Machine learning-based adaptive personalization in virtual reality stroke rehabilitation: a systematic review.
Arar Al Tawil, Siti Hazyanti Mohd Hashim, Aseel Aburub and 3 others
ML-adaptive VR improved upper limb scores by 7.47 points on the Fugl-Meyer scale, past the 5.25-point threshold for meaningful change.
But 68% of the 25 studies had fewer than 30 patients, so this is proof-of-concept, not established practice.
Key findings
1The pooled FMA-UE improvement was 7.47 points (95% CI: 5.38–9.57, p < 0.001), exceeding the minimal clinically important difference of 5.25 points.
2The evidence base is limited: 68% of studies recruited fewer than 30 participants, 77% did not report an intention-to-treat analysis, and the predominant designs were pilot or feasibility studies rather than adequately powered RCTs.
3Real-time closed-loop adaptation was the dominant approach, used in 20 studies (80%), with reinforcement learning the most common algorithm category (8 studies).
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
68% of included studies recruited fewer than 30 participants, so individual trials are underpowered and the pooled estimate rests on small samples. 77% of studies did not report an intention-to-treat analysis, and most were unblinded, increasing risk of performance and detection bias. The evidence base is dominated by pilot and single-arm designs (24% pilot, 28% quasi-experimental); only 32% were RCTs. Moderate heterogeneity (I² = 58%) across studies with different algorithms, VR platforms, and intervention durations limits the precision of the pooled estimate. Too few studies to assess publication bias; funnel plot asymmetry could not be formally evaluated. The authors explicitly state the findings represent proof-of-concept evidence, not established clinical effectiveness, and call for larger, blinded, adequately powered RCTs before broad clinical recommendations can be made.
Declared interests
The authors declared that no financial support was received for the work or its publication.
The easy way to misread this
Do not read the 7.47-point FMA-UE improvement as established clinical effectiveness. Sixty-eight percent of the included studies had fewer than 30 participants, 77% lacked an intention-to-treat analysis, and the authors themselves describe the evidence as proof-of-concept rather than established effectiveness. The moderate GRADE rating reflects these methodological gaps, not just statistical heterogeneity.
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 →