Multifactorial Biomarkers for "Talk and Deteriorate" after Head Trauma Identified Using Machine Learning

Talk and Deteriorate refers to a clinical course where a patient is able to speak immediately after a traumatic brain injury but subsequently deteriorates in consciousness. Talk and Deteriorate outcomes are poor, and reliable prediction may help improve them. We searched for multifactorial biomarkers-combinations of multiple risk factors-assuming they would predict Talk and Deteriorate more reliably than single factors. A total of 680 consecutive patients aged ≥16 years with intracranial traumatic lesions and a Glasgow Coma Scale of 13-15 upon admission were included. Information on 27 factors at admission and whether Talk and Deteriorate developed was collected. The correlation between Talk and Deteriorate incidence and each factor was evaluated using univariate analysis. Multifactorial biomarkers associated with a high Talk and Deteriorate incidence were identified using a brute force method combined with clustering based on combinations of factors and various thresholds for each. Among the 680 patients, 89 (13.1%) developed Talk and Deteriorate. A total of 94 multifactorial biomarkers with a Talk and Deteriorate incidence rate of ≥50.0% were identified. These multifactorial biomarkers, which included factors previously suspected to be related to Talk and Deteriorate, were associated with a higher Talk and Deteriorate incidence than each factor alone. Thus, this study indicates that predicting Talk and Deteriorate is more reliable when based on multifactorial biomarkers than on single factors. Factors not previously reported as independent biomarkers contributed as components of multifactorial biomarkers, while previously reported factors played more important roles within them. Thus, this study revealed multifactorial contributions of factors related to Talk and Deteriorate. Future studies focusing on these biomarkers, incorporating pathophysiologic insights, will improve Talk and Deteriorate outcomes.