Developing appropriate methods for addressing missing data in health economic evaluation

Abstract: Missing data occurs in nearly every cost-effectiveness study, for example, because patients are lost to follow-up or fail to respond to quality-of-life or resource use questionnaires. A major concern that arises with missing data is that individuals with missing information tend to be systematically different from those with complete data, and inferences based on the complete cases will be biased. These concerns face both cost-effectiveness analyses based on a single study and those that synthesise data from several sources in decision-analytic models. While standard multiple imputation (MI) methods have been proposed for handling missing data in cost-effectiveness analysis, these may be insufficient in many settings. For example, they assume that individual observations are independent (which may be implausible in multicentre/multinational studies or meta-analysis of individual-participant data) or that the imputation model is correctly specified. Here, I consider the use of multilevel MI and doubly-robust methods such as the ‘robust MI’ to address these concerns, assuming data are missing at random (MAR). Further, I propose a practical framework for exploring the robustness of the results to departures from MAR and consider approaches for explicitly addressing potential missing not at random (MNAR) mechanisms such as pattern-mixture modelling. The alternative methods are illustrated in a re-analysis of patient-reported outcome measures (PROMs) for comparing provider performance.