Machine learning for surgical time prediction. (September 2021)

• Record Type: Journal Article
• Authors: Martinez, Oscar; Martinez, Carol; Parra, Carlos A.; Rugeles, Saul; Suarez, Daniel R.
• Is Part Of: Computer methods and programs in biomedicine. Volume 208(2021)
• Publication Date: 2021-09
• Subjects: Machine learning -- Surgical time prediction -- Linear regression -- Support vector machine -- Regression trees -- Assembly methods; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2021.106220
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: Our manuscript offer a comparison of different ML techniques to estimate the duration of a surgery using a large data set of surgery records (200.000 records during 14 years.), allowing not only comparing the techniques but also exploring the possible variables and factors influencing the surgery durations. While machine-learning methods can use the clinical records of an OR service, like manual programing, it is less susceptible to staff-personal biases, as it is the case of manual programing. Manual programming of OR services is still very common worldwide, and it usually overestimates the duration of the surgeries avoiding optimal OR scheduling. The methods presented in this study allow a convenient way to estimate the duration of individual surgeries directly form the records of the service. Those durations later can be used to schedule surgeries using Operation Research mythologies aiming at finding (sub-) optimal solutions. Usually those methodologies use statistics (average, mean, etc.) to estimate a "characteristic" duration of group of surgery, hampering their outcomes. Abstract: Background and Objective: Operating Rooms (ORs) are among the most expensive services in hospitals. A challenge to optimize the OR efficiency is to improve the surgery scheduling task, which requires the estimation of surgical time duration. Surgeons or programming units (based on people's experience) typically do the duration estimation using an experience-based strategy, which may include some bias, such as overestimating the surgery time, increasing ORs' operational cost. Methods: This paper analyzes a machine learning-based solution for surgical time predictions. We apply and compare four machine-learning algorithms (Linear Regression, Support Vector Machines, Regression Trees, and Bagged Trees) to predict the surgical time duration at a tertiary referral university hospital in Bogotá, Colombia. Historical data from 2004 until 2019 was used to train the algorithms. Comparison among algorithms was given in terms of the Root Mean Square Error (RMSE) of the predicted surgery duration and the algorithms' computing time. The algorithm with the best performance was compared to the currently used experience-based method. Results: All the ML algorithms predict the surgery duration with an error between 26 and 37 min. The best overall performance was obtained using Bagged Trees (26 min RMSE, 3.16 min training time, 0.49 min testing time) when using a subset of the DB with the nine specialties containing 80% of the surgeries. Bagged Trees also outperformed the experience-based method with a lower RMSE; however, it also shifted from a predominant overestimation to underestimating surgeries' duration. Conclusions: Different ML algorithms for predicting the surgical time duration, showing and comparing their performance. Bagged Trees showed the best performance in terms of RMSE and computing time. Depending on the initial data, Bagged Trees outperformed the experience-based method, but future work is necessary to suit it, like any other ML algorithm, to the hospitals' needs.