Development of Prognostic Models for Bladder and Bowel Dysfunction in Traumatic Spinal Cord Injury Patients Using...

Area of Science:

• Neurology
• Rehabilitation Medicine
• Data Science

Background:

• Spinal cord injury (SCI) frequently leads to bladder and bowel dysfunction.
• Predictive models for neurological outcomes post-SCI are increasing, but models for bladder/bowel dysfunction are limited.

Purpose of the Study:

• To develop and validate machine learning models predicting bladder and bowel dysfunction outcomes in traumatic SCI patients.
• To integrate these predictive models into a user-friendly web application for clinical use.

Main Methods:

• Utilized data from 4,181 traumatic SCI patients (1991-2015) from the Japan Association of Rehabilitation Database.
• Employed machine learning (PyCaret) with feature selection (Boruta) and SHAP values for interpretability.
• Evaluated model performance using Area Under the Curve (AUC), focusing on predicting natural urination and defecation at discharge.

Main Results:

• Gradient boosting models achieved high accuracy: AUC of 0.9064 for bladder function and 0.8714 for bowel function.
• Key predictors identified for both functions included L3 motor function, time from injury to admission, and Functional Independence Measure bowel management score.
• A web application integrating these predictive models was developed and made publicly available.

Conclusions:

• Machine learning models demonstrate high accuracy in predicting bladder and bowel dysfunction after traumatic SCI.
• L3 motor function, time to admission, and bowel dysfunction severity are critical predictors for both bladder and bowel outcomes.
• The developed web application provides a valuable tool for predicting and managing SCI-related bladder and bowel dysfunction.