Predictive Model Assessment for Efficient Healthcare Resource Scheduling and Decision Making Open Access
Downloadable ContentDownload PDF
Health systems and regulatory bodies continuously attempt to reduce costs and eliminate waste to improve the way healthcare is delivered. Quality of care has become the key principle driving change and healthcare outcomes. As the healthcare industry transitions to value-based reimbursement models from fee-for-service models, providers need pragmatic solutions to schedule resources adequately, primarily for events that involve multiple encounters. In this study, ordinary least squares, maximum likelihood, and supervised machine learning predictive models were compared to predict the length of time required for an inpatient encounter and the predicted values were used to determine whether patients can be flagged for readmission within 30 days of discharge. The evidence suggests that the maximum likelihood models outperform the ordinary least squares and supervised machine learning models. The maximum likelihood models better fit the data based on the Akaike information criterion, Bayesian information criterion, and root-mean-square error. The evidence further suggests that in addition to other predictors, the length of stay is significant in predicting the likelihood of 30-day readmittance. The logistic regression model is preferred to the KNN machine learning algorithm because of its ability to obtain classification probabilities. Further research opportunities include expanding the formulated models by incorporating other factors that determine the length of stay and the likelihood of patient readmittance.