P19. Development and validation of machine learning algorithms for predicting adverse events following surgery for metastatic spine tumors: metastatic adverse events scoring system (MAES).

Abstract

BACKGROUND CONTEXT Preoperative prognostication of the adverse events (AEs) following surgery for metastatic spine tumors can enhance the risk stratification, thus can guide in implementing targeted treatment to minimize these events. PURPOSE To develop and validate a set of predictive variables of the AEs following surgery for metastatic spine tumors. STUDY DESIGN/SETTING The patient cohort was identified from the American College of Surgeons, National Surgical Quality Improvement Program (2005-2016). PATIENT SAMPLE The overall 30-d AEs of the 3,566 patients who underwent surgery for metastatic spine tumors were 8.7% (n=309). OUTCOME MEASURES Machine learning algorithm. METHODS We developed eight machine learning algorithms, and the algorithm with the best performance across discrimination, calibration and overall performance was used for predicting the overall risks of 30-day AEs. RESULTS The median age of our cohort was 60.0 years (range, 16-89 years).The model with 16-predictive factors which included: age, body mass index, tumor location, dural involvement, type of surgery, functional health status, chronic smoker, chronic obstructive pulmonary disorder, chronic steroid use, any bleeding disorders, preoperative serum creatinine, preoperative serum albumin, preoperative hematocrit, preoperative alkaline phosphatase, preoperative platelet count and preoperative white blood cell count performed best on the discrimination, calibration, Brier score and decision analysis to develop a machine learning algorithm. Logistic regression showed higher AUCs than least absolute shrinkage and selection operator across these different models. The predictive probability derived from the best model was uploaded on an open access web application which can be found at: https://spine.massgeneral.org/drupal/AdverseEvents-MetastaticSpineTumor. CONCLUSIONS Machine learning algorithms provide promising results for prediction of 30-day AEs following surgery for metastatic spine tumors. These algorithms can provide useful factors for patient-counselling, assessing perioperative risk factors, and predicting postoperative 30-day outcome after spine surgery. FDA DEVICE/DRUG STATUS This abstract does not discuss or include any applicable devices or drugs. Preoperative prognostication of the adverse events (AEs) following surgery for metastatic spine tumors can enhance the risk stratification, thus can guide in implementing targeted treatment to minimize these events. To develop and validate a set of predictive variables of the AEs following surgery for metastatic spine tumors. The patient cohort was identified from the American College of Surgeons, National Surgical Quality Improvement Program (2005-2016). The overall 30-d AEs of the 3,566 patients who underwent surgery for metastatic spine tumors were 8.7% (n=309). Machine learning algorithm. We developed eight machine learning algorithms, and the algorithm with the best performance across discrimination, calibration and overall performance was used for predicting the overall risks of 30-day AEs. The median age of our cohort was 60.0 years (range, 16-89 years).The model with 16-predictive factors which included: age, body mass index, tumor location, dural involvement, type of surgery, functional health status, chronic smoker, chronic obstructive pulmonary disorder, chronic steroid use, any bleeding disorders, preoperative serum creatinine, preoperative serum albumin, preoperative hematocrit, preoperative alkaline phosphatase, preoperative platelet count and preoperative white blood cell count performed best on the discrimination, calibration, Brier score and decision analysis to develop a machine learning algorithm. Logistic regression showed higher AUCs than least absolute shrinkage and selection operator across these different models. The predictive probability derived from the best model was uploaded on an open access web application which can be found at: https://spine.massgeneral.org/drupal/AdverseEvents-MetastaticSpineTumor. Machine learning algorithms provide promising results for prediction of 30-day AEs following surgery for metastatic spine tumors. These algorithms can provide useful factors for patient-counselling, assessing perioperative risk factors, and predicting postoperative 30-day outcome after spine surgery.