Description: Parametric models of healthcare costs are common in program evaluation and general cost prediction. The most common approach is to identify and apply a single regression model, such as when applied to logarithmic or square root transformed costs, or a single conditional distribution, such as when applying maximum likelihood estimation to a gamma distribution. This presentation will extend the cost modeling toolbox to include multipart models based on partitioning the range of costs into multiple groups. GECDAC has found this approach can perform better than common models for predicting VA healthcare costs in among Veterans.
Intended Audience: The interested audience would likely be those who engage in parametric cost prediction models. Consequently, those who predict VA and VA program costs.