An evaluation of a simple model for predicting surgery duration using a set of surgical procedure parameters

The predicted surgery duration is the main data for operating room scheduling. Existing studies on surgery duration prediction have mostly addressed a large set of predictors. However, the available data for predictors are limited or cannot be easily obtained. In practice, the patient's identity and the surgical procedure name are definitely available when the surgeon reserves the surgery schedule. Other detailed data will be available only after the clinical observation of the patient, which is conducted a few hours before the surgery. Furthermore, the variability in surgery duration contributes to the complexity of the operating room scheduling. This study evaluated a simple model to predict the duration of surgery. The model used fewer predictors, which were the surgical procedure parameters, and reduced the variability of the surgery duration numerical value. The parameters comprised a set of hospital parameters collected for the purpose of surgery billing, representing the surgery complexity and resources needed. Using the estimation algorithm, our results showed that a set of surgical procedure parameters as the sole predictors resulted in slightly better performance than combining them with patient features. To reduce the variability of the surgical duration numerical values, we used discretization to convert them into categorical values represented by time bins. We proposed a modified calculation of error and accuracy based on the mean absolute error (MAE) of the estimation algorithm to evaluate the classification algorithm for predicting surgery duration using categorical values. Our study indicated that the use of categorical values resulted in a performance equivalent to that obtained using numerical values. Our simple model could facilitate a hospital to develop a framework for predicting surgery duration using the limited data available for surgery billing. The impacts of operating room scheduling using predicted surgery duration categorical values on patient waiting time and resource utilization in the operating room will be considered in a further study. © 2021 The Authors