Continuous real-time prediction of surgical case duration using a modular artificial neural network

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Surgical Workflow Optimization

Background:

• Accurate surgical duration prediction is crucial for perioperative decision-making and cost reduction.
• Current methods lack real-time adaptability, impacting efficiency.
• Machine learning offers a potential solution for dynamic surgical time forecasting.

Purpose of the Study:

• To develop and evaluate a machine learning model for real-time surgical duration prediction.
• To compare the model's performance against traditional methods and scheduled durations.
• To assess the clinical utility in identifying potential overtime situations.

Main Methods:

• Retrieved preoperative and intraoperative data from 70,826 surgeries (March 2019 - October 2019).
• Developed a modular artificial neural network (ANN) for continuous forecasting.
• Compared ANN performance using Continuous Ranked Probability Score (CRPS) against a Bayesian approach and scheduled duration.

Main Results:

• The modular ANN demonstrated the lowest time error (CRPS: mean=13.8 min), outperforming the Bayesian approach (mean difference=6.4 min).
• ANN achieved 89% accuracy in predicting cases exceeding 15:00, significantly higher than Bayesian (80%) and scheduled (78%) methods.
• The model provided superior real-time information for perioperative decision support.

Conclusions:

• A real-time neural network model effectively predicts surgical duration using integrated data.
• This AI-driven approach offers significant improvements over existing methods for accuracy and efficiency.
• The model presents opportunities to reduce surgical costs and optimize operating room resource allocation.