The Autonomous Clinic: Navigating AI, Sensors and the Equity Gap in Modern Physiotherapy

Wearable sensors have entered physiotherapy through two distinct routes. The first is passive monitoring, in which body-worn devices quantify movement and activity without attempting to alter them. The second is real-time biofeedback, in which sensor-derived signals are returned to the patient during performance with the explicit intention of changing it. These applications are frequently discussed as a single technological trajectory, yet they rest on different assumptions, require different forms of validation and have accumulated evidence of markedly different quality. This critical narrative review examines the strength, consistency and methodological adequacy of the evidence linking wearable sensing to physiotherapy outcomes, and interrogates the widely repeated but seldom tested premise that better measurement of movement produces better rehabilitation. Literature was identified through open scholarly databases and indexes, supplemented by citation searching, and appraised for design adequacy, measurement transparency, control condition credibility and outcome relevance. Three findings dominate the synthesis. Measurement performance is parameter-dependent rather than device-dependent: spatial gait descriptors achieve agreement adequate for clinical inference, whereas temporal descriptors, joint kinematics and measurements taken on impaired limbs retain error margins that overlap the changes clinicians seek to detect. Passive monitoring reveals a reproducible dissociation between what patients can do under observation and what they do in daily life, and interventions that improve one do not reliably improve the other. Real-time biofeedback produces effects that track the specificity of the biomechanical target rather than the sophistication of the sensing hardware; feedback aimed at a mechanistically justified, individually determined target has altered loading, symptoms and tissue-level outcomes, whereas feedback appended to an already effective behavioural intervention has repeatedly added nothing. Confidence in generalisation is limited by small samples, brief follow-up, weak control conditions and concentration of evidence in few research groups. Priorities include target-specific efficacy trials, retention-focused designs, error thresholds anchored to clinical decisions, and implementation research addressing workforce readiness and equitable access.