Retrospective cohort links combined procedure to early gains in frozen shoulder

In a two-center retrospective cohort study of adults with refractory frozen shoulder, manipulation under anesthesia combined with platelet-rich plasma and corticosteroid injection was associated with better early pain, function, disability and forward-flexion outcomes than arthroscopic capsular release. The clearest potentially meaningful signal appeared at three months, when the combined strategy was associated with a 9.09-point higher American Shoulder and Elbow Surgeons score. Most differences were no longer evident by 12 months, and the nonrandomized design does not establish that either treatment caused the observed outcomes.

The analysis included 137 patients treated through routine care between January 2019 and December 2024 at The Second Affiliated Hospital of Anhui Medical University and Shannan City People’s Hospital in China. Sixty-seven received manipulation under anesthesia followed by intra-articular platelet-rich plasma and a 40 mg triamcinolone acetonide injection, while 70 underwent arthroscopic capsular release. Eligibility required persistent pain and restricted active and passive shoulder motion despite at least three months of conservative treatment, as well as at least 12 months of follow-up. Of 200 screened records, 63 were excluded: 15 did not meet diagnostic criteria, 12 had full-thickness rotator cuff tears, 10 had previous shoulder surgery, 18 had less than 12 months of follow-up and eight had insufficient records for analysis.

Treatment was not assigned randomly. Patients and clinicians selected between the complete pathways through shared decision-making, with factors including symptom duration, imaging findings, patient preference, anticipated invasiveness, recovery, safety, resources and financial circumstances. The manipulation group received general anesthesia, gradual short-lever-arm movement after muscle relaxation and scapular stabilization, and reassessment of passive motion. Autologous platelet-rich plasma was prepared from approximately 8 mL of peripheral blood and 4–5 mL of leukocyte-reduced material was injected into the glenohumeral joint after manipulation. The arthroscopy group underwent radiofrequency ablation and shaver-assisted release of the rotator interval, coracohumeral ligament and anterior and inferior capsule, with posterior release added when posterior contracture was identified. The available procedural summary did not establish whether the corticosteroid was mixed with platelet-rich plasma or injected separately, or record the final injectate volume.

Both groups followed the same rehabilitation principles under the treating surgeon and rehabilitation team. Exercises began after recovery from anesthesia when pain was tolerable, initially emphasizing passive and assisted active forward flexion, abduction, external rotation and internal rotation. Stretching was performed several times daily, with active movement and rotator-cuff and periscapular strengthening added as symptoms and mobility improved. Outcomes were evaluated before treatment and at 1, 3, 6 and 12 months. Pain was measured with the visual analog scale, function with the American Shoulder and Elbow Surgeons score and disability with the Shoulder Pain and Disability Index. Clinicians also measured passive flexion, abduction, external rotation and internal rotation in degrees. The one-month pain comparison was highlighted after data collection rather than prospectively registered, making it an exploratory comparison rather than a prespecified primary outcome.

Both pathways were followed by substantial improvement, but several early measures favored the combined strategy. At one month, the adjusted mean visual analog scale score was 0.57 points lower in the manipulation group, with a 95% confidence interval of −0.75 to −0.38. Its American Shoulder and Elbow Surgeons score was 2.00 points higher, with a 95% confidence interval of 0.88 to 3.12, while its Shoulder Pain and Disability Index score was 2.66 points lower, with a 95% confidence interval of −3.44 to −1.89. The corresponding Holm-adjusted probability values were 0.001 or lower. At three months, the combined group’s American Shoulder and Elbow Surgeons score was 9.09 points higher, with a 95% confidence interval of 7.97 to 10.22 and a Holm-adjusted P value below 0.001. Pain and disability no longer differed significantly at that point, and no later pain comparisons were statistically significant.

The magnitude of the early findings requires caution because statistical significance does not necessarily indicate clinical importance. Contextual reference values came from a separate cohort of nondiabetic patients undergoing arthroscopic capsular release; its distribution-based within-patient minimal clinically important differences at six months were 1.1 points for pain and 8.2 points for function. Those estimates arose from a different population, time point and within-patient construct and are not validated between-group thresholds for the present cohort. The one-month pain and function differences in this study were smaller than those contextual values, whereas the three-month function estimate was numerically larger. A reported 17-point smallest detectable difference for the Shoulder Pain and Disability Index in adhesive capsulitis was not directly comparable, but the one-month difference of 2.66 points was smaller than that reference. The clinical importance of the early estimates therefore remains uncertain.

