Temporal trajectory and predictors of hemiplegic shoulder pain after stroke: a 12-month longitudinal clinical-ultrasound study

Abstract

BACKGROUND

Hemiplegic shoulder pain (HSP) is a frequent complication after stroke. Although HSP is considered a multifactorial condition, the temporal sequence through which different mechanisms contribute to pain remains poorly understood, mainly due to the lack of longitudinal studies starting from the acute phase and integrating imaging data.

AIM

To describe the 12-month temporal trajectory of hemiplegic shoulder pain after stroke and to identify time-specific clinical and ultrasonographic predictors of pain intensity.

DESIGN

Prospective Longitudinal Observational Study.

SETTING

Inpatient stroke unit and neurorehabilitation facilities.

POPULATION

Patients with first-ever ischemic or hemorrhagic stroke and contralateral hemiparesis.

METHODS

Patients were evaluated within 72 hours after stroke onset and at 1, 3, 6, and 12 months. Shoulder pain during passive mobilization was assessed using the Numerical Rating Scale. Clinical evaluation and standardized shoulder ultrasonography were performed at each time point. Univariable and multivariable linear regression analyses identified independent predictors of pain.

RESULTS

Seventy-seven patients were enrolled. Pain increased from the acute phase to a peak at 3 months and progressively decreased thereafter. Reduced passive range of motion was consistently associated with pain. In the acute phase, pain was associated with glenohumeral subluxation. Adhesive capsulitis emerged at 1 month, while at 3 months pain was associated with subluxation, capsulitis, and tendon tears. In the chronic phase, tendon tears became the main determinant of residual pain.

CONCLUSIONS

Hemiplegic shoulder pain follows a temporally differentiated multifactorial pattern, with distinct mechanisms prevailing at different stages after stroke.

CLINICAL REHABILITATION IMPACT

Identifying phase-specific mechanisms of hemiplegic shoulder pain supports targeted, time-specific preventive and therapeutic strategies in neurorehabilitation.

Introduction

Hemiplegic shoulder pain (HSP) is a frequent and clinically relevant complication after stroke, with reported prevalence rates ranging widely from approximately 5% to 84%, depending on diagnostic criteria, timing of assessment, and characteristics of the study population.^1-6 HSP negatively affects upper-limb motor recovery, reduces functional autonomy, and promotes limb disuse, thereby contributing to long-term disability. Pain is commonly elicited during passive shoulder mobilization and may persist for months or years in a substantial proportion of patients, highlighting its detrimental impact on rehabilitation outcomes.^7-9

From a pathophysiological perspective, neurological damage represents the initiating event in a cascade leading to shoulder pain. Muscle weakness, altered muscle tone, and sensory deficits disrupt normal scapulohumeral mechanics, generating a biomechanical imbalance around the glenohumeral joint. This imbalance increases mechanical stress on periarticular soft tissues – including the joint capsule, tendons, ligaments, and bursae – predisposing them to structural lesions that generate nociceptive input and contribute to the development of pain.^10-15

Soft-tissue abnormalities of the hemiplegic shoulder can be effectively assessed using ultrasonography, which allows detailed and dynamic evaluation of periarticular and articular structures. Owing to its accessibility, safety, and ability to provide real-time assessment, ultrasound is considered the imaging modality of choice for investigating post-stroke shoulder pathology and its evolution over time.^{12, 16, 17}

Ultrasound studies have reported a wide spectrum of structural abnormalities associated with HSP, including glenohumeral subluxation, adhesive capsulitis, tendinopathies and tears of the rotator cuff and of the long head of the biceps tendon (LHBT), subacromial-subdeltoid bursitis, and joint effusion.^{16, 18, 19}

These findings support the multifactorial nature of HSP. However, most of the available evidence derives from cross-sectional studies, which provide only a “snapshot” of shoulder pathology at a single time point and therefore cannot capture the temporal dynamics through which different mechanisms contribute to pain.^{17, 19, 20}

Longitudinal clinical and ultrasound studies on HSP are scarce,^{16, 21, 22} and only a single study has adopted a longitudinal design starting from the acute phase after stroke; notably, this investigation relied exclusively on clinical assessment, without any imaging evaluation of shoulder structures.²³ Taken together, to date, no study has systematically investigated the parallel evolution of pain and structural shoulder abnormalities beginning in the acute phase of stroke.

