Machine Learning Methods Predict Individual Upper-Limb Motor Impairment Following Therapy in Chronic Stroke.
Ceren Tozlu, Dylan Edwards, Aaron Boes and 7 others
PMID 32193984WHAT IT FOUND
Machine learning predicted a chronic stroke patient's post-therapy upper-limb score better than it predicted whether they would make a clinically meaningful gain.
Baseline arm score and a brain-stimulation measure of motor threshold were the strongest predictors.
Key findings
01Elastic-Net was the best machine learning method for predicting each patient's post-therapy upper-limb score.
02The starting upper-limb score was the strongest predictor, followed by the difference in motor threshold; both were selected in every Elastic-Net result.
03Predicting whether a patient would make a clinically meaningful gain was less accurate than predicting the post-therapy score.
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
The analysis used only 102 patients from 2 of the original 12 institutions, and 2 patients had missing data. The paper reports repeated cross-validation and says further validation with larger, multi-modal data is needed. The therapy included warm-up, rest, active or sham rTMS, rest, and task-oriented rehabilitation, so the contribution of any single component cannot be separated. The sham and real rTMS groups did not differ in change in UE-FMA (p = 0.46), so improvement cannot be attributed to rTMS. Only structural T1 MRI was used; diffusion or functional MRI, dexterity, attention, visuo-spatial neglect, sensory deficits, motivation, and depression were not included. The NeMo Tool was based on healthy controls who were younger than the stroke population. Patients with language, extinction, or inattention scores above set thresholds were excluded.
Declared interests
The authors declared no competing interests.
The easy way to misread this
Do not read this as proof that rTMS or machine learning should change patient care. The patients received warm-up, rest, active or sham rTMS, and task-oriented rehabilitation, and the sham and real rTMS groups did not differ in UE-FMA change. The models were evaluated by cross-validation, and the paper says further validation is needed.