Investigating the impact of outpatient services on length of stay: an easily interpretable approach

Limiting hospital length of stay (LOS) can prevent patient complications and reduce costs. The identification of activities that could impact LOS, and which may be attributable to outpatient services (OSs - i.e., diagnostic exams or specialist consultations provided by external wards), can be very useful for hospital administrators, and help optimize healthcare processes. In this work, we introduce TEXLoS, a tool able to study the association of OS activities with LOS. The problem is afforded in a Business Process Management perspective, allowing the integration and the adoption of state of the art techniques for trace classification and process model discovery. The focus of TEXLoS output is on explainability, which is the key to the adoption of the tool in practice. In the paper, we present the main steps of the TEXLoS architecture, and the results of its application to the data of Azienda Ospedaliera SS. Antonio e Biagio e Cesare Arrigo in Alessandria, Italy, where a balanced accuracy of 93% has been reached, along with a Matthews Correlation Coefficient of 0.68, confirming the high performance in classification. Interpretability of the provided output has also been successfully validated by a group of end user.