OBJECTIVES: A number of mapping algorithms from a non-preference-based measure onto a target health utility measure have been developed and applied in cost-utility analysis (CUA). Conditions for a mapping algorithm to work well in CUA, however, are still unclear from the statistical perspectives. In the present study, we formulate the mapping problem as a missing data problem and clarify such conditions.

METHODS: We define a valid mapping algorithm as a mapping algorithm that can produce unbiased estimates of expected health utility in the experimental and control treatments in cost-utility analysis. This definition reflects the purpose of mapping (i.e., not for individual health utility prediction but for CUA). As mapping can be viewed as imputation of missing health utility data, we derive a sufficient set of statistical conditions for a valid mapping algorithm based on the theory of missing data analysis. We conducted a simulation study to investigate properties of a mapping algorithm under situations where the derived conditions are satisfied and violated.

RESULTS: The derived sufficient conditions indicate importance of the “complete overlap” of the source measure to the target health utility measure and a covariates that is omitted from a mapping algorithm but has an effect on the target health utility measure not captures by the source measure. The conditions cannot be verified from observed data in CUA, but can be supported using external data. A simulation study showed that when at least one of the derived conditions was violated, a mapping algorithm provided biased health utility estimates in CUA, and that prediction accuracy did not necessarily reflect performances of a mapping algorithm in CUA.

CONCLUSIONS: The derived conditions for a valid mapping algorithm provide a guidance for better practices in developing and selecting a mapping algorithm for CUA.