Development and Validation of Machine Learning Algorithms for Predicting Adverse Events After Surgery for Lumbar Degenerative Spondylolisthesis

Abstract

Preoperative prognostication of adverse events (AEs) for patients undergoing surgery for lumbar degenerative spondylolisthesis (LDS) can improve risk stratification and help guide the surgical decision-making process. The aim of this study was to develop and validate a set of predictive variables for 30-day AEs after surgery for LDS. The American College of Surgeons National Surgical Quality Improvement Program was used for this study (2005-2016). Logistic regression (enter, stepwise, and forward) and LASSO (least absolute shrinkage and selection operator) methods were performed to identify and select variables for analyses, which resulted in 26 potential models. The final model was selected based on clinical criteria and numeric results. The overall 30-day rate of AEs for 80,610 patients who underwent surgery for LDS in this database was 4.9% (n= 3965). The median age of the cohort was 58.0 years (range, 18-89 years). The model with the following 10-predictive factors (age, gender, American Society of Anesthesiologists grade, autogenous iliac bone graft, instrumented fusion, levels of surgery, surgical approach, functional status, preoperative serum albumin [g/dL] and serum alkaline phosphatase [IU/L]) performed well on the discrimination, calibration, Brier score, and decision analyses to develop machine learning algorithms. Logistic regression showed higher areas under the curve than did LASSO methods across the different models. The predictive probability derived from the best model is uploaded on an open-access Web application, which can be found at: https://spine.massgeneral.org/drupal/Lumbar-Degenerative-AdverseEvents. It is feasible to develop machine learning algorithms from large datasets to provide useful tools for patient counseling and surgical risk assessment.