Abstract
Background/Aims The Operating Room (OR) is a large source of revenue and one of the most costly departments in a hospital. Scheduling surgeries into an OR is complicated by the inherent uncertainty associated with each surgery. The case length of a surgery at Geisinger is predicted using a moving average of the 10 previous procedures performed by a given surgeon. A process capability analysis was performed to gauge the ability of each surgical procedure to be within ± 15 minutes of scheduled time. This analysis demonstrated a low process capability across all surgeries. This research aims to create a process to better predict the surgical case length by leveraging the Electronic Health Records, which can enable more efficient scheduling and use of the ORs.
Methods Based on a literature review and the results of an internally conducted survey of OR staff, a dataset was constructed with 135 predictors. A test dataset was randomly separated from the training dataset for validation. Predictive models were developed using Stepwise Linear Regression (LR) and Artificial Neural Networks (ANNs). Multilayer Perceptron ANNs with 2 hidden layers using a sigmoid transfer function and Delta Bar Delta learning algorithm tended to perform the best. The final model contains the 39 most sensitive predictor variables from the ANN model and the LR model.
Results In all cases, the predictive models significantly improved the case duration accuracy. The ANN models outperformed the LR models on 3 of the 5 high-volume procedures. The greatest improvement over the baseline occurred for the ANN model for Arthroplasty Total Hip, where case duration accuracy improved from 32.1% (80 of 249 test cases) to 59.0% (147 of 249 test cases) for an improvement of 83.7% (improving to 59.0% from 32.1%).
Conclusions The ANN and LR models can be used to significantly enhance the predictability of surgical schedules. Even with the significantly enhanced predictability of surgical case lengths, the 5 investigated high-volume surgeries are still not necessarily process capable with respect to ± 15 minute specification limit.
