Machine Learning Prediction of Prevertebral Soft Tissue Swelling after Single-Level Anterior Cervical Surgery : A Proof-of-Concept Study

Abstract

OBJECTIVE: Prevertebral soft tissue swelling (PSTS) is a significant complication of anterior cervical spine surgery (ACSS) that causes dysphagia, dysphonia, and possibly life-threatening airway obstruction. This study aims to develop and internally validate an interpretable machine learning model to predict significant PSTS after single-level ACSS using a small, single-center dataset. METHODS: We retrospectively analyzed data from 62 patients who underwent single-level ACSS in our center, from January 2014 to December 2022. Postoperative swelling over 7.0 mm (above 75th percentile) was defined as significant PSTS. We developed an elastic net regularized logistic regression model over 1000 iterations with nested 5-fold cross validation for hyperparameter tuning and bootstrap validation. The model's interpretability was assessed using SHapley Additive exPlanations (SHAP) approach, and its clinical utility by decision curve analysis. RESULTS: A total of 16 (25.8%) out of 62 patients developed significant PSTS. Our model achieved a bootstrap-validated area under the ROC curve value of 0.84 (95% confidence interval, 0.73-0.92) with good calibration (Hosmer-Lemeshow p=0.42). According to our results, important predictors of PSTS included 1) low preoperative serum albumin (SHAP importance, 0.42), 2) upper level surgery above C5 (0.38), and 3) male sex (0.31). Decision curve analysis demonstrated net benefit across probability thresholds of 15-45%. CONCLUSION: Our machine learning model effectively predicted risk of significant PSTS following single-level ACSS despite small sample size. Such findings offer new opportunities for risk stratification and prevention strategies. This study is a proof of concept for artificial intelligence (AI)-based risk stratification strategy; further external validation could improve its relevance in the clinical setting.

Ähnliche Arbeiten