Respiratory Variability in Disorders of Consciousness: Relationship and Clinical Applications.

BACKGROUND

Respiratory variability (RV) reflects the dynamic modulation of breathing patterns by the central nervous system and may serve as a physiological marker of consciousness. However, its predictive value in disorders of consciousness (DOC) remains unclear.

OBJECTIVE

To investigate the relationship between RV and the level of consciousness, and to evaluate the potential of RV-based analysis for predicting clinical outcomes in DOC patients.

METHODS

Patients with disorders of consciousness and healthy controls were assessed using inertial measurement unit (IMU) sensors to record triaxial acceleration signals. RV indicators were extracted from respiratory waveforms. A generalized additive model (GAM) was applied to adjust for confounding variables. Group differences were evaluated using Bootstrap resampling and the Mann-Whitney U test. Machine learning models-including random forest, elastic net, support vector machine, and partial least squares regression-were employed to predict Coma Recovery Scale-Revised (CRS-R) scores and clinical outcomes.

RESULTS

Significant differences in multiple RV indicators were observed between patient and control groups (p < 0.05), indicating an association between RV and the pathophysiology of consciousness disorders. Among the tested models, the random forest algorithm achieved the best predictive performance for CRS-R scores (mean squared error = 3.76, R2 = 0.76) and for clinical outcomes (AUC = 0.74, sensitivity = 0.86), outperforming other models.

CONCLUSIONS

Respiratory variability, particularly when analyzed via random forest modeling, shows strong potential for prognostic assessment and clinical decision support in disorders of consciousness. RV may represent a non-invasive biomarker reflecting the neural control of respiration and consciousness state.