AI-Guided Rehab Edges Ahead in Frozen Shoulder Pilot Trial
A mobile app built around digital therapeutics produced a small but measurable advantage in frozen shoulder rehabilitation, with the clearest signal emerging in side external rotation after 12 weeks. In a single-center, assessor-blinded, randomized controlled pilot study, 18 patients with frozen shoulder were assigned to an AI-based application, ANAPA ME, or to conventional home exercise videos on YouTube. Every participant also received a single ultrasound-guided intra-articular corticosteroid injection before starting a 12-week exercise program. The setup matters because adhesive capsulitis often responds slowly, and even modest differences in home adherence or exercise precision can influence whether the shoulder gradually loosens or remains functionally stuck.
At week 12, the digital therapeutics group achieved significantly greater side external rotation range of motion, with a between-group p value of 0.04, and the improvement from baseline also favored the app arm at p = 0.042. The quality-of-life signal moved in the same direction. On the EuroQol-5 Dimension-5 Level instrument, the study showed a significant group-by-time interaction, p = 0.048, and greater improvement from baseline in the app-treated patients, p = 0.013. These findings point toward a plausible rehabilitation advantage when an exercise program is not simply delivered, but continuously structured and potentially personalized through software.
The shoulder did not improve uniformly across every endpoint. Pain scores, disability measures, and the Constant Shoulder Score did not separate significantly between groups, despite the motion and quality-of-life findings. In a pilot trial this small, that pattern is not surprising. Pain in frozen shoulder can be noisy and heavily influenced by day-to-day variation, while range of motion, particularly external rotation, often serves as a sensitive mechanical readout of capsular restriction and recovery. A digital platform may be better at reinforcing repeated, technically correct mobilization than at shifting broader symptom inventories over a short horizon.
Why External Rotation Moved First
Frozen shoulder is a disorder of stiffness as much as pain. Fibrotic tightening of the capsule, especially in the rotator interval and surrounding structures, typically narrows external rotation early and often leaves it as one of the last motions to recover. A guided app can plausibly help by improving exercise timing, repetition, and technique while reducing the drift that commonly appears when patients are left to self-direct home rehabilitation from generic video content. The biologic logic is straightforward: after corticosteroid injection quiets inflammation, the remaining challenge is mechanical remodeling, and that process depends on whether stretching and mobility work are sustained accurately enough to remodel the contracted capsule.
Because both study arms received the same injection, the difference between them centers on rehabilitation delivery rather than pharmacology. That design gives the exercise platform more weight than it would have in an uncontrolled comparison. It also keeps the interpretation tight: the app was not replacing standard care, but refining how standard care was executed at home. The result was not a sweeping improvement across every clinical scale, yet the direction of benefit was consistent where shoulder mechanics are most unforgiving.
A Pilot Signal, Not a Practice Standard
The trial was intentionally small, and that size limits how much confidence clinicians can place in the magnitude of benefit. With only 18 patients, chance imbalances can shape outcomes, and a single-center design raises the possibility that local patient behavior, therapist oversight, or training routines influenced the results. Even so, the study offers an important early read on what digital rehabilitation may contribute in musculoskeletal care: not a dramatic reinvention of treatment, but a structured way to improve adherence and execution outside the clinic.
The choice of comparator also deserves attention. YouTube-based home exercises are accessible and familiar, but they are static. An AI-based application can potentially adapt prompts, sequence exercises, and keep patients engaged in ways that generic video content cannot. In a condition where improvement often depends on repeated movement over weeks rather than a single intervention, those differences may be clinically meaningful even when pain scales remain unchanged.
The study was registered in Korea before patient enrollment, and funding came from OrthoCare Co., Ltd., with the sponsor having no role in study design, data collection, analysis, interpretation, or manuscript writing. Patent applications related to the ANAPA ME application are pending. Those disclosures do not invalidate the signal, but they frame the need for replication in larger, independent cohorts.
What the Next Trial Will Need to Prove
For frozen shoulder, the key question is no longer whether rehabilitation matters. It does. The real question is whether AI-guided delivery can consistently outperform conventional digital instruction once the sample is large enough to test durability, adherence, and real-world usability across more patients and more severe stiffness. Future studies will need to determine whether the external rotation gain persists, whether it translates into meaningful daily function, and whether similar effects appear without the co-intervention of steroid injection.
For now, this pilot study places mobile application-based digital therapeutics on the map as a plausible adjunct in frozen shoulder rehabilitation. The strongest early signal came from the motion that frozen shoulder takes away first and often returns last: external rotation.