Interpretable Machine Learning for Stroke Recovery: Predicting Discharge and 3-Month Functional Outcomes.
Inês Carvalho Martins Augusto, Nuno Antonio, Ana Marreiros and 2 others
PMID 41712473WHAT IT FOUND
In 116 stroke patients, age and initial stroke severity were consistent predictors of poor function at discharge and three months.
For three months, discharge destination was selected by the models. This is a prediction model, not evidence that treatment changes outcome.
Key findings
01The best model, extreme gradient boosting, had AUCs of 79% for mRS at discharge and 87% for mRS at three months.
02Age and NIHSS consistently predicted immediate and short-term disability outcomes.
03For three-month mRS, the logistic regression and support vector machine models selected discharge destination.
STILL TO COME
How it was doneWhat they foundWhat it means for PTs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
Only 116 of the 125 collected patients were analysed after cleaning, so the result is small. The data came from a single medical center, so it may not transfer to other stroke services. The study was retrospective, so selection bias may affect who was included. The models were checked with internal cross-validation, not external validation, and the small sample raises a risk of overfitting and unstable model choice. The paper did not include imaging or some clinical variables, so it may miss important predictors. There was no simple baseline model, so the added value of the complex pipeline is unclear. The model predicts association, not cause, and it does not test whether changing rehabilitation improves function.
Declared interests
This work was supported by national funds through FCT, and the authors declared no potential conflicts of interest.
The easy way to misread this
Do not read the AUC values as proof this model will work in your clinic. It was built from 116 patients at a single medical center and not externally validated, and it predicts likely poor function, not the effect of any therapy.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →