events. We will describe the use of mixed models and multiple imputation to handle missingness under MAR, and finally discuss methods for conducting missing data sensitivity analyses, including reference based imputation methods.

The course will cover:

• Introduction to estimands and missing data in trials; review of missing data assumptions & terminology (e.g. missing at random)
• Performing analyses under missing at random for continuous outcome data, using mixed models and multiple imputation (including consideration of retrieved dropout multiple imputation)
• Performing analyses under missing at random for binary data, using full conditional specification for multiple imputation with a GEE analysis model

• Sensitivity analyses using multiple imputation, including reference based imputation methods

Please note: Each of the above will be presented in a one hour lecture, followed by a two hour interactive computer practical. Computer practicals will be taught using R and so having R or R Studio installed on your personal laptop/computer is required to participate in the practicals.

Please note: There are 4 sessions that make up this Course. They will take place on Mon. 9th, Tues. 10th, Thurs. 12th & Fri. 13th October 2023, and will be run online via Zoom.

Who is this event intended for? This course is intended for clinical trial statisticians who are interested in learning more about statistical methods for handling missing data in clinical trial analyses.

What is the benefit of attending? By the end of the course participants will be familiar with the key concepts (e.g. missing at random) and statistical methods (e.g. multiple imputation) relevant when estimating treatment effects in trials where some data are missing.