Prevalence, risk factors, prediction of robust callus formation and accelerated fracture healing in traumatic brain injury patients: a five-year study

Background

Traumatic brain injury (TBI) usually induces robust callus formation at early stage and then subsequent acceleration of fracture union, as supported by both clinical and preclinical studies. However, risk factors and predictive tools to identify TBI patients most likely to experience this accelerated healing response are lacking and subject to future development.This study aimed to study the prevalence, risk factors, and develop machine learning (ML) models to predict robust callus formation and healing acceleration of fractures in TBI patients.

Methods

Between January 2018 and 2023, patients sustaining concomitant TBI and diaphyseal fractures who were admitted into the First Affiliated Hospital of Shantou University Medical College were evaluated retrospectively. The TBI patients were categorized into robust callus formation group (RCF) and normal callus formation group (NCF) based on follow-up radiographic fracture callus index assessments. Risk factors for RCF occurrence were first identified using traditional univariate and multivariate regression model, and predictive models were developed using 12 ML models (including traditional logistic regression model). The performance and interpretations of ML models were evaluated using the area under the receiver operating characteristic curve (AUC) and Shapley Additive Explanations (SHAP).

Results

Of the 723 patients reviewed, 150 cases were enrolled for final analysis. The prevalence of robust callus formation was 40.67 % (61/150) with significantly wider callus index (2.01 ± 0.61 vs 1.17 ± 0.12, P < 0.001) and acceleration in time to initial callus formation (22.92 ± 11.98 days vs 90.18 ± 34.52 days, P < 0.001). Brain contusions (OR 5.914, 95 % CI:2.479–14.108,P < 0.001), greater TBI severity levels evaluated using Glasgow Coma Scale (GCS, OR 3.074, 95 % CI:1.149–8.222,P = 0.025) and Marshall CT classifications (OR 2.845, 95 %CI:1.095–7.390,P = 0.032) were identified as independent risk factors for RCF occurrence. The gradient boosting decision tree (GBDT) algorithm demonstrated optimal predictive performance using TBI-specific variables, achieving an AUC of 0.86 ± 0.03. SHAP analysis revealed brain contusion, GCS scores, and Marshall CT classification scores as the three most influential clinical features.

Conclusions

For the first time, this study provided the prevalence and risk factors contributing to RCF occurrence in TBI patients with combined diaphyseal fractures, and also developed ML models for its prediction, for which it may optimize orthopedics treatment strategies and decision making in these unique set of TBI patients.

The translational potential of this article

The findings from this study offer crucial insights to enhance clinical decision-making and treatment approaches for managing fractures in TBI patients. Furthermore, our research establishes a groundwork for future investigations into the mechanisms linking TBI and enhanced osteogenesis, potentially aiding in addressing intricate bone regeneration obstacles like non-unions and critical-size defects.