Moving rehabilitation home is valuable only if the treatment still works. A randomised neck-pain trial published on 15 October found that a therapist-supervised home programme met its noninferiority target against outpatient rehabilitation.[1]
Researchers at two Chinese hospitals enrolled 116 people for 24 weeks of matched exercises. The home system used movement recognition and threshold rules; therapists controlled treatment adjustments, not an autonomous generative model.[1]
Primary result at 24 weeks
| Measure | Reported result |
|---|---|
| Outcome scale | Neck Disability Index, 0–100; lower is better |
| Home minus clinic difference | −2.5 points |
| 95% confidence interval | −6.2 to +1.2 points |
| Noninferiority margin | +10 points |
What the comparison establishes
Noninferiority asks a different question from superiority. Rather than requiring the new delivery method to win, it asks whether a clinically unacceptable disadvantage can be ruled out within a chosen margin. CONSORT’s reporting guidance treats the choice of margin, comparator and interpretation as central features of this type of trial.[2]
The distinction is visible in the interval above. Its most favourable end supports a home-treatment advantage; its other end allows a small disadvantage. The range remains below the trial’s ten-point boundary. That is why “met the noninferiority target” is more faithful to the result than either “proved identical” or “AI beat physiotherapy.”[1][2]
Both groups improved. Two authors worked for the developer but did not assess outcomes or analyse results. The study did not establish cost savings or the automated component’s separate contribution.[1]
The human benefit is in the delivery option
The meaningful result is that a supported home route can be evaluated against the clinical service it might partly replace. It is not enough for a system to recognise an exercise or issue a reassuring message. A delivery model has to preserve the outcome for which the patient is undertaking the work in the first place.
This also changes how to think about the role of automation. A therapist can remain accountable while technology helps connect sessions between appointments. The relevant comparison is the complete care pathway, not a contest between a person and a machine. That is a more demanding, and more useful, definition of success than showing that software can classify a movement.
The result does not mean everyone with neck pain should switch to home exercise. It concerns the studied condition, programme and population, rather than all possible causes of pain. For service designers, the next practical question is whether the same outcomes can be maintained while measuring the burdens the trial did not resolve: staff time, equipment costs and patients’ ability to keep using the system.
What this study contributes is a credible alternative worth testing further, not a replacement therapist. Its positive finding rests on keeping the clinical goal fixed while changing where, and with what support, the work is done.
Sources & further reading
- AI-Enabled Telerehabilitation for Chronic Nonspecific Neck Pain: Multicenter Randomized Controlled Trial JMIR Formative Research · 15 Oct 2026 · Accessed 28 Oct 2026
- Reporting of Noninferiority and Equivalence Randomized Trials: Extension of the CONSORT 2010 Statement JAMA / CONSORT Group · 26 Dec 2012 · Accessed 28 Oct 2026