Length of Stay Outlier Detection through Cluster Analysis: A Case Study in Pediatrics

The increasing availability of detailed inpatient data is enabling the development of data-driven approaches to provide novel insights for the management of Length of Stay (LOS), an important quality metric in hospitals. This study examines clustering of inpatients using clinical and demographic attributes to identify LOS outliers and investigates the opportunity to reduce their LOS by comparing their order sequences with similar non-outliers in the same cluster. Learning from retrospective data on 353 pediatric inpatients admitted for appendectomy, we develop a two-stage procedure that first identifies a typical cluster with LOS outliers. Our second stage analysis compares orders pairwise to determine candidates for switching to make LOS outliers similar to non-outliers. Results indicate that switching orders in homogeneous inpatient sub-populations within the limits of clinical guidelines may be a promising decision support strategy for LOS management.