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

Area of Science:

• Medical Informatics
• Operations Research
• Artificial Intelligence

Background:

• The COVID-19 pandemic significantly reduced endoscopic procedure volumes, creating substantial patient backlogs.
• Delays in endoscopic investigations impact patient care and require efficient solutions.

Purpose of the Study:

• To develop a machine learning-based scheduling tool for endoscopic procedures.
• To enhance resource utilization, system efficiency, and patient throughput.
• To mitigate procedural delays and address patient backlogs.

Main Methods:

• Phase 1: Applied machine learning (XGBoost regression) to predict procedure duration using patient and environmental data.
• Phase 2: Developed a scheduling module utilizing a greedy heuristic and Mixed Integer Programming (MIP) for resource optimization.

Main Results:

• XGBoost regression model achieved a mean absolute error of 5.67 minutes for duration prediction.
• MIP simulations increased patient scheduling by 5.9% and reduced mean waiting time from 19.5 to 17.3 days for 1,000 patients.
• Real-world data simulations showed MIP scheduling 8 more patients than the baseline, with improved resource utilization.

Conclusions:

• A machine learning scheduling tool shows significant potential for improving endoscopic procedure backlogs.
• Clinical validation is required to confirm the tool's effectiveness in real-world settings.
• Future work includes prospective data collection and integration into clinical workflows for practical utility.