Improved Prediction of Procedure Duration for Elective Surgery.

Area of Science:

• Medical Informatics
• Health Services Research
• Surgical Workflow Optimization

Background:

• Accurate estimation of surgery duration is critical for operating theatre management and elective patient scheduling.
• Existing prediction algorithms require further analysis at a specialty level for performance insights.

Purpose of the Study:

• To analyze the performance of existing surgery duration prediction algorithms at a specialty level.
• To evaluate algorithm performance after data filtering to remove unreliable data.
• To develop and validate novel ensemble approaches for improved prediction accuracy.

Main Methods:

• Performance analysis of prior surgery duration prediction algorithms.
• Data filtering techniques to exclude unreliable data points.
• Development and validation of new ensemble machine learning models.

Main Results:

• Filtering unreliable data reduced Mean Absolute Percentage Error (MAPE) by 44% (from 0.68 to 0.38) with the Random Forests algorithm.
• The novel ensemble approach achieved a MAPE of 0.31, a 55% reduction from the original error.
• The ensemble approach improved prediction accuracy by 18% compared to Random Forests on filtered data.

Conclusions:

• Data filtering and ensemble methods significantly enhance the accuracy of surgery duration prediction.
• Optimized prediction models can lead to more efficient operating theatre utilization and scheduling.
• Further research into specialty-level algorithm performance and ensemble techniques is warranted.