Passive forward flexion also showed an early difference in favor of manipulation, with an adjusted mean difference of 8.31 degrees at one month and a 95% confidence interval of 5.52 to 11.10. The groups had no significant follow-up differences in abduction or external rotation. At 12 months, the main model identified a 2.10-degree internal-rotation difference favoring the combined strategy, with a 95% confidence interval of 1.01 to 3.20 and a Holm-adjusted P value below 0.001. This result was not confirmed as statistically significant in the inverse probability of treatment-weighted sensitivity analysis. The weighted mean difference was 2.03 degrees, with a 95% confidence interval of 0.02 to 4.03 and an adjusted P value of 0.193. Because the difference was small, lacked a validated disease-specific clinical-importance threshold and was sensitive to the weighting analysis, the authors treated it as uncertain rather than a durable advantage.

By 12 months, the groups had nearly identical outcomes for pain, function, disability, forward flexion, abduction and external rotation. Mean visual analog scale scores were 0.40 in both groups, while mean American Shoulder and Elbow Surgeons scores were 92.72 after the combined strategy and 93.21 after arthroscopic release. Patient satisfaction was 98.5% in the combined group, based on 66 respondents, and 90.0% in the arthroscopy group, based on 63. The difference was not statistically significant, with an odds ratio of 7.33, a 95% confidence interval of 0.88 to 61.31 and P = 0.063. No serious procedure-related complication or reintervention was identified in either group. The zero-event findings had exact upper 95% confidence limits of 5.36% and 5.13%, and the cohort was too small to estimate rare harms precisely. Records also did not systematically capture mild or transient events such as pain flares, short-lived sensory symptoms or recurrent stiffness that did not require another intervention.

The hospital-recorded aggregate treatment amount was lower for the combined pathway. It averaged 7,975.49 Chinese yuan, compared with 14,897.57 Chinese yuan for arthroscopic capsular release. After adjustment for age, sex, dominant-side involvement, symptom duration, diabetes and study site, the mean difference was −6,903.88 Chinese yuan, with a 95% confidence interval of −7,196.76 to −6,611.00 and P below 0.001. This figure should not be treated as a standardized direct medical cost or proof of cost-effectiveness: itemized procedure, hospitalization, consumable and medication charges were unavailable, as were the billing window and price-standardization procedures. Rehabilitation expenses, return to work, quality-adjusted life-years and other indirect costs were not measured. Differences in billing systems and prices between the centers could also remain despite statistical adjustment for study site.

Repeated outcomes were analyzed with mixed-effects models adjusted for baseline values, age, sex, symptom duration, diabetes, dominant-side involvement and study site. These models accounted for repeated observations within patients, while Holm correction was applied across the four postoperative comparisons for each outcome. Inverse probability of treatment weighting was used as a sensitivity analysis for differences associated with treatment allocation. A ridge-penalized propensity model was required because conventional logistic estimation showed near-complete separation. Stabilized weights were truncated at the 2.5th and 97.5th percentiles, improving balance across recorded covariates; all post-weighting absolute standardized mean differences were below 0.10, and the effective sample size was 128.5. Weighted longitudinal estimates were directionally consistent with the primary models for early pain, function, disability and forward flexion, but did not confirm the 12-month internal-rotation finding.

The findings describe two routine-care strategies rather than the independent effects of their components. Manipulation may disrupt adhesions, corticosteroid may reduce early inflammation and platelet-rich plasma was included as a biological adjunct, but the study included no comparison groups receiving manipulation alone, platelet-rich plasma alone or corticosteroid alone. The evidence therefore cannot determine whether any component produced the early association or whether differences in manipulation technique, injection delivery, arthroscopic tissue trauma or postoperative discomfort explain the trajectory. The retrospective design is the central limitation: shared decision-making may have influenced allocation, and neither regression adjustment nor inverse probability weighting can remove unmeasured confounding. Detailed imaging findings, patient preferences, financial circumstances and surgeon judgment were not consistently available, while excluding records with short follow-up or incomplete data may have introduced selection bias. Assessors were not blinded, rehabilitation adherence was not recorded, and one arthroscopy patient had missing postoperative abduction measurements for undocumented reasons. The results can inform shared decision-making, but they do not establish causal superiority, comparative safety, equivalence, noninferiority, component-specific efficacy or cost-effectiveness.

Subject of Research: Comparative early recovery and recorded treatment amounts after combined manipulation and arthroscopic capsular release for refractory frozen shoulder

Article Title: MUA with PRP and corticosteroid injection versus arthroscopic capsular release for refractory frozen shoulder: a retrospective cohort study

Article References: Ye, C., Ciren, L., Wang, Y., Yang, L., Wang, K., Lan, S., Sun, K., Wang, C., Pan, S., & Li, J. (2026). MUA with PRP and corticosteroid injection versus arthroscopic capsular release for refractory frozen shoulder: a retrospective cohort study. Frontiers in Medicine, 13, Article 1957034. https://doi.org/10.3389/fmed.2026.1957034

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1957034

Keywords: frozen shoulder, adhesive capsulitis, manipulation under anesthesia, arthroscopic capsular release, platelet-rich plasma, corticosteroid injection, retrospective cohort, shoulder function, patient-reported outcomes, treatment amount, corticosteroid, injection