Did longer AI-assisted Parkinson’s rehab improve movement?
No clear motor advantage emerged between one, two and three months of home training. The peer-reviewed Chinese trial randomized 120 people, analysed 71 and had no usual-care group, so its exploratory cognitive signal cannot establish benefit.
The short version
- What happened
- The trial randomized 120 people with early-stage Parkinson’s disease equally to one, two or three months of home training, scheduled for 15–30 minutes on five days each week.
- Who it affects
- People with Parkinson’s should not infer that three months of this system improves movement more than a shorter course.
- What remains unproven
- The primary motor comparison was negative.
Keep in mind
The primary motor comparison was negative.
Only 71 of 120 randomized participants were included in the final analysis, creating substantial uncertainty from attrition and reduced power.
There was no usual-care, non-AI digital or in-person comparison group, so the study cannot isolate the effect of the AI component.
The cognitive responder difference was post hoc and exploratory, while related correlations did not survive correction for multiple testing.
The sample covered early-stage Parkinson’s disease at five Chinese centres and does not establish safety or effectiveness for broader populations.
Source analysis · 1 linked source. Primary analysed source date: 7 October 2026.
Read the full report · 6 min The evidence, methods, implications and limitations
The answer is no clear motor advantage
Longer exposure to the AI-assisted home programme did not produce a statistically significant advantage on the trial’s primary motor measure at three months. That is the central finding. The study compared different durations of the same rehabilitation approach, not AI-assisted rehabilitation against usual care, in-person physiotherapy or no exercise. It therefore cannot show whether the system itself improved movement; it can only ask whether continuing it for longer separated the three assigned groups.
The negative primary result matters because duration is a practical decision for patients, clinicians and health systems. A three-month digital programme demands more sustained effort, supervision and technical access than a one-month programme. This trial does not provide evidence that the longer commitment yields greater motor improvement. Its exploratory cognitive finding is interesting, but the paper itself says the result does not permit causal inference and needs a larger, prespecified test.[1]
Three groups changed duration, not the technology
Researchers at five Chinese centres enrolled people with early-stage Parkinson’s disease, defined as Hoehn–Yahr stage 2.5 or lower, and randomized 120 participants in a 1:1:1 ratio. The short, medium and long groups were assigned one, two or three months of training. Sessions were scheduled for 15 to 30 minutes, five days a week. Raters assessing outcomes were blinded to allocation, which reduces one important source of measurement bias even though participants necessarily knew how long they trained.
The system facilitated remote exercise rather than making a diagnosis or replacing a clinician. The published abstract does not support interpreting the algorithm as an autonomous treatment. A useful deployment question is therefore whether its exercise selection, movement feedback and remote monitoring add value beyond a comparable non-AI home programme. Because every arm received the AI-facilitated approach and only duration varied, this study cannot isolate that contribution.[1]
The denominator fell from 120 randomized to 71 analysed
Final analysis included 71 of the 120 randomized participants—about 59%. Losing 49 people after allocation is not a minor detail. Attrition reduces statistical power and can upset the balance produced by randomization if the people who remain differ from those who leave. The abstract describes attrition as high, so the absence of a significant difference should not be read as proof that duration never matters; the trial may have been unable to detect a modest effect.
For people considering home rehabilitation, retention is also an outcome in its own right. A digital exercise programme that works only for those who persist may have limited population benefit. Future reports should make clear why participants withdrew, whether dropout differed by group, how much assigned training was actually completed and whether analyses that include all randomized participants lead to the same conclusion.[1]
Motor function was primary; cognition was exploratory
The prespecified primary outcome was change from baseline to three months on Part III of the Movement Disorder Society Unified Parkinson’s Disease Rating Scale, a clinician-rated measure of motor signs. Secondary outcomes covered cognition, mood, gait, activities of daily living and safety. The primary comparison was negative: the paper reports no significant between-group difference in motor change at three months.
A post-hoc responder analysis found a higher proportion of cognitive responders in the three-month group than in the one-month group. Post-hoc thresholds are selected after the main analysis and are more vulnerable to chance findings, especially when many outcomes are examined. Weak correlations between changes in cognition and motor parameters did not remain statistically significant after false-discovery-rate correction. The responsible conclusion is not that longer AI rehabilitation improves cognition, but that the pattern may justify a specifically designed trial.[1]
No recorded harm is reassuring but not conclusive
The researchers recorded no falls or adverse events during the trial. That is reassuring for the participants who completed the observed programme. It does not establish safety for everyone with Parkinson’s disease: the sample was limited to early-stage disease, the final denominator was 71, and people unable or unwilling to continue may have different risks from completers.
Safe scaling would need screening for mobility, cognition, vision, hearing, home hazards and digital access, alongside a clear route to a human clinician when the system detects difficulty or when symptoms change. Absence of recorded events in a small exploratory trial should support further evaluation, not removal of clinical safeguards.[1]
What would change the assessment
Confidence would rise with a larger trial that compares a fixed AI-assisted programme with usual care, a matched non-AI digital exercise programme and, where feasible, therapist-led rehabilitation. Allocation, primary outcomes, cognitive responder thresholds and missing-data handling should be fixed in advance. An intention-to-treat analysis should preserve every randomized participant, while adherence and reasons for withdrawal should be reported by arm.
The most useful outcomes extend beyond short rating-scale changes: falls, participation in daily life, caregiver burden, access, technical failures, quality of life and durability after training stops. Independent replication outside the five Chinese centres would show whether language, housing, clinical pathways and internet access alter performance. Until then, the study supports feasibility and further research, not a claim that a longer AI-assisted course improves movement or cognition.[1]