PROJECT TITLE :

A Two-Stage Model to Predict Surgical Patients’ Lengths of Stay From an Electronic Patient Database

ABSTRACT:

Increasing healthcare expenses and a growing demand for services necessitate a more efficient use of healthcare resources. The care delivery process is less efficient when resource requirements are unpredictable. Our goal is to eliminate the uncertainty surrounding patients' resource needs, which we do by grouping patients into similar resource user groups. Using electronic medical records, we create a two-stage classification algorithm to classify patients into lower variability resource user categories in this research. Patients can be classified into reduced variability resource user groups using a variety of statistical methods. Because of its unique properties, classification and regression tree (CART) analysis is a better tool for assessing healthcare data. It can, for example, naturally manage the interaction between predictor variables, is nonparametric in nature, and is mostly unaffected by the dimensionality curse. We discovered that the CART analysis is also beneficial for identifying patient characteristics that can explain resource requirement fluctuation. Furthermore, we discovered that several factors, such as the primary prescribed procedure code, admission point, and operating surgeon, might account for up to 53.43 percent of the variation in patient durations of stay (LoS). Reducing the uncertainty in patients' LoS estimates allows us to better manage patient flow and, as a result, achieve higher throughput.