Reducing endoscopic procedure backlog by improving efficiency: a predictive model and machine learning-based scheduling approach.

Background: The COVID-19 pandemic led to a significant decrease in endoscopic procedure volumes, resulting in a backlog of patients awaiting investigation. Our study thus aimed to develop a machine learning-based scheduling tool to improve resource utilization, enhance system efficiency, and increase patient throughput, ultimately reducing procedural delays. Methods: In the first phase, machine learning methods were applied to historical data to predict procedure duration based on patient characteristics and environmental factors. In the second phase, a scheduling module was built using a greedy heuristic and a Mixed Integer Programming (MIP) model to optimize resource utilization. Results: We showed that among the tested models, an XGBoost regression model was selected with a mean absolute error of 5.67 minutes on the test set. The simulation results demonstrated that MIP increased the number of patients scheduled by 5.9% while reducing mean waiting time from 19.5 days to 17.3 days over a waiting list of 1,000 patients, evaluated within a 2-week period (10 working days). Simulations using real patient data showed that the MIP scheduled 8 more patients than the baseline. Numerical results confirmed higher resource utilization rates in adaptive schedules. Conclusions: Our study highlights the potential of a machine learning-based scheduling tool to enhance resource allocation, thus helping address backlogs in endoscopic procedures. Real-world clinical validation is now necessary to substantiate the tool's effectiveness. Future work should prioritize prospective data collection to refine the predictive model and seamlessly integrate the tool into clinical workflows, ensuring its practical utility and success.