181 Comparing Machine Learning Models for Predicting Operative Time in Neurosurgical Procedures: Institutional Versus Nationwide Dataset Approaches

INTRODUCTION: The estimation of operative time for a planned surgery is usually based on the experience of the surgical team or the retrospective studies relevant to the case. Efficient management of operative time and surgical theater utilization is important for hospitals to provide timely and cost-effective patient care. METHODS: We utilized two datasets: one from an institutional cohort and another from the ACS-NSQIP database, with each dataset comprised details of patient and neurosurgical procedures. Due to positively skewed operative time distributions, 5% of the upper end in both datasets were removed as outliers. We applied several ML models, including Linear Regression (LR), Support Vector Regression (SVR), Deep Neural Network (DNN), and Extreme Gradient Boosting Regression (XGBR), and conducted extensive hyperparameter tuning. Model performance was evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R2 metrics. RESULTS: The models developed from the institutional dataset generally provided better predictions (DNN RMSE: 53.44, MAE: 38.33, R2: 0.74; XGBR RMSE: 52.24, MAE: 37.25, R2: 0.76) compared to those from the ACS-NSQIP dataset (XGBR RMSE: 61.98, MAE: 47.90, R2: 0.38). The disparity in model performance highlights the influence of dataset characteristics on predictive accuracy. The XGBR model consistently showed the best performance across both datasets, suggesting its robustness in handling diverse and complex data features. CONCLUSIONS: Our findings indicate that while ML models can significantly aid in predicting operative times, the choice of dataset has a profound impact on their effectiveness. Institutional data, with potentially more clinical granularity, tends to enable more accurate predictions. These insights support the hypothesis that institutional-specific models might be more effective for OR scheduling optimizations.