P30. Development of a machine learning algorithm for prediction of complications and readmission after lumbar spinal fusion

Abstract

<h3>BACKGROUND CONTEXT</h3> Lumbar spinal fusion represents a large and growing fraction of the health care system, with a 15-fold increase in cases since 2002. The average age and number of medical comorbidities of patients undergoing lumbar fusion has similarly increased. Accurately risk-stratifying patients who undergo lumbar fusion would be of great utility, given the significant cost and morbidity associated with developing major perioperative complications. There is a paucity of accurate and validated prediction models that can be used to preoperatively risk-stratify patients for lumbar fusion. Advanced machine learning (ML) methods have grown in popularity across numerous medical disciplines due to their ability to recognize complex, non-linear relationships. <h3>PURPOSE</h3> We aim to develop an optimized ML algorithm for prediction of major perioperative complication after lumbar fusion. Our secondary aim is to compare its performance against logistic regression (LR). <h3>STUDY DESIGN/SETTING</h3> Retrospective, cohort study. <h3>PATIENT SAMPLE</h3> Patients 18 years or older at a non-federal California hospital who underwent lumbar spinal fusion. <h3>OUTCOME MEASURES</h3> Readmission within 30 days, major perioperative complications (readmission within 30 days, venous thromboembolism within 30 days, myocardial infarction within 7 days, pneumonia within 7 days, systemic infection within 7 days, surgical site bleeding within 90 days, and wound complications within 90 days). <h3>METHODS</h3> This is a retrospective cohort study of adult patients who underwent lumbar spinal fusion at any California hospital between 2015-2017. We build LR and ML models that span different classes of modeling approaches. Discrimination and calibration were assessed using area under the receiver operating characteristic curve (AUROC) and Brier score, respectively. We ranked the contribution of the included variables to model performance. <h3>RESULTS</h3> A total of 38,788 patients met inclusion criteria for this study. There were 4,470 major complications (11.5%). The XGBoost algorithm demonstrates the highest discrimination (AUROC: 0.687 + 0.01) compared to LR (0.675 + 0.01), outperforming the three other ML models. This model was well calibrated with a Brier score of 0.094 + 0.001. The variables most important to XGBoost performance include angina pectoris, metastatic cancer, teaching hospital status, history of concussion, number of comorbidities, and workers' compensation insurance. Teaching hospital status and concussion history were not found to be among the most important features in the LR model. <h3>CONCLUSIONS</h3> Major perioperative complications and unplanned readmissions after lumbar fusion are a source of significant cost and morbidity. Wereport an optimized ML algorithm for prediction of major perioperative complications and 30-day readmission after lumbar spinal fusion. The predictors most important for XGBoost differ from those for LR, suggesting that the superior performance of XGBoost for prediction of major complications in this dataset is due to the ability of advanced ML methods to capture relationships between variables that LR is unable to detect. By providing accurate prognostic information, this algorithm may facilitate preoperative shared decision-making and aid with appropriate patient selection. A physician must be able to provide patients with comprehensive and accurate information regarding risks and benefits in order for the patient to provide true informed consent. This tool may identify and address potentially modifiable risk factors, helping to accurately risk-stratify patients and decrease likelihood of major complications. <h3>FDA DEVICE/DRUG STATUS</h3> This abstract does not discuss or include any applicable devices or drugs.