Machine learning for early dynamic prediction of functional outcome after stroke
Problem
Predicting functional outcome after stroke is complicated by fluctuations in the first days, yet is needed for treatment planning and resource allocation.
Why it matters
Accurate early prediction guides life-critical treatment decisions and hospital resource use.
What the AI did
A transformer model integrated continuous clinical, physiological, imaging and biological data from 72h of admission to generate hourly mortality and morbidity predictions, with SHAP explanations and external validation on MIMIC-III.
Results
not stated in source
Impact
Model provides hourly outcome predictions validated on an external database; exact performance metric truncated.
Record
| Venue | Communications Medicine |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Julian Klug, Guillaume Leclerc, Elisabeth Dirren, Emmanuel Carrera |
| Citations | 30 (OpenAlex, at import time) |
| References | 94 |
| Model or method | Transformer model |
| Organization | Geneva University Hospital |
| Primary source | https://doi.org/10.1038/s43856-024-00666-w |
| Open access | gold |
| Keywords | stroke, transformer, outcome prediction, machine learning, MIMIC-III, SHAP |
Importance rating
significant — thousands to millions affected
Abstract states outcome prediction after stroke is critical for treatment planning and resource allocation; externally validated hourly predictions support life-critical decisions, but no population scale stated.