When a shoulder will not move, the instinct is to make it stronger. It feels like the obvious thing. The arm has gone weak and useless, so you load it, and you expect the strength to bring the movement back with it.
A trial published in 2023 tested that instinct against the alternative, and the instinct came second.
The design, which is the reason to take it seriously
A rehabilitation department in Beijing recruited forty people with idiopathic frozen shoulder, meaning no injury and no other cause found. Average age fifty four. Symptoms for about three months, so early, in the painful phase rather than the stiff one.
Everybody was randomised, and an independent assessor who did not know which group anyone was in did all the measuring, before and after.
Both groups got exactly the same core treatment. Five minutes of wall climbing to warm up, then forty minutes of standard physiotherapy: Maitland joint mobilisation, glenohumeral distraction and gliding at two to three oscillations a second, stretching with a wand in every direction, and active range of motion work. Identical for both.
Then the two groups split for twenty minutes.
- The strengthening group did what most people would do. Theraband isometrics and one to two kilogramme dumbbells, isotonic work through flexion, extension, abduction and both rotations.
- The neuromuscular group did not strengthen anything. They stood on a motorised platform that tilted and rotated underneath them, holding an elastic belt, with the shoulder set in six different positions in turn. A screen showed them where their centre of gravity was, and the task was to keep it inside a target while the floor moved.
Once a day, five days a week, eight weeks. Forty sessions each.
Read those numbers twice, because there are two separate stories in them. The first is that both groups transformed. External rotation went from roughly 23 degrees to somewhere in the sixties or seventies in both arms of the trial. Nobody was left where they started, and the shared ingredient was the forty minutes of hands-on mobilisation and stretching that both groups received.
The second story is the gap. Seventeen degrees more flexion, ten more of external rotation, nine more of internal rotation, and a full point less pain, bought with the same twenty minutes spent differently.
The shoulder did not need to be told to be stronger. It needed to be told where it was.
Why control might beat strength here
A shoulder is not held together by its bones. The socket is shallow by design, which is what buys the joint its enormous range, and the price of that design is that stability has to be supplied actively, by muscle, continuously, and with good timing.
When a shoulder stops moving because moving it hurts, the joint stops sending its usual stream of position information. The tissue is not the only thing that changes. The map goes stale.
Strengthening a muscle does not restore the map. Asking somebody to hold a position while the floor moves under them does, because that is a task the nervous system has to solve rather than a load the muscle has to carry. The trial's own explanation runs along these lines, and it also points to the same pattern found in other joints: neuromuscular work has outperformed or matched strengthening after knee ligament reconstruction, in degenerative knee disease, and in shoulder dislocation.
What did not differ, which matters
Two movements showed no significant advantage for the neuromuscular group.
- Extension improved in both groups and the difference between them was not significant (p=0.204).
- Abduction came in at exactly p=0.05, which is the boundary rather than a result. Both groups gained roughly sixty degrees of abduction. Reporting that as a win for either side would be dishonest.
So the advantage is real but it is specific: it showed up in pain, in flexion, and in both rotations, and it did not show up in extension or abduction.
The part about injections, since it comes up
Nobody in this trial received a corticosteroid injection. Both groups got manual therapy and exercise, and both groups improved substantially, from around seven out of ten in pain to between two and three and a half.
That is worth saying plainly, because frozen shoulder is often discussed as though an injection were the necessary first move. This trial did not test that question, and it cannot answer it. What it does show is that a shoulder in the painful early phase can drop most of its pain and recover most of its rotation on hands-on treatment and the right exercise alone.
The trial's discussion also points to earlier work reporting that exercises restoring neuromuscular control matched corticosteroid injection for shoulder pain in the short term. That is a claim inside somebody else's reference list rather than something this trial measured, and it is flagged here as exactly that, not offered as established.
How to hold that honestly
Forty people. The authors say so first, and they are right to. Forty is enough to detect a difference this size and nowhere near enough to settle a question.
The machine. The neuromuscular exercise was delivered on a motorised multi-axis platform with a screen giving live feedback on the person's centre of gravity. That is a specific and expensive piece of equipment, and the trial cannot tell us how much of the advantage came from the principle and how much from the device. The authors also note they do not know whether the frequency, intensity and duration they chose were the right ones, because there was nothing to compare against.
Forty sessions. Once a day, five days a week, for eight weeks, in a hospital department. Very few people outside a trial receive that, and no reader should measure their own progress against it.
Eight weeks and then nothing. There was no follow up. Frozen shoulder runs for many months, so a difference at week eight is not the same as a difference at the end, and the authors say the long term effect remains unknown.
Pain and degrees only. No disability or function questionnaire was used, so the trial does not tell us whether the extra rotation changed what people could actually do.
And people with diabetes and thyroid disease were excluded. That keeps the trial clean but narrows it, because roughly 30 percent of people who present with a frozen shoulder have diabetes. The group most likely to walk into a clinic with this condition is the group this trial deliberately left out.
A boundary on the condition itself, too. A shoulder that became stiff straight after a significant injury, or one that arrived with fever, weight loss or a history of cancer, is a different conversation and needs examining rather than reading about.
What we take from it in practice
The hands-on work is doing more than it gets credit for. Every person in this trial had forty minutes of mobilisation, stretching and active range of motion, five days a week, and every person improved a great deal. Whatever the twenty minutes at the end added, it was added on top of that. The exercise did not replace the treatment.
How the twenty minutes is spent is a real decision, not a detail. Same time, same person, same shoulder, measurably different result. That is the practical content of this trial and it costs nothing to act on.
The principle travels even when the machine does not. Almost nobody has a motorised platform. What the platform was doing was forcing the shoulder to solve an unpredictable positional problem instead of lifting a predictable load, and unstable surfaces, unpredictable perturbations, eyes closed, changing the base of support, and holding a position while something else moves are all ways of asking the same question with equipment that fits in a treatment room. That is an argument for the principle, not evidence for any particular substitute, and it should be said in that order.
Do not read this as an argument against strength. The strengthening group also improved enormously, and strength has its own reasons for existing that have nothing to do with the eight weeks of a frozen shoulder. What the trial questions is strengthening as the priority in an early, painful, stiff shoulder, when the joint has lost its sense of where it is.
Expect a long road anyway. Symptoms here had lasted about three months and the trial ran for two more. Frozen shoulder is measured in many months. What appears to be changeable is how much of that time is spent in pain, and how much rotation is there at the end of it.
References
- Wang L, Yu G, Zhang R, Wu G, He L, Chen Y. "Positive effects of neuromuscular exercises on pain and active range of motion in idiopathic frozen shoulder: a randomized controlled trial." BMC Musculoskeletal Disorders. 2023;24:50. doi:10.1186/s12891-023-06173-8. Trial registration ChiCTR2100054453.
- Zreik NH, Malik RA, Charalambous CP. "Adhesive capsulitis of the shoulder and diabetes: a meta-analysis of prevalence." Muscles, Ligaments and Tendons Journal. 2016;6(1):26-34. doi:10.11138/mltj/2016.6.1.026