An explainable multimodal machine learning model for diagnosing disorders of consciousness: evidence from a large multicenter Chinese cohort

BACKGROUND: Accurate assessment of consciousness in patients with disorders of consciousness (DoC) remains a major clinical challenge in both specialized neurorehabilitation and acute care settings, where behavioral evaluations alone can lead to misdiagnosis rates of up to 40%. This study aimed to develop and validate an explainable multimodal machine learning (ML) model to improve diagnostic precision in DoC. METHODS: Leveraging one of the largest multicenter cohorts of patients with DoC in China to date (n = 583), together with an independent external validation set (n = 70), we examined whether routinely collected clinical indicators together with EEG features could improve diagnostic reliability. Using LASSO-based feature selection, we developed three predictive ML models (clinical, EEG, and a combined clinical-EEG model). Nine ML algorithms were tested, with hyperparameters optimized by grid search and cross-validation. Model performance was assessed using ROC curves, calibration plots, and decision curve analysis (DCA). Model interpretability was examined using SHapley Additive exPlanations (SHAP). RESULTS: The combined clinical-EEG model (Model 3) consistently outperformed single-modality models. The final logistic regression-based Model 3 achieved an AUC of 0.77 (95% CI: 0.68–0.85) in the training set, 0.76 (95% CI: 0.61–0.92) in the internal validation set, and 0.75 (95% CI: 0.64–0.87) in the external validation set. SHAP analysis identified eight key features that contributed most to classification, providing biologically and clinically interpretable insights. To support clinical translation and time-sensitive decision-making, we developed a web-based tool enabling automated EEG feature extraction and individualized diagnostic predictions. CONCLUSIONS: This study developed and validated an explainable ML model integrating clinical and EEG features to classify consciousness levels in DoC, based on the large-multicenter DoC cohorts in China. Taken together, these findings suggest that a transparent multimodal framework can strengthen the clinical evaluation of consciousness and may offer a practical template for improving assessment in complex care environments. TRAIL REGISTRATION: ClinicalTrials.gov: NCT05949528. Registered 14 March 2022, https://clinicaltrials.gov/study/NCT05949528 .