Analysis examines the connection between neurological impairments and functional outcomes in spinal cord injury, suggesting a multi-class modeling approach.
Key Points
The study aims to model the relationship between neurological status and functional independence after spinal cord injury using machine learning techniques.
Compared ordinal and nominal classification models for predicting functional independence.
Analyzed motor and sensory scores alongside demographics (age, sex) from a multicenter study.
Evaluated model performance based on Spinal Cord Independence Measure tasks (grooming, bladder management, indoor mobility).
Stratified analyses into early, intermediate, and late post-injury phases.
Used Shapley Additive Explanations for model interpretability.
Model accuracy increased over time, with earlier phase results (46–71%) improving in the late phase (50–85%).
Random forest model demonstrated the highest performance on average (0.93 ± 0.03).
Ordinal models produced fewer severe misclassifications compared to nominal models.
Motor scores were better predictors of functional outcomes than sensory scores, indicating specific neurological functions linked to mobility and self-care.
The models provide a scalable approach for predicting functional independence from neurological assessments.