BACKGROUND CONTEXT Emergency department (ED) and urgent care (UC) visits after spine surgery are costly and potentially avoidable health care encounters. By processing an increasingly vast array of digitized electronic medical record information using powerful machine learning (ML) methods, it may be possible to identify patient-specific features for complications that would otherwise remain unknown. To date, no study has leveraged ML to identify the strongest predictors of postoperative ED and UC visits after lumbar microdiscectomy. PURPOSE Identify risk factors for postoperative emergency department or urgent care visits after lumbar microdiscectomy. STUDY DESIGN/SETTING Retrospective cohort study at a multiple site, academic center. PATIENT SAMPLE Patients 18 years or older undergoing single-level lumbar microdiscectomy between September 2002 to December 2021 were included. OUTCOME MEASURES The primary outcome was at least one ED or urgent care UC visit within 90 days postoperatively. METHODS A total of 29 variables including patient demographics, medical comorbidities, history of spine surgery, preop drug usage and procedure characteristics were processed. Three machine learning algorithms – a balanced random forest (RF), a class-weighted L1-penalized logistic regression (LR), and a class-weighted gradient boosted tree (GBT) – were constructed and evaluated on the analysis data using an 80/20 train/test split and 5-fold cross validation for hyperparameter tuning. Performance on the test set was assessed via the Area Under the Receiver Operating Curve (AUC) statistic. The modeling process was repeated over 50 iterations and 95% confidence intervals for AUROC were constructed to account for stochasticity inherent to the model fitting as well as data splitting processes. Feature importance scores – mean decrease in impurity (MDI) for RF, regression coefficients for LR, and gain scores for GBM – were also computed across the 50 iterations, with emphasis given to the feature importance scores from the highest performing model. RESULTS A total of 5,679 patients were included, of which 367 (6.46%) had at least one ED or UC visit within 90 days. The RF, LR and GBM had AUC performance of 0.82 +/- 0.02, 0.82 +/- 0.02, and 0.74 +/- 0.00 respectively, indicating overall strong performance for each model. The top five most important features that were most predictive of ED/UC visits in the LR model included: BMI, use of intraoperative microscope OR duration, history of prior lumbar surgery and history of neurologic condition. The top five features least indicative of ED visits included: intraoperative durotomy, history of hepatic condition, preoperative opioids within 60 days of surgery, minimally invasive approach and history of dermatologic condition. CONCLUSIONS All three ML models exhibited strong performance for prediction of an ED/UC visit within 90 days of lumbar microdiscectomy, with BMI, use of intraoperative microscopy, OR duration and history of neurologic conditions and prior lumbar surgery being the most predictive. Identification of these risk factors may enable development of targeted patient-specific preoperative optimization. FDA DEVICE/DRUG STATUS This abstract does not discuss or include any applicable devices or drugs. Emergency department (ED) and urgent care (UC) visits after spine surgery are costly and potentially avoidable health care encounters. By processing an increasingly vast array of digitized electronic medical record information using powerful machine learning (ML) methods, it may be possible to identify patient-specific features for complications that would otherwise remain unknown. To date, no study has leveraged ML to identify the strongest predictors of postoperative ED and UC visits after lumbar microdiscectomy. Identify risk factors for postoperative emergency department or urgent care visits after lumbar microdiscectomy. Retrospective cohort study at a multiple site, academic center. Patients 18 years or older undergoing single-level lumbar microdiscectomy between September 2002 to December 2021 were included. The primary outcome was at least one ED or urgent care UC visit within 90 days postoperatively. A total of 29 variables including patient demographics, medical comorbidities, history of spine surgery, preop drug usage and procedure characteristics were processed. Three machine learning algorithms – a balanced random forest (RF), a class-weighted L1-penalized logistic regression (LR), and a class-weighted gradient boosted tree (GBT) – were constructed and evaluated on the analysis data using an 80/20 train/test split and 5-fold cross validation for hyperparameter tuning. Performance on the test set was assessed via the Area Under the Receiver Operating Curve (AUC) statistic. The modeling process was repeated over 50 iterations and 95% confidence intervals for AUROC were constructed to account for stochasticity inherent to the model fitting as well as data splitting processes. Feature importance scores – mean decrease in impurity (MDI) for RF, regression coefficients for LR, and gain scores for GBM – were also computed across the 50 iterations, with emphasis given to the feature importance scores from the highest performing model. A total of 5,679 patients were included, of which 367 (6.46%) had at least one ED or UC visit within 90 days. The RF, LR and GBM had AUC performance of 0.82 +/- 0.02, 0.82 +/- 0.02, and 0.74 +/- 0.00 respectively, indicating overall strong performance for each model. The top five most important features that were most predictive of ED/UC visits in the LR model included: BMI, use of intraoperative microscope OR duration, history of prior lumbar surgery and history of neurologic condition. The top five features least indicative of ED visits included: intraoperative durotomy, history of hepatic condition, preoperative opioids within 60 days of surgery, minimally invasive approach and history of dermatologic condition. All three ML models exhibited strong performance for prediction of an ED/UC visit within 90 days of lumbar microdiscectomy, with BMI, use of intraoperative microscopy, OR duration and history of neurologic conditions and prior lumbar surgery being the most predictive. Identification of these risk factors may enable development of targeted patient-specific preoperative optimization.