Augmenting Physical Therapy, What the Evidence Actually Supports

Bottom line

Physical therapy is being augmented from three directions at once, remote delivery, computer-vision movement analysis, and app-based exercise programs. The evidence for these is uneven. Telerehabilitation is well studied and roughly matches in-person care for many conditions. Motion-tracking AI is improving quickly on technical accuracy but has almost no evidence tying it to better patient outcomes. Digital musculoskeletal programs show real engagement gains, though much of the strongest data comes from the companies selling them. The useful question is not whether a tool is impressive. It is whether it closes the two gaps that actually limit recovery, access to a therapist and adherence to the plan.

The two bottlenecks that limit outcomes

Physical therapy works when a patient can reach a therapist and then does the prescribed work between visits. Both conditions fail often.

Access is the harder constraint. The WHO reports that in many countries more than half of people who need rehabilitation do not receive it, and the shortfall is widest in low-income settings (strong, WHO synthesis via scoping review). In some African countries, between 62.5 and 82.5 percent of people who need physiotherapy go without it (moderate, regional review). The workforce numbers explain why. Malawi has around 0.8 physiotherapists per 100,000 people, against nearly 1,350 per 100,000 in the United States (limited, single-country report).

Adherence is the second constraint, and it undermines even patients who reach care. Across musculoskeletal populations, home exercise nonadherence commonly falls between 30 and 50 percent (moderate, research synthesis). One outpatient study in Ethiopia found only 35.3 percent of patients adherent to their home program (limited, single observational study). A prescribed exercise plan that is not done is not treatment.

Any augmentation worth adopting should move one of these two numbers. That is the test we apply to each technology below.

Telerehabilitation has the strongest evidence, and a real ceiling

Telerehabilitation, meaning physical therapy delivered remotely by video, phone, or connected app, is the most studied form of augmentation. A 2024 systematic review and meta-analysis pooled 37 trials and 4,288 participants and found telerehabilitation more effective than control conditions across pain, function, and quality of life, with standardized effect sizes from 0.24 to 0.91 (strong, meta-analysis). The largest functional benefit appeared in people recovering from hip fracture (standardized mean difference 0.87, 95 percent CI 0.34 to 1.41).

The important caveat is in the comparison. When telerehabilitation was measured directly against face-to-face therapy rather than against weaker controls, its advantage disappeared (strong, same meta-analysis). The honest reading is that remote delivery is roughly non-inferior to in-person care for many conditions, not superior. That is still a strong result, because non-inferior care that reaches a patient beats superior care that does not.

Evidence quality is a separate concern. A 2024 umbrella review found that 69 percent of existing telerehabilitation reviews were rated critically low quality on the AMSTAR 2 appraisal tool (moderate, umbrella review). The direction of the evidence is consistent, but the underlying trials are heterogeneous and often small.

AI motion tracking has an accuracy-versus-outcome gap

The most visible new augmentation is computer vision that watches a patient exercise and scores their form. Modern pose-estimation systems can extract joint positions from an ordinary phone camera. Against laboratory motion capture, these systems reach mean absolute errors of roughly 2 to 4 degrees for lower-limb gait angles (limited, single validation study). Smartphone systems for shoulder rehabilitation have shown reasonable agreement with optical motion capture in small samples (limited, 15-participant study).

Those numbers describe technical accuracy, not clinical benefit. High reported accuracy for detecting an exercise or counting a repetition does not establish that automated feedback improves recovery, reduces reinjury, or changes function. Markerless systems also degrade on out-of-plane movement and vary with camera placement and lighting, which matters in a home rather than a lab (moderate, methods review). Motion-tracking AI is maturing as a measurement tool. As a treatment, it remains unproven, and current outcome evidence is emerging at best.

Its plausible near-term value is narrow and real, giving a remote patient immediate and consistent feedback on movement quality so that between-visit exercise is done correctly. That is an adherence-and-quality play, not a replacement for clinical judgment.

Digital musculoskeletal programs work, but read the funding

A commercial category has formed around app-delivered exercise, coaching, and sometimes sensor feedback for musculoskeletal conditions. The pooled evidence for digital health interventions in musculoskeletal pain is positive but modest. A meta-analysis of 56 trials and 9,359 participants found small effects on pain, disability, and quality of life, at moderate quality (moderate, meta-analysis). Small and real is a fair summary.

Vendor evidence sits on top of that base and should be read carefully. A manufacturer-sponsored randomized trial of one digital program, published in npj Digital Medicine with 140 participants, reported engagement and completion roughly double that of in-person therapy with comparable clinical outcomes (moderate but commercially sponsored, company summary). Claims that such programs reduce downstream joint-replacement surgery rest on observational, industry-linked data whose own authors call for randomized trials to establish causation (limited, observational analysis). The engagement gains are believable. The harder economic claims are not yet settled.

What augmentation should mean in a resource-constrained system

Set against the two bottlenecks, a clear pattern emerges. The augmentation with the best evidence, telerehabilitation, is valuable mainly because it extends reach, and reach is precisely what fails in systems with few therapists. Sub-Saharan Africa carries a heavy disease burden with only about 3.5 percent of the world's health workers, so a tool that lets one physiotherapist safely supervise more patients addresses the actual shortage (moderate, scoping review).

This is the frame Curely builds toward. Advanced healthcare intelligence should not be a privilege of well-staffed systems. In physical therapy that means using remote delivery and automated movement feedback to stretch a scarce clinician across more patients and more days, while the therapist keeps authority over diagnosis, progression, and safety. The evidence does not support handing rehabilitation to an algorithm. It does support using these tools to keep patients doing correct work between the visits a therapist cannot physically attend.

Takeaway

Judge physical therapy augmentation by whether it improves access or adherence, and by the strength and independence of its evidence. Telerehabilitation clears that bar as a non-inferior way to reach more patients. Digital exercise programs clear a lower bar with genuine engagement gains and modest outcome effects. Motion-tracking AI does not clear it yet, though it is the strongest candidate for improving between-visit quality once outcome trials catch up. In systems where therapists are scarcest, the case for careful augmentation is strongest, because there the alternative is often no therapy at all.