(I-562) Development of a Personalized Machine Learning-based Prediction Model for Short Term Postoperative Outcomes in Patients Undergoing Posterior Cervical Fusion
Clinical Research Coordinator Mount Sinai Health System New York, New York, United States
Introduction: By predicting short-term postoperative outcomes before surgery, patients who undergo posterior cervical fusion (PCF) surgery may benefit from more precise patient care plans that reduce the likelihood of unfavorable outcomes. Based on our literature search, no study has explored the ability of ML algorithms to predict prolonged length of stay (LOS), non-home discharges, and readmissions in a single study following PCF surgery. We developed machine learning models for predicting short-term postoperative outcomes with additional explainability.
Methods: The American College of Surgeons National Surgical Quality Improvement Program database was used to identify patients who underwent PCF surgery. Prolonged length of stay, non-home discharges, and readmissions were the three outcomes that were investigated. These outcomes were predicted using four supervised machine learning algorithms: XGBoost, LightGBM, CatBoost, and Random Forest. We used Shapley additive explanations (SHAP) to assess the relative importance of predictor variables, in addition to performance plots and metrics.
Results: A total of 6277 patients that underwent PCF surgery were included in the analysis. The machine learning algorithms were able to accurately predict between 74.6% and 78.5% of the patients who had prolonged LOS with AUROC values between 0.746 and 0.759; between 80.0% and 82.8% of the patients who had non-home discharges with AUROC values between 0.804 to 0.819; and between 93.6% and 94.4% of the patients who had readmissions with AUROC values between 0.720 and 0.724 in the test set. The most important features of the best algorithms for each outcome were as follows: preoperative hematocrit for prolonged LOS, age for non-home discharges, and preoperative sodium for readmissions.
Conclusion : Machine learning techniques have a significant potential for predicting postoperative outcomes following PCF surgery. The development of predictive models as clinically useful decision-making tools may significantly improve risk assessment and prognosis as the amount of data in spinal surgery keeps growing. Here, we present predictive models for PCF surgery that are meant to accomplish the aforementioned goals.
