A Meta-analysis of Predicting Disorders of Consciousness After Traumatic Brain Injury by Machine Learning Models.

ABSTRACT

Objective: This study pursued a meta-analysis to evaluate the predictive accuracy of machine learning (ML) models in determining disorders of consciousness (DOC) among patients with traumatic brain injury (TBI). Methods: A comprehensive literature search was conducted to identify ML applications in the establishment of a predictive model of DOC after TBI as of August 6, 2023. Two independent reviewers assessed publication eligibility based on predefined criteria. The predictive accuracy was measured using areas under the receiver operating characteristic curves (AUCs). Subsequently, a random-effects model was employed to estimate the overall effect size, and statistical heterogeneity was determined based on I2 statistic. Additionally, funnel plot asymmetry was employed to examine publication bias. Finally, subgroup analyses were performed based on age, ML type, and relevant clinical outcomes. Results: Final analyses incorporated a total of 46 studies. Both the overall and subgroup analyses exhibited considerable statistical heterogeneity. Machine learning predictions for DOC in TBI yielded an overall pooled AUC of 0.83 (95% CI 0.82-0.84). Subgroup analysis based on age revealed that the ML model in pediatric patients yielded an overall combined AUC of 0.88 (95% CI 0.80-0.95); among the model subgroups, logistic regression was the most frequently employed, with an overall pooled AUC of 0.85 (95% CI 0.83-0.87). In the clinical outcome subgroup analysis, the overall pooled AUC for distinguishing between consciousness recovery and consciousness disorders was 0.84 (95% CI 0.82-0.85). Conclusion: The findings of this meta-analysis demonstrated outstanding accuracy of ML models in predicting DOC among patients with brain injuries, which presented substantial research value and potential of ML application in this domain.