Efficient operating room planning using an ensemble learning approach to predict surgery cancellations

Abstract

Cancelled surgeries cause insufficient use of operating rooms, surgeons, nurses, equipment and other hospital resources. Operating rooms are one of the most expensive resources in any healthcare system. The average amount of cost for any operating room for a less complex procedure is around $29/min, while the more complex procedures cost up to $80/min. Many studies have been conducted to analyze and understand the reasons behind surgery cancellation; however, a handful of studies aimed to predict which patients have a large risk of cancellation. We used four different traditional data mining techniques—Conditional Inference Tree, C5.0, logistic regression, and Radial Basis Function Kernel Support Vector Machine—to predict surgical cancellations to create efficient surgical patients’ schedules. Then, a stacking generalization ensemble machine was developed and compared to the traditional methods. Three different scheduling scenarios were developed and tested using discrete event simulation (DES). The stacking generalization ensemble machine outperformed all traditional data mining techniques and the benchmark algorithm with an accuracy of 95% and area under the curve (AUC) of 96%. The proposed simulation-based optimization scheduling technique increased the operating room use by 13%. Surgeries cancellation prediction leads to efficient scheduling that ultimately leads to reductions in expenditure.

Highlights:

• Created a robust framework to increase OR utilization

• Predicted surgery cancellation using data mining approaches

• Created a novel stacking ensemble machine

• Enhanced the prediction power through ensemble learning that outperformed the other algorithms

• Determined overbooking opportunities through Simulation-based optimization

Keywords:

• Cancellations
• data mining
• ensemble learning
• operating rooms cancellation
• surgeries