Neural Networks Accurately Predict Precise Metrics of Hospital Resource Utilization for Total Hip Arthroplasty: A Retrospective Database Study

AbstractO_ST_ABSBackgroundC_ST_ABSTotal hip and knee arthroplasties (THAs and TKAs) are some of the most common and successful surgeries. Predicting their duration of surgery (DOS) and length of stay (LOS) has massive implications for costs and resource management. The purpose of this study was to predict the DOS and LOS of THAs using machine learning models (MLMs) based on preoperative factors. MethodsThe American College of Surgeons (ACS) National Surgical and Quality Improvement (NSQIP) database was queried for elective unilateral THA procedures. Multiple MLMs were constructed to predict DOS and LOS. Models were evaluated according to mean squared error (MSE), buffer accuracy, and classification accuracy. To ensure useful predictions, the results of the models were compared to a mean regressor and previous MLM predictions for primary TKAs. Results196,942 patients were included. The neural network had the best MSE, buffer and training accuracies for both DOS and LOS. For DOS testing, the neural network MSE was 0.916, with the 30-minute buffer and [≤]120 min, >120 min accuracies being 75.4% and 88.5%. For LOS testing, the neural network MSE was 0.567, with the 1-day buffer and [≤]2 days, >2 accuracies being 70.3% and 80.9%. Slightly reduced performance was found for THA compared to TKA for DOS and LOS (3 to 5%), with similar important features identified. ConclusionMLMs based on preoperative factors successfully predicted the DOS and LOS of elective unilateral THAs, with similar performance to TKA. Future work should include operational factors to apply these models to real world resource optimization.