Implementation of Pattern-Mixture Models Using Standard SAS/STAT Procedures

Methods for dealing with missing data in clinical trials have been receiving increasing attention from the regulators, practitioners and academicians in the pharmaceutical industry over the past years. New guidelines and recommendations place a great emphasis not only on the importance of carefully selecting primary analysis methods based on clearly formulated assumptions regarding the missingness mechanism, but also on the necessity to perform a range of sensitivity analyses that stress-test the results of the primary analysis under different sets of assumptions. There are many methods that could be employed for sensitivity analyses, but some of them have not yet gained a wide-spread usage, partly because of the complex underlying theory, and partly because of lack of relatively easy approaches to their implementation. In this paper, we present a new way of using standard SAS/STAT ® procedures for multiple imputation (MI) in order to implement a class of methods based on Pattern-Mixture Models (PMMs). PMMs provide a general and flexible framework for sensitivity analyses that allows formulating assumptions regarding missing data in a transparent and clinically interpretable manner. Our implementation strategy is based on the core functionality available in PROC MI and MIANALYZE procedures and does not require any additional statistical programming outside DATA steps and sorting. A specific PMM-based method that we discuss here relies on clear and realistic clinical assumptions, while the general principles of our approach can be used to implement a range of other methods with different sets of assumptions.