Functional Outcome Prediction After Spinal Cord Injury Using Ensemble Machine Learning.

Area of Science:

• Neurology
• Rehabilitation Medicine
• Artificial Intelligence in Healthcare

Background:

• Spinal Cord Injury (SCI) presents significant challenges in predicting functional recovery.
• Accurate prognostication is crucial for tailoring rehabilitation strategies and managing patient expectations.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting functional outcomes in SCI patients.
• Utilize features available at the time of rehabilitation admission for early prognostication.

Main Methods:

• Retrospective single-center study involving 210 SCI patients.
• Collected demographic, injury, and functional data (SCIM, UEMS, LEMS).
• Engineered a meta-model combining Random Forest, SVM, Neural Network, and Gradient Boosting using ridge regression.

Main Results:

• The developed meta-model achieved high accuracy with RMSE of 9.7453, R² of 0.8835, and MAE of 7.4743 on the testing set.
• The meta-model outperformed individual base ML models in predicting functional outcomes.
• Key predictive features included age, acute length of stay, UEMS, LEMS, and SCIM subtotal scores.

Conclusions:

• Machine learning models can effectively predict functional outcomes post-SCI using readily available admission data.
• This approach enables early goal setting in rehabilitation programs.
• The model holds potential for evaluating the efficacy of advanced therapies like regenerative medicine.