Predicting post-hospital discharge health care costs.

Abstract

Patients in this study represented an important group for health services research; their postdischarge costs were high, resulted from readmissions (especially for nonelective reasons), and were incurred by a minority of patients. Moreover, intensive interventions delivered by ambulatory care providers has the potential to reduce overall costs if patients at highest risk for readmission could be identified prospectively (Smith et al., 1988; Weinberger et al., 1988). We developed a model to predict post-hospital discharge costs in these patients using strategies to maximize their predictive capability. Although the model appeared to account for more variance in costs than currently available models in the derivation set, its performance in the validation set, albeit statistically significant, was disappointing. Because we considered a broad array of predictors, expanding the number of patient-oriented variables may not be fruitful. Instead, future research may need to consider more homogenous subgroups of patients in whom specific laboratory tests would have clinical significance; variance in providers' behaviors; and studies in health maintenance organizations, where control over resource utilization may make costs more predictable. Finally, empirically derived models must be tested in an independent sample. Without validating predictive models, the models' ability to predict health care costs may be overestimated.