Data Mining for Generalizable Pre-admission Prediction of Elective Length of Stay

An elective admission involves an hospital admission for which a doctor schedules the day and time of the visit in advance. That is, it is not an emergency or urgent admission. Factors in the scheduling of such admissions and also resource impacts include bed availability and utilization, and as such it is of benefit to be able to predict in advance the length of stay of patients. In this work we report on the use of data mining on a large, cross-provider dataset to develop potentially generalizable machine learning models to predict all- condition, high length of stay for elective admissions, prior to the time of admission based upon patient, admitting condition, hospital and locale attributes. The best of these models achieves an AUC score of 0.900. We also carry out evaluations to consider the generalizability of such models via measuring their predictive performance on a dataset from a different time period, and the models demonstrate that they maintain equivalent predictive levels on data from a different period.