Surgery Duration Prediction Using Multi-Task Feature Selection.

Area of Science:

• Healthcare Operations Research
• Medical Informatics
• Applied Machine Learning

Background:

• Optimizing operating room (OR) activity is complex for hospital managers.
• Traditional OR scheduling is insufficient; personalized medicine is needed.
• Accurate surgery duration prediction is crucial for OR efficiency.

Purpose of the Study:

• To introduce a scientific tool for predicting surgery durations and enhancing OR performance.
• To improve patient benefit and hospital efficiency through better OR management.
• To develop a method for personalized medicine in surgical scheduling.

Main Methods:

• Utilized multi-task regression for surgery duration prediction.
• Selected a common subset of predictive covariates across tasks.
• Allowed model coefficients to vary between regression tasks (surgeon, operation type, or interaction).

Main Results:

• The multi-task regression approach outperformed baseline models for surgeon-based and operation-type-and-surgeon-based tasks.
• The method accurately estimates surgery durations, aiding resource identification.
• Performance lagged behind baseline for operation-type-based tasks.

Conclusions:

• Accurate surgery duration estimation improves patient throughput and resource optimization.
• The proposed tool advances personalized medicine and operational efficiency in healthcare.
• This research provides a valuable method for dynamic OR management.