Utilizing Predictive Analytics to Understand Neurogenic Bladder Symptom Score (NBSS) Variations in Adults With Acquired Spinal Cord Injury.

Abstract

Introduction: Individuals with spinal cord injury (SCI) have varying bladder health trajectories after their injury. We explored whether a predictive machine learning model could identify which variables impact urinary symptoms.

Methods: We used 238 variables from the Neurogenic Bladder Research Group SCI registry for a Decision Tree analysis (eCHAID technique). The primary outcomes were the baseline Neurogenic Bladder Symptom Score (NBSS), and the change from the baseline NBSS at 1-year follow up (measured as better/worse than the median change).

Results: Among the 1479 participants, mean baseline NBSS was 24.16 ± 0.28 (standard error of the mean). Our decision tree that evaluated the NBSS at baseline predicted that individuals with a suprapubic tube/urostomy as their primary bladder management method and good bowel QOL at baseline had the lowest (best) mean baseline NBSS at 13.44 ± 0.83. In contrast, females with baseline spontaneous voiding had the highest (worst) mean baseline NBSS at 34.42 ± 1.05. Our second decision tree evaluated the change in the NBSS at 1-year follow-up. Of the 711 participants that performed better than the median change (i.e., improved), 45% were accounted for jointly by women who did not use bladder relaxing medications at baseline, and men without a history of prior urinary tract infections who used a single bladder management method at follow-up. The predictive capacity the decision tree was 57%.

Conclusions: Decision tree models help identify combinations of patient characteristics which correlate with urinary symptoms after SCI. However, there was a limited predictive capacity of the decision tree to forecast future bladder symptoms.