Incidence, risk factors, and machine learning prediction models of rib fractures in patients with traumatic thoracic vertebral fractures

Objective

This study aimed to comprehensively describe the clinical characteristics of rib fractures in patients with traumatic thoracic vertebral fractures (TVFs), and to develop machine learning (ML) models for predicting the risk of rib fractures.

Methods

We retrospectively reviewed patients diagnosed with TVFs at a single hospital between January 2007 and November 2024, enrolling 1420 patients and 20 variables. Chest CT scans were used to confirm the presence of rib fractures and to examine their distribution characteristics. Several ML models, including Support Vector Machine (SVM), XGBoost, Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Naive Bayes (NB), Neural Network (NN), and Ensemble Learning (EL), were applied. Model performance was evaluated using indicators such as area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), F1 score, density, discrimination slope, and a scoring system. Additionally, the prediction performance of the ML models was compared with that of three experienced clinicians.

Results

Rib fractures were identified in 222 patients (15.6 %), with a total of 1035 rib fractures recorded. Only 22.5 % were single rib fractures, and the distribution of unilateral and bilateral fractures was comparable (54.5 % vs. 45.5 %). Multivariate logistic regression revealed four significant predictors of rib fractures: gender (P = 0.004), cardiovascular disease (P = 0.003), trauma mechanism (P < 0.001), and the number of thoracic fractures (P < 0.001). Among all models, the EL model demonstrated the best predictive performance, achieving an accuracy of 0.920, F1 score of 0.767, sensitivity of 0.683, specificity of 0.977, PPV of 0.875, NPV of 0.928, and the highest overall score (48). Notably, its performance surpassed that of all three clinicians.

Conclusions

Rib fractures are relatively common in patients with TVFs and may be underdiagnosed, especially in the absence of clear symptoms. The EL model developed in this study offers strong predictive capability and may serve as a valuable clinical decision-support tool to identify high-risk patients and reduce the likelihood of missed diagnoses.