A Bivariate Regression-Based Cost-Effectiveness Analysis

In the economic evaluations of health interventions, the cost-effectiveness of a treatment often exhibits considerable variability across patients, underscoring the need for incorporating covariates into the analysis. A subgroup analysis is often recommended based on a stratification approach. Such methods may be insufficient and somewhat arbitrary, with subgroups defined in a post hoc fashion. The present work incorporates patient heterogeneity into economic evaluations using patient-level covariates in a bivariate regression model, allowing for interactions among covariates. Covariate-specific metrics are then defined and investigated, which include the incremental net benefit and a cost-effectiveness probability. Lower confidence limits are obtained using fiducial and parametric percentile bootstrap approaches, which accurately maintain the coverage probability. However, confidence limits obtained using the delta method are inaccurate. Based on expected values of the lower confidence limits, the parametric percentile bootstrap approach is recommended. An application on the cost-effectiveness of treatments for schizophrenia is used to demonstrate that even though cost-effectiveness may not hold at the population level, it can hold at certain covariate values. The proposed analysis can be used to identify covariate combinations that are major determinants of the cost-effectiveness for a new intervention. R-codes are provided as supplementary information, with an illustration of their execution.