A prediction model for postoperative respiratory failure in patients with traumatic cervical spinal cord injury using interpretable machine learning.

purposeRespiratory failure (RF) has been the primary cause of morbidity and mortality following trauma in patients with traumatic cervical spinal cord injury (TCSCI), often requiring assisted ventilation. This study aims to utilize interpretable machine learning (IML) models to predict the risk of postoperative respiratory failure in patients with TCSCI.

methodsA retrospective analysis was conducted on the clinical data of patients with TCSCI in our hospital from 2010 to 2021. The LASSO regression was used for feature selection through ten-fold cross-validation. Six machine learning (ML) models were developed, and their performance was evaluated based on metrics such as discrimination, sensitivity, specificity, accuracy, and F1 score. The Shapley additive explanation (SHAP) algorithm was employed to assess the impact of different clinical features on the optimal model, enhancing the interpretability of the model.

resultsA total of 19.4% of patients (84/432) developed respiratory failure (RF). The LASSO regression algorithm ultimately selected five variables for model development. Among the six models, the XGBoost model demonstrated the best predictive performance based on metrics such as discrimination, sensitivity, specificity, accuracy, and F1 score. The SHAP algorithm analysis showed that the factors with the highest to lowest impact on the output of the XGBoost model were ASIA, prognosis of nutrition index (PNI), cervical fracture-dislocation, neurological injury segment, and severe trauma.

conclusionInterpretable machine learning (IML) models can serve as reliable tools for predicting postoperative RF. The XGBoost model has the best predictive performance, which is helpful for clinicians to identify high-risk patients and implement early intervention in clinical decision-making.