A New Multimodal Neuroprognostic Model for Chronic Disorders of Consciousness: Integrating Behavioral, Hormonal, and Imaging Features

Previous studies have suggested that endocrine abnormalities following brain injury may influence the long-term recovery of patients with chronic disorders of consciousness (DOC). However, it remains unclear whether combining endocrine measurements with established behavioral and imaging metrics can further enhance DOC prognostication. To address this, we aim to develop a precise neuroprognostic model by integrating hormonal, behavioral, and resting-state fMRI (rs-fMRI) assessments. In this retrospective observational study, 43 patients with DOC were enrolled, each of whom was assessed using the Coma Recovery Scale-Revised (CRS-R), pituitary-related hormone levels, and rs-fMRI. Based on the Glasgow Outcome Scale (GOS), patients were classified into a favorable prognosis subgroup (GOS ≥ 3, n = 19) and a poor prognosis subgroup (GOS < 3, n = 24). We calculated two rs-fMRI features for each brain region: static functional connectivity and dynamic temporal stability. A Support Vector Machine classifier was then applied using these multimodal feature subsets to predict patient prognosis. Our multimodal model achieved a prediction accuracy of 0.91 (sensitivity = 0.84, specificity = 0.96) for DOC prognosis, outperforming control models that used fewer feature subsets, which had accuracy ranging from 0.58 to 0.84. Additionally, brain regions primarily from the frontoparietal networks contribute most to the prediction, along with motor function scores of the CRS-R and free triiodothyronine hormone levels. Our preliminary findings suggest that integrating multiple domains enhances the accuracy of DOC prognosis predictions. Our model shows promise as an accurate and convenient tool to aid clinical decision-making regarding DOC prognosis, though further external validation is needed.