CDA

CDA predicts short-term adverse postoperative outcomes after cervical disc arthroplasty using machine learning models trained on ACS National Surgical Quality Improvement Program (NSQIP) data to enable individualized risk assessment.

Key Features:

• Machine Learning Algorithms: Four distinct machine learning algorithms were used to train predictive models for postoperative outcomes.
• Outcomes Predicted: The models predict prolonged hospital stays, major complications, nonhome discharges, and 30-day readmissions.
• Data Source: Models were developed using the American College of Surgeons (ACS) NSQIP database comprising 6,604 patients who underwent cervical disc arthroplasty (CDA).
• Performance Metrics: The models achieved a mean area under the receiver operating characteristic curve (AUROC) of 0.814 and an accuracy of 87.8%.
• Interpretability: SHapley Additive exPlanations (SHAP) analyses identified key predictors, with white race reported as a significant feature across all algorithms.

Scientific Applications:

• Individualized Risk Assessment: Provide patient-level predictions of short-term adverse outcomes following cervical disc arthroplasty to inform clinical risk stratification.
• Postoperative Planning: Inform postoperative care planning and resource allocation by identifying patients at higher risk for prolonged stay, complications, nonhome discharge, or readmission.

Methodology:

Models were trained on historical ACS NSQIP data for 6,604 CDA patients using four machine learning algorithms, and SHAP values were computed to identify influential predictor variables.