Overview
- Published in npj Digital Medicine, the University of Waterloo study modeled three weeks of routine lab results to predict injury presence, severity and in-hospital mortality after spinal cord trauma.
- The models achieved an ROC-AUC of 0.79 for mortality prediction on day one, rising to 0.89 by day 21, reached 0.81 for severity by day seven, and attained 0.71 for detecting SCI by day 21, outperforming SAPS II after the first few days.
- Predictions were possible as early as 24 to 72 hours after admission, offering objective assessments when early neurological exams are infeasible or unreliable.
- Training used the MIMIC critical-care dataset (n=2,615) with external validation on the TRACK-SCI cohort (n=137), supporting generalizability across clinical contexts.
- Authors highlight that accuracy improves as more labs accrue and note potential to distinguish motor complete from incomplete injuries, while calling for larger external validation, prospective trials and workflow integration.