In a recent study published in the Surgery journal, researchers aimed to develop a machine learning model that could accurately predict the length of surgical cases across various services and locations. By utilizing limited data available at the time of case creation, accurate predictions could lead to improved scheduling efficiency and cost-effective utilization of resources in the operating room.
To construct the predictive model, the researchers employed a similarity cascade technique to analyze the complexity of cases and the impact of the operator on their duration. This information was then incorporated into a gradient-boosting machine learning model. The team further adjusted the model’s loss function to strike a balance between overestimating or underestimating the case length.
To facilitate widespread deployment and usage, a streamlined production process was implemented, allowing the model to be seamlessly employed throughout their school without encountering any obstacles.
Upon evaluating the model’s performance, the results from August to December 2022 demonstrated its superiority over traditional scheduler predictions. In a comprehensive analysis encompassing 33,815 surgery cases across outpatient and hospital platforms, the model successfully forecasted 11.2% fewer cases that were too short, 5.9% more cases that fell within 20% of the actual case length, and only 5.3% more cases that exceeded the projected duration.
The implementation of the model also benefitted schedulers, as it enabled them to estimate the length of 3.4% more cases accurately within a 20% margin of the actual duration, while minimizing the occurrence of cases that extended beyond the forecasted time by 4.3%.
In conclusion, the researchers developed a unique framework that has proven dependable in predicting surgical case lengths at the time of their posting. Furthermore, this framework could serve as a foundation for future machine learning models aimed at enhancing surgical scheduling and resource allocation in the healthcare industry.
The source of the article is from the blog krama.net