Consequently, the temporal sequence through which HSP emerges and progresses, as well as the relative contribution of specific pathophysiological mechanisms at different stages of recovery, remains poorly understood.

To address this gap, our group recently conducted the first longitudinal study integrating systematic clinical and ultrasound assessments starting from the acute phase of stroke (≤72 hours). In that preliminary cohort, we showed that shoulder pain could already be detected in the acute phase and that glenohumeral subluxation and adhesive capsulitis were the only ultrasound findings associated with pain during the first three months after stroke.²⁴

That preliminary work, however, was limited by the small sample size and by follow-up restricted to three months. To overcome these limitations and to provide a more comprehensive characterization of the temporal evolution of HSP, we extended the cohort and prolonged follow-up to six and twelve months.

The objectives of the present study were therefore: 1) to characterize the 12-month temporal trajectory of HSP from the acute to the chronic phase; 2) to identify the principal clinical and ultrasound predictors of pain at each stage; and 3) to determine their independent contributions using multivariable analyses. By mapping how different mechanisms emerge and interact over time, this study aims to refine the pathophysiological model of HSP and to inform mechanism-based, phase-specific strategies for early screening and prevention.

Materials and methods

Ethical considerations

The study was approved by the Regional Ethics Committee of Liguria (N.CET–Liguria), protocol number 254/2024 (DB id 13923), on October 14^th, 2024, and was conducted in accordance with the principles of the Declaration of Helsinki as revised in 2024. All participants provided written informed consent prior to inclusion in the study.

Participants

Consecutive patients admitted to IRCCS Ospedale Policlinico San Martino (Genoa, Italy) for a first-ever ischemic or hemorrhagic stroke with contralateral hemiparesis were prospectively recruited between 18^th October and 31^st December 2024. Inclusion criteria were age ≥18 years, stroke onset within 72 hours, and a clinically confirmed contralateral motor deficit demonstrated by anti-gravity testing. Exclusion criteria were severe impairment of consciousness, cognitive or behavioral disturbances precluding cooperation, and a history of shoulder pain prior to stroke onset.

Seventy-seven patients were enrolled (mean age 68.6±10.9 years; 53 males, 68.8%). Fifty-three (68.8%) had an ischemic stroke and 24 (31.2%) a hemorrhagic stroke; hemiparesis involved the right side in 49 (63.6%) and the left in 28 (36.4%). At the enrollment (baseline), National Institutes of Health Stroke Scale (NIHSS) score was 11.9±5.3 (mild impairment <5; moderate impairment 5-15; severe impairment >15).

Study design

This was a longitudinal observational study. Clinical and ultrasound evaluations were performed within 72 hours (T0) and at one (T1), three (T2), six (T3), and 12 months (T4) after stroke.

Attrition over follow-up resulted in 8 dropouts at T1, 3 at T2, 6 at T3, and 5 at T4.

Because of its observational design, the study did not interfere with standard clinical care. All participants received rehabilitation and shoulder pain treatments as deemed clinically appropriate by the treating team, according to routine practice. Study participation did not modify rehabilitation protocols or pain management strategies.

Clinical evaluation

Pain intensity during passive shoulder mobilization (flexion, abduction, and external rotation) was assessed using the Numerical Rating Scale (NRS; 0-10, where 0 indicates no pain and 10 the worst imaginable pain). Pain was rated separately for each movement direction, and the highest value was retained as the representative pain score for each patient.

The passive and active range of motion (ROM) of the shoulder was measured with a manual goniometer for flexion (0-180°), abduction (0-180°), and external rotation (0-80°). For each participant, the ROM achieved in each movement was expressed as a percentage of its normal reference value. These percentages were then averaged to derive a single composite ROM index, reflecting the participant’s overall shoulder mobility on a 0-100% scale. This composite ROM index was used for subsequent analyses.

Muscle strength of the shoulder flexors, abductors, and external rotators was evaluated using the Medical Research Council (MRC) scale (0-5, from no contraction to normal strength). Muscle tone of the shoulder extensors, adductors, and internal rotators was measured using the Modified Ashworth Scale (MAS) (0-4, from normal tone to the shoulder fully rigid in the tested movement). For both strength and tone, the mean score across the three assessed movement directions was calculated and used for subsequent analyses.

