Can machine learning improve on the early prediction of upper limb recovery after stroke?
G J van der Gun, C G M Meskers, E R Andrinopoulou and 5 others
PMID 41146217WHAT IT FOUND
An XGBoost model predicted six-month arm recovery from stroke more accurately than a mixed-effects model, with a median error of 4.3 points on the 57-point ARAT scale when applied in the first week.
This error was below the 6-point threshold considered clinically meaningful.
Key findings
01When applied within the first week post-stroke, the XGBoost model achieved a median absolute error of 4.3 ARAT points for six-month upper-limb capacity, compared to 13.7 points for the mixed-effects model.
02The XGBoost model's 80% prediction intervals captured 79% of observed outcomes at two weeks, whereas the mixed-effects model's intervals captured only 66%, showing XGBoost better reflects true uncertainty.
03Prediction errors were highest for patients with low baseline ARAT scores, and the model still showed suboptimal results for approximately 5% of patients with baseline scores below 10 points.
STILL TO COME
How it was doneWhat they foundWhat it means for PTsWhat it means for OTs
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
The model was only internally validated on the same dataset it was trained on; external validation in other populations is crucial before clinical implementation. The dataset consisted of patients with ischaemic stroke, mostly included before the widespread use of endovascular thrombectomy, so it may not reflect current treatment standards. The model performs poorly for a small subset (~5%) of patients with very low baseline ARAT scores (<10 points), potentially misinterpreting early voluntary movement. The study used only clinical variables; it did not include biomarkers like TMS or neuroimaging, which might improve predictions for difficult cases. The sample was drawn from Dutch cohorts, and generalisability to other healthcare systems or populations is unknown.
Declared interests
The study was funded by the Netherlands Organization for Health Research and Development, the European Research Council, the Dutch Society of Physical Therapy, and the Dutch Brain Foundation, among others. No commercial conflicts of interest are explicitly declared in the text.
The easy way to misread this
Do not use this model in clinical practice yet. It has only been tested on the same data used to build it and lacks external validation, meaning its accuracy in your specific patient population is unknown.
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 →