A retrospective cohort of 135 early-stage spinal cord injury (SCI) patients with neurogenic bladder (NB) was studied to predict indwelling catheter conversion (ICC) within 1 year using five machine learning survival models, with SHAP analysis to identify key early predictors.
The Random Survival Forest (RSF) model using 5 SHAP-selected variables achieved a validation C-index of 0.8742 and time-dependent AUCs of 0.9352, 0.8834, and 0.8803 at 3, 6, and 12 months. It stratified patients into favorable, intermediate, and unfavorable groups with 3-month ICC rates of 97.30%, 50.23%, and 2.22% (log-rank p < 0.0001). Median indwelling catheter time was 72 days; 75.56% of patients achieved ICC during follow-up.
Single-center retrospective design limits generalizability. Relatively small sample (n=135) may affect model stability. No external validation cohort was used.
Five early, routinely obtainable variables — H-reflex, lower extremity motor score, UTI, time from lesion to rehab facility, and SCIM bladder score — can stratify SCI-NB patients by likelihood of catheter conversion, helping clinicians personalize bladder management plans early in rehabilitation.
Explore related topics
