Pitfalls of adjusting for mean baseline utilities/costs in trial-based cost-effectiveness analysis with missing data
Failure to account for baseline utilities/costs imbalance between treatment groups in cost-effectiveness analyses can result in biased estimates and mislead the decision making process. The currently recommended adjustment approach is linear regression, with estimates that are typically evaluated at the mean of the baseline utilities/costs. However, a problem arises whenever there are some missing follow-up values and the evaluation is restricted to the complete cases. Should the mean of the complete cases or the available cases baseline utilities/costs be used in generating the adjusted estimates? To our knowledge there is no current guideline about this choice in the literature, with standard software implementations often implicitly selecting one of the methods. We use two trials as motivating examples to show that the two approaches can lead to substantially different conclusions for healthcare decision making and that standard approaches which automatically resort to complete case analysis are potentially dangerous and biased. Analysts should therefore consider methods that can explicitly incorporate missing data assumptions and assess the robustness of the results to a range of plausible alternatives.
READ FULL TEXT