Working Paper: NBER ID: w25233
Authors: David Cutler; Kaushik Ghosh; Irina Bondarenko; Kassandra Messer; Trivellore Raghunathan; Susan Stewart; Allison B. Rosen
Abstract: Partitioning medical spending into conditions is essential to understanding the cost burden of medical care. Two broad strategies have been used to measure disease-specific spending. The first attributes each medical claim to the condition listed as its cause. The second decomposes total spending for a person over a year to the cumulative set of conditions they have. Traditionally, this has been done through regression analysis. This paper makes two contributions. First, we develop a new method to attribute spending to conditions using propensity score models. Second, we compare the claims attribution approach to the regression approach and our propensity score stratification method in a common set of beneficiaries age 65 and over drawn from the 2009 Medicare Current Beneficiary Survey. Our estimates show that the three methods have important differences in spending allocation and that the propensity score model likely offers the best theoretical and empirical combination.
Keywords: medical spending; health economics; propensity score models; cost attribution
JEL Codes: I1
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| propensity score model (C52) | accurate allocation of spending to comorbid conditions (H51) |
| claims-based approach (G22) | more costs attributed to acute medical conditions (I12) |
| propensity score model (C52) | lower out-of-sample mean squared error (C51) |
| regression approach (C29) | underestimation of spending on prevalent and less severe conditions (H51) |