Upper limb disability was assessed using the Disabilities of the Arm, Shoulder, and Hand (DASH) questionnaire (0-100, higher scores indicating greater disability). Stroke severity was measured using the NIHSS (0-42, higher scores indicating greater impairment).

All clinical measures were assessed at all five time points (T0, T1, T2, T3, and T4), except for the NIHSS, which was administered only at baseline (T0).

Ultrasonographic evaluation

Ultrasound assessments were performed using a portable ultrasound machine (Minisono Alpinion, Seoul, Republic of Korea) equipped with a linear probe (frequency range 3-12 MHz) connected to a Microsoft Surface Pro 7 tablet. All ultrasound examinations were performed by two experienced physiatrists, who conducted the assessments jointly to ensure real-time agreement on image acquisition and interpretation. Any uncertainties or discrepant findings were discussed immediately and resolved by consensus at the time of examination. Each shoulder was scanned in both long- and short-axis views with the patient seated. When useful, dynamic maneuvers (abduction/external rotation or passive motion) were applied to improve visualization of tendon integrity and capsular mobility. Both shoulders were examined, and the contralateral side served as an internal reference. Images and short cine-loops of key findings were archived.

Tendon abnormalities were assessed in the supraspinatus, infraspinatus, subscapularis, teres minor, and LHBT, and classified as tendinopathy or tear. Tendinopathy was defined as focal or diffuse tendon thickening with reduced or heterogeneous echogenicity and preserved fiber continuity. Tears were classified as partial-thickness or full-thickness.²⁵ When both conditions coexisted in the same tendon, the diagnosis of tear prevailed. For each participant, the presence of tendinopathy and tear was recorded separately for each tendon, and the total number of tendons affected by tendinopathy and by tear was calculated, yielding two summary counts used for subsequent analyses.

Adhesive capsulitis was diagnosed when the inferior glenohumeral capsule (axillary recess) measured >4 mm or showed >60% thickening compared with the contralateral side. Additional signs (coracohumeral ligament thickening, rotator-interval changes, biceps-sheath effusion) were recorded when present but not required for diagnosis.^{26, 27}

Glenohumeral subluxation (inferior diastasis) was defined as an acromion-greater tuberosity distance >4 mm compared with the contralateral side.²⁸

Subacromial-subdeltoid (SASD) bursitis was defined as an anechoic or hypoechoic fluid collection distending the bursal walls, with or without associated synovial thickening, visible in both longitudinal and transverse planes.¹⁷

LHB tenosynovitis was identified by the presence of an anechoic or hypoechoic effusion surrounding the tendon within its sheath, often accompanied by synovial hypertrophy or Doppler signal increase.²⁸

Intra-articular effusion was defined as an anechoic or hypoechoic fluid collection within the glenohumeral joint recesses, typically visualized in the posterior or anterior approach, distinguishable from surrounding capsule or labrum structures.²⁸ When both the effusion of LHB sheath and the effusion of the posterior recess were visualized, only joint effusion was considered.

Glenohumeral subluxation, capsulitis, SASD bursitis, LHB tenosynovitis, and intra-articular effusion were all classified dichotomously (present/absent) and analyzed accordingly. All ultrasonographic measures were assessed at all five time points (T0, T1, T2, T3, and T4).

Statistical analysis

All statistical analyses were performed using Jamovi (version 2.3.28). The significance level was set at P<0.05 (two-tailed).

To identify HSP predictors, both univariable and multivariable linear regression analyses were conducted, with NRS pain intensity treated as a continuous dependent variable. This approach was chosen to preserve information on pain severity and maximize statistical power. Clinical and ultrasound variables were selected a priori based on clinical relevance or physiological plausibility. At each time point, NRS scores were analyzed in relation to the clinical and ultrasound variables measured at the corresponding time point. The only exception was the NIHSS score, recorded only at baseline (T0), which was tested as a predictor of pain intensity at all subsequent time points to assess whether initial stroke severity predicted later pain levels.

In multivariable models, the same variables were entered simultaneously to estimate their independent contribution to pain intensity. For each model, unstandardized coefficients (β), standard errors (SE), standardized coefficients, t-values, and p-values were reported. Model fit was expressed as the coefficient of determination (R²).

