Missing data in health economic evaluations: handling the associated uncertainty

Abstract: The purpose of this thesis is to explore the problems associated with missing data in studies included in health economic evaluations in order to promote validity and improve the base for decision making. The literature study focus on adjustment methods from a realized dataset with individual data, but also covers methods to prepare and prevent the problem. Missing data, especially nonresponse in survey, is common and threatens validity for two reasons. First, it can bias results, for example when occurring more frequently in the right tail of cost distributions that are skewed due to rare expensive events. Second, variance estimation can be distorted. A battery of methods should preferably be applied to make the reason for missing data observed within the gathered data, since otherwise crucial modelling assumptions are needed. Robustness of models is therefore a subject for sensitivity analysis. Most efficient methods are based on maximum likelihood and multiple imputation, but are primarily large sample tools unless applied with Bayesian estimation. Their potential in the examined dataset were low due to the limited information. This thesis will hopefully provide help in reducing uncertainty accompanied with missing data, which in the end would lead to improved decision making.