Objective: To evaluate the predictive accuracy of the Diagnostic Cost Group (DCG) model using health survey information.
Data sources/study setting: Longitudinal data collected for a sample of members of a Dutch sickness fund. In the Netherlands the sickness funds provide compulsory health insurance coverage for the 60 percent of the population in the lowest income brackets.
Study design: A demographic model and DCG capitation models are estimated by means of ordinary least squares, with an individual's annual healthcare expenditures in 1994 as the dependent variable. For subgroups based on health survey information, costs predicted by the models are compared with actual costs. Using stepwise regression procedures a subset of relevant survey variables that could improve the predictive accuracy of the three-year DCG model was identified. Capitation models were extended with these variables.
Data collection/extraction methods: For the empirical analysis, panel data of sickness fund members were used that contained demographic information, annual healthcare expenditures, and diagnostic information from hospitalizations for each member. In 1993, a mailed health survey was conducted among a random sample of 15,000 persons in the panel data set, with a 70 percent response rate.
Principal findings: The predictive accuracy of the demographic model improves when it is extended with diagnostic information from prior hospitalizations (DCGs). A subset of survey variables further improves the predictive accuracy of the DCG capitation models. The predictable profits and losses based on survey information for the DCG models are smaller than for the demographic model. Most persons with predictable losses based on health survey information were not hospitalized in the preceding year.
Conclusions: The use of diagnostic information from prior hospitalizations is a promising option for improving the demographic capitation payment formula. This study suggests that diagnostic information from outpatient utilization is complementary to DCGs in predicting future costs.