Machine Learning Predicts Mortality and Respiratory Failure in Patients Admitted With Rib Fractures.

INTRODUCTION

Rib fractures are common and associated with significant morbidity and mortality. Accurate prediction of adverse outcomes remains challenging, with existing clinical risk models demonstrating poor performance on external validation. This study applies machine learning to predict mortality and respiratory failure in trauma patients with rib fractures.

METHODS

Adult trauma patients with rib fractures were queried from the National Trauma Data Bank for the years 2021 and 2022. The primary outcomes were in-hospital mortality and respiratory failure, defined as the need for mechanical ventilation. Models were developed using the Light Gradient Boosting Machine and Extreme Gradient Boosting algorithms and incorporated only admission variables. Bayesian optimization fine-tuned hyperparameters, and model performance was evaluated using repeated stratified k-fold cross-validation. Performance metrics included accuracy, precision, recall, area under the receiver operating characteristic curve, and area under the precision-recall curve. SHapley Additive exPlanations and partial dependence plots were used to interpret model predictions.

RESULTS

Overall, 260,771 patients were admitted with rib fractures over the study period. Of these, 3.4% died and 10.1% required mechanical ventilation. The models demonstrated strong predictive performance, with area under the receiver operating characteristic curve values of 0.90 and 0.87 for the Light Gradient Boosting Machine model predicting mortality and respiratory failure respectively. Age, severity of hemorrhage, degree of pulmonary dysfunction, and neurologic status were most influential in predicting adverse outcomes.

CONCLUSIONS

Machine-learning models accurately predict mortality and respiratory failure in trauma patients with rib fractures. These models offer a sophisticated, data-driven approach to risk stratification. Future work should focus on enhancing interpretability to better facilitate clinical adoption.