Surgeon behaviors compared to predictive modeling to reduce surgical readmissions
STUDY DESIGN: Patient data from a tertiary academic medical center’s surgical services were analyzed using a time-to-event model to identify criteria that statistically explain the timing of discharges. The data were then used to develop a time-varying prediction model of unplanned hospital readmissions. The readmissions model was validated via a series of in-sample and out-of-sample tests using bootstrapped, partitioned patients. Results from the discharge and readmission models were then compared to each other.
RESULTS: The predictive discharge and readmission regression models were generated from a database of 10,272 patients totaling 62,204 patient-days with 850 readmissions (8.27%). 13 daily clinical measures were found to be significant in both regression models. The readmission model demonstrated moderate discrimination with all patient partitions [mean in-sample C statistic = 0.806 (95%CI=0.805-0.807); out-of-sample mean C statistics = 0.780 (95%CI=0.778-0.781)]. Comparison of discharge behaviors versus the predictive readmission model suggested important discordance with certain clinical measures (e.g., patient demographics, day of discharge laboratory values) not being accounted for to optimize readmissions.
CONCLUSIONS: Predictive models suggest considerable discordance exists between how surgeons behave in practice and what may optimize discharge decision-making. The real-world application of such a predictive model is a question in itself. Behavioral experiments are necessary to ascertain whether medical decision makers will use the decision support software. If so, is the decision support tool effective in reducing readmissions? These further issues will need to be addressed through ongoing research.