Machine Learning Models to Predict Risk of Venous Thromboembolism in Patients undergoing Lumbar Spine Surgery

Introduction: Venous thromboembolism (VTE) is a significant complication in patients undergoing lumbar spine surgery, with risks influenced by various patient and procedural factors. Predictive models may assist in identifying high-risk patients and tailoring preventive measures.

Objective: To develop a machine learning model to predict VTE risk in patients undergoing lumbar spine surgeries, specifically decompression and fusion procedures, using patient data from the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP).

Methods: A retrospective cohort of patients from ACS-NSQIP between 2016 and 2022 undergoing lumbar decompression and fusion was analyzed. VTE risk prediction was performed using the XGBoost model, with patient demographics, comorbidities, lab values, and surgical details as input variables. To address class imbalance, undersampling was used to equalize positive and negative occurrences of VTE. Model performance was evaluated primarily by Area Under the Receiver Operating Characteristic Curve (AUC-ROC).

Results: The XGBoost model achieved an AUC-ROC of 0.757 for decompression and 0.702 for fusion procedures. For decompression, model precision, recall, and F1 scores were 0.714, 0.638, and 0.674, respectively, with an accuracy of 0.691 and specificity of 0.743.

Conclusion : The XGBoost model demonstrated moderate predictive performance in identifying VTE risk in lumbar spine surgery patients. Future work should aim to enhance predictive accuracy by exploring additional patient-specific and procedural features.