Comprehensive Summary
This study evaluates the effectiveness of machine learning (ML) models in accurately predicting operative times for primary total joint arthroplasty (TJA), focusing on total knee arthroplasty (TKA) and total hip arthroplasty (THA). Researchers analyzed 22,769 cases (12,179 TKA and 10,590 THA) performed across a multi-center academic institution between 2017 and 2022. Four ML algorithms were tested: linear ridge regression, random forest, XGBoost, and explainable boosting machine. The XGBoost model demonstrated the highest accuracy, significantly outperforming traditiona scheduling methods. In TKA cases, the model reduced "excess time blocks" by 85 blocks (972 down to 887) and "wait time blocks" by 96 blocks (974 down to 878), saving 2,715 minutes (45.25 hours), or 181 blocks worth, of operating room (OR) time. For THA, "wait time blocks" decreased by 134 (976 down to 842), representing 2,070 minutes (34.5 hours) saved, though improvements in "excess time blocks" were not statistically significant. These results indicate that ML tools, particularly XGBoost, ca enhance scheduling accuracy, optimize OR utilization, and reduce wasted time and resources
Outcomes and Implications
This study underscores the significant role of machine learning in improving surgica scheduling efficiency and resource allocation. The XGBoost model saved a total o 4,785 minutes (79.75 hours) across TKA and THA cases, translating into substantial cost savings at an average OR cost of $36 per minute. For TKA, the model reduced overbooking by 8.7% and under booking by 9.8%, and for THA, under booking was reduced by 13.7%, highlighting its precision in time management. The most influentia predictor was the median case time of the previous 30 surgeries, with TKA median times showing a strong correlation to booking accuracy. BMI, ASA classification, an robotic assistance also emerged as significant predictors, with robotic-assisted TK increasing case duration by an average of 28 minutes and robotic THA by 18 minutes The data revealed additional insights, such as increased case times for patients with higher BMI and ASA classifications. For example, patients classified as ASA IV for required an average of 11.5 minutes more than ASA II patients, and BMI increases correlated with a 0.1-minute rise in case time per unit BMI. Furthermore, robotic procedures added considerable time, yet the model accurately integrated these factors into predictions, reducing scheduling errors