OBJECTIVES: Health technology assessment (HTA) in the UK often requires that health-related quality of life is considered in a cost-utility analysis (CUA). Most studies contain missing data at some level. Missing utility data can misrepresent the denominator of the ICER. Despite this, there are no definitive guidelines on how to manage missingness for UK HTA, and analysts rely on judgment to address missingness. We intend for this research to formalise our recommendations for dealing with missing utility data for UK HTA.

METHODS: A simulated individual patient dataset, similar to those used to estimate utility values for a CUA submission for HTA in the UK, was developed. With this dataset, we simulated missingness at various levels and by different mechanisms. We assessed the performance of: complete case analysis (CCA), mean square estimation (MSE), linear mixed modelling (LMM) and multiple imputation via chained equations (MICE), and used this to make recommendations for handling and reporting of missing data in IPD that will be submitted to decision-makers in the UK.

RESULTS: Regardless of the mechanism or the level of missingness, MICE and LMM always resulted in substantially less error when calculating health state utility. Where data were missing at 30%, CCA and MSE were often associated with a substantially higher mean difference than MICE and LMM. The standard deviation was always substantially depressed regardless of mechanism or level of missingness when LMM was used.

CONCLUSIONS: It is our recommendation that an assessment of the mechanism and magnitude of missingness be made and reported for all data sets that are used to inform economic models for HTA in the UK as a minimum standard. Further, we recommend that CCA is almost never appropriate and our research, echoed by others supports the use of MICE as standard.