Predicting surgery durations in non-elective emergency operating rooms using machine learning techniques: A case study of La Pitié Salpêtrière hospital

Non-elective surgery patients are on the rise worldwide, generating mounting pressure on the already scarcely available resources (financial, human, and equipment). One of the main issues that has been identified in the literature is the high levels of unexpected occupancy of non-elective operating rooms (ORs) and the major scheduling problems generated by the lack of accuracy in the prediction of the surgery duration for emergency surgery patients. Therefore, accurately predicting surgery duration will positively impact healthcare performance metrics and allow hospitals to better schedule nonelective patients and manage ORs in a more efficient way. This paper aims to increase the efficiency and improve the patient flows in Emergency Operating Rooms (EOR) by predicting the surgery duration by applying Machine Learning (ML) algorithms. Using anonymized hospital data, we extract the patients' records admitted to EOR of La Pitié-Salpêtrière Hospital in Paris between January 2015 and December 2019. After a thorough data cleansing process, we implement and compare the accuracy of four standard machine learning algorithms: linear regression (LR), random forest (RF), support vector machine (SVM) for regression, and artificial neural networks (ANN). Machine Learning methods provide accurate predictions of surgery duration in La Pitié Salpêtrière Hospital. Particularly, ANN model is the most pertinent model for the available database; it is the most accurate for two specialties and gives good results for the other specialties. Then, it can be used in the optimization of EOR by reducing the length of stay (LOS) in the hospital.