Spinal Cord Injury AIS Predictions Using Machine Learning.

Area of Science:

• Spinal Cord Injury Research
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Spinal cord injury (SCI) poses significant challenges in predicting patient recovery.
• Accurate prediction of outcomes is crucial for patient management and rehabilitation planning.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the American Spinal Injury Association Impairment Scale (AIS) scores at hospital discharge.
• To identify critical features, including AIS scores and demographic data, that predict SCI patient recovery.

Main Methods:

• Utilized the National Spinal Cord Injury Statistical Center (NSCISC) database (1972-2016) with 20,790 patients.
• Processed 417 features down to 53 for machine learning models.
• Tuned eight machine learning models, including Ridge Classifier, and employed Shapely analysis for feature importance.

Main Results:

• Ridge Classifier achieved the highest test set accuracy of 73.6% in predicting AIS scores.
• Admission AIS scores and neurologic category were the strongest predictors of recovery.
• Age, sex, marital status, and race were identified as important demographic predictors.

Conclusions:

• Machine learning models, particularly Ridge Classifier, show promise in predicting SCI recovery.
• Admission AIS scores combined with demographic data offer a robust approach to outcome prediction.
• Shapely analysis provided valuable insights into feature importance for SCI recovery.