Improving preoperative prediction of surgery duration

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Operations Research

Background:

• Operating room (OR) inefficiencies lead to significant financial losses, staff dissatisfaction, and disrupted patient care.
• Accurate operation duration estimates are crucial for improving OR efficiency.
• Current methods for estimating surgery duration, such as surgeon estimates or averaging a few past cases, lack precision.

Purpose of the Study:

• To develop and evaluate machine learning (ML) approaches for more accurate prediction of operation duration.
• To provide a reliable alternative to current, less precise methods of estimating surgery length, especially when surgeon estimates are unavailable.

Main Methods:

• Utilized over four years of elective surgery records (n=52,171) from a major Australian hospital.
• Developed and trained predictive models using individual patient characteristics and historical surgery data.
• Compared various ML algorithms, including Extreme Gradient Boosting (XGBoost) and Random Forest (RF), for operation duration prediction.

Main Results:

• The XGBoost model demonstrated statistically significant reductions in prediction error compared to other ML models.
• The XGBoost model decreased total absolute error by 6854 minutes (approximately 114 hours) relative to existing hospital methods.
• The developed ML approach offers a more accurate prediction of operation duration.

Conclusions:

• ML methods show significant potential for improving the accuracy of operation duration estimations over current hospital practices.
• The proposed ML approach can be deployed using readily available features at the operating room scheduling stage.
• Accurate operation duration prediction using ML can enhance operating room efficiency and resource management.