Predicting the short term direct medical costs incurred by patients with rheumatoid arthritis.
OBJECTIVE: With increasing interest in revising the mechanisms of health care funding, the ability to anticipate patients' medical expenditures as well as to identify potentially modifiable predictors would be informative for health care providers, payers, and policy makers. METHODS: Eight hundred fifty-eight patients with rheumatoid arthritis from 2 Canadian centers reported semi-annually on their health services utilization and health status for up to 12 years. Annual direct costs were calculated using 1994 Canadian prices. Regression models for the variation in total direct costs and the individual resource components (i.e., physicians, tests, medications, acute and non-acute hospital care) were estimated using previous values of age, sex, disease duration, education, methotrexate availability, employment status, global well being, pain, duration of morning stiffness, and functional disability as predictor variables. The models were developed using all available data except the last 2 observations (i.e., data collected on the last 2 self-report questionnaires) from each patient, which were reserved for model validation. The predictive abilities of the models were assessed by comparing the most recent costs with those predicted by the model using values of the predictor variables from the previous time period. Further, to assess whether the models conferred any advantage over cost estimates based only on previous costs, most recent observed costs were also compared with costs observed in the preceding time period. RESULTS: Self-reported indices of either global well being, pain, or functional disability predicted total direct costs as well as the costs of the 5 individual resource components. Being younger, female, disabled from the work force, having shorter disease duration, and receiving more formal education also predicted higher costs in at least on health resource category. However, being older predicted higher acute and non-acute care hospital costs. Regression models incorporating longitudinal data did not perform better than average costs in the preceding time period in predicting future short term costs. CONCLUSION: Global well being, pain, functional disability, and previous costs are the most important predictors of short term direct medical costs. Although we have demonstrated that regression models do not perform better than previous costs in predicting future short term costs, previous costs are a much less informative predictor than health status variables. Variables such as functional disability and pain identify potentially modifiable disease features and suggest interventions that may improve patient well being and reduce costs.