Estimation of encounter-level hospitalization costs: accuracy of a multivariate prediction model.

Wagner, Chen, and Barnett (2003 [this issue]) rightly identify the U.S. Department of Veterans Affairs’ (VA’s) historical lack of uniform cost data as an important barrier to economic analyses within the VA. Their article is an important contribution, and the cost estimates they generate should facilitate many types of health services research. At the same time, there are some aspects of their method that may limit the appropriateness of the cost estimates for certain types of analyses. We discuss several such aspects here. The model estimates costs in the long run, meaning it assumes all costs are variable. In the VA, however, managers often must make decisions that are essentially short run in nature. For example, managers are often unable to close facilities or even buildings due to political constraints. If the model had adopted a short-run perspective (i.e., treating some costs as fixed rather than variable), some of the parameter estimates (for example, the marginal cost of an extra day of hospitalization) would be expected to decline substantially. The reason for this is that fixed costs are high relative to variable costs in the short run (Schwartz and Mendelson 1991, 1994; Taheri et al. 1998, 1999; Taheri, Butz, and Greenfield 2000a, 2000b; Williams 1996; Reinhardt 1996; Roberts et al. 1999). For truly long-run analyses, in which it is appropriate and important to consider costs that managers consider fixed in the short run, cost estimates using the Wagner, Chen, and Barnett (2003) methodology are likely to play a