Only for the estimation of HSP prevalence, and in order to ensure comparability with previous studies,^{22, 29} HSP was defined using an NRS cut-off ≥ 4. In contrast, all univariable and multivariable analyses treated NRS pain intensity as a continuous variable.

Data availability

The data associated with this study are not publicly available due to ethical and privacy restrictions but are available from the corresponding author on reasonable request.

Results

Of the 77 patients enrolled at baseline (T0), eight were lost to follow-up between T0 and T1, three between T1 and T2, six between T2 and T3, and five between T3 and T4. In total, 55 patients completed the 12-month assessment.

Of those who dropped out of the study, seven died, one moved to another city, and 14 withdrew because they found the follow-up too burdensome.

During follow-up, some patients received additional treatments for shoulder pain as clinically indicated. Details of the interventions performed at different time points are provided in the Supplementary Digital Material 1 (Supplementary Table I). All patients underwent standard rehabilitation programs, including shoulder mobilization and therapeutic exercises, throughout the follow-up period.

Temporal trajectory of HSP

At baseline (T0), HSP was minimal, with a mean NRS of 0.95 and a median value of 0.00.

Compared with baseline, pain increased significantly at T1 (mean 2.54, median 1) and further rose at T2, where the highest values were observed (mean 3.98, median 5). Subsequently, pain levels decreased at T3 (mean 3.40, median 2.5) and showed a further reduction at T4 (mean 2.73, median 2). Overall, the Friedman test confirmed a significant change in pain over time (χ²=29.4, P<0.001). Post-hoc Wilcoxon tests with Holm correction showed that pain levels at all follow-up time points were significantly higher than baseline (all adjusted P<0.001).

At T0, 11 patients presented with an NRS score ≥4 (prevalence 14.3%). The number of newly incident patients with an NRS score ≥4 since the previous time point was 17 at T1, 13 at T2, 5 at T3, and 1 at T4. Conversely, the number of patients whose NRS score passed from ≥4 to less than 4 was 0 at T1, 2 at T2, 11 at T3, and 6 at T4. This resulted in a steady rise in HSP prevalence at T1 (39.1%) and T2 (57.6%), with a subsequent progressive decline in prevalence at T3 (46.7%) and T4 (34.5%).

The temporal trajectory of pain intensity is shown in Figure 1.

Figure 1.

—Distribution of NRS scores over time with HSP prevalence annotations.

Longitudinal predictors of hemiplegic shoulder pain across timepoints

Timepoint T0

In the univariable analysis, HSP was associated with reduced passive ROM and with glenohumeral subluxation. In the multivariable model, reduced passive ROM remained independently associated with pain, with the regression coefficient indicating that each 1% loss of passive ROM corresponded to an average increase of 0.034 points on the NRS pain scale. Glenohumeral subluxation also remained independently associated with higher pain intensity, corresponding to an adjusted increase of 3.53 points on the NRS scale. Baseline neurological severity (NIHSS) emerged as an additional independent predictor of pain after adjustment, with higher NIHSS scores being associated with lower pain intensity (β=-0.1152 per NIHSS point), while no significant association was observed in the univariable analysis (Figure 2).

Figure 2.

—Forest plot graph of predictive factors for shoulder pain at baseline (T0): A) results of the univariate regression analyses; B) results of the multivariate regression model. Vertical lines represent regression coefficients (β), horizontal lines indicate 95% confidence intervals, and the vertical dotted line indicates the null effect (β=0).

Timepoint T1

In the univariable analysis, HSP was associated with reduced passive ROM, glenohumeral subluxation, adhesive capsulitis, and tendon tears. In the multivariable model, reduced passive ROM remained independently associated with pain, with the regression coefficient indicating that each 1% loss of passive ROM corresponded to an average increase of 0.038 points on the NRS pain scale. Glenohumeral subluxation also remained independently associated with higher pain intensity, corresponding to an adjusted increase of 2.94 points on the NRS scale. Adhesive capsulitis likewise remained independently associated with pain, with an adjusted increase of 4.63 points on the NRS pain scale (Figure 3).

Figure 3.

—Forest plot graph of predictive factors for shoulder pain at T1: A) results of the univariate regression analyses; B) results of the multivariate regression model. Vertical lines represent regression coefficients (β), horizontal lines indicate 95% confidence intervals, and the vertical dotted line indicates the null effect (β=0).

Timepoint T2

In the univariable analysis, HSP was associated with reduced passive and active ROM, increased spasticity, decreased muscle strength, adhesive capsulitis, glenohumeral subluxation, and higher disability scores. In the multivariable model, reduced passive ROM remained independently associated with pain, with the regression coefficient indicating that each 1% loss of passive ROM corresponded to an average increase of 0.041 points on the NRS pain scale. Glenohumeral subluxation also remained independently associated with higher pain intensity, corresponding to an adjusted increase of 3.65 points on the NRS scale, and adhesive capsulitis likewise remained independently associated with pain, with an adjusted increase of 3.51 points on the NRS pain scale. Tendon tears emerged as an additional independent predictor of pain after adjustment, with each additional tendon tear corresponding to an average increase of 0.80 points on the NRS pain scale, while no significant association was observed in the univariable analysis. Younger age also emerged as an independent predictor of higher pain intensity after adjustment (Figure 4).

Figure 4.

—Forest plot graph of predictive factors for shoulder pain at T2: A) results of the univariate regression analyses; B) results of the multivariate regression model. Vertical lines represent regression coefficients (β), horizontal lines indicate 95% confidence intervals, and the vertical dotted line indicates the null effect (β=0).

Timepoint T3

In the univariable analysis, HSP was associated with higher disability scores, increased spasticity, reduced passive and active ROM, decreased muscle strength, adhesive capsulitis, glenohumeral subluxation, intra-articular effusion, tendon tears, and greater neurological severity. In the multivariable model, reduced passive ROM remained independently associated with pain, with the regression coefficient indicating that each 1% loss of passive ROM corresponded to an average increase of 0.12 points on the NRS pain scale. The number of tendons affected by tears also remained independently associated with pain, with each additional tendon tear corresponding to an average increase of 0.60 points on the NRS pain scale. Hemorrhagic stroke emerged as an additional independent predictor of pain after adjustment, corresponding to an adjusted increase of 0.95 points on the NRS pain scale (Figure 5).

Figure 5.

—Forest plot graph of predictive factors for shoulder pain at T3: A) results of the univariate regression analyses; B) results of the multivariate regression model. Vertical lines represent regression coefficients (β), horizontal lines indicate 95% confidence intervals, and the vertical dotted line indicates the null effect (β=0).

Timepoint T4

In the univariable analysis, HSP was associated with reduced passive and active ROM, increased spasticity, decreased muscle strength, adhesive capsulitis, glenohumeral subluxation, the presence of tendon tears, and higher disability scores. In the multivariable model, reduced passive ROM remained independently associated with pain, with the regression coefficient indicating that each 1% loss of passive ROM corresponded to an average increase of 0.081 points on the NRS pain scale. Glenohumeral subluxation also remained independently associated with higher pain intensity, corresponding to an adjusted increase of 1.79 points on the NRS scale. The number of tendons affected by tears likewise remained independently associated with pain, with each additional tendon tear corresponding to an average increase of 0.78 points on the NRS pain scale (Figure 6).

Figure 6.

—Forest plot graph of predictive factors for shoulder pain at T4: A) results of the univariate regression analyses; B) results of the multivariate regression model. Vertical lines represent regression coefficients (β), horizontal lines indicate 95% confidence intervals, and the vertical dotted line indicates the null effect (β=0).

Representative ultrasonographic examples of the main shoulder abnormalities observed during follow-up, including inferior capsular thickening, subacromial-subdeltoid bursitis, tendon tears, intra-articular effusion, and glenohumeral subluxation, are provided in the Supplementary Digital Material 2 (Supplementary Figure 1-5).

Discussion

In this cohort, evaluated longitudinally from the first 72 hours after stroke (T0), HSP exhibited a characteristic temporal trajectory, with a sharp increase during the first three months (T2), followed by a progressive decrease over time. This pattern is highly consistent with previous longitudinal studies and meta-analyses, which similarly report an early onset, a peak around the third-fourth month, and a subsequent decrease over time.^{5, 29-31} The later decline in prevalence does not necessarily reflect spontaneous remission