Machine learning for surgical time prediction

Operating Rooms (ORs) are among the most expensive services in hospitals. A challenge to optimize OR efficiency is improve surgery scheduling task, which requires estimation of surgical time duration. Surgeons or programming units (based on people's experience) typically do duration using an experience-based strategy, may include some bias, such as overestimating time, increasing ORs' operational cost. This paper analyzes a machine learning-based solution for predictions. We apply and compare four machine-learning algorithms (Linear Regression, Support Vector Machines, Regression Trees, Bagged Trees) predict at tertiary referral university hospital Bogotá, Colombia. Historical data from 2004 until 2019 was used train algorithms. Comparison given terms Root Mean Square Error (RMSE) predicted algorithms' computing time. The algorithm with best performance compared currently method. All ML error between 26 37 min. overall obtained Trees (26 min RMSE, 3.16 training 0.49 testing time) when subset DB nine specialties containing 80% surgeries. also outperformed method lower RMSE; however, it shifted predominant overestimation underestimating surgeries' Different predicting duration, showing comparing their performance. showed RMSE Depending initial data, method, but future work necessary suit it, like any other algorithm, hospitals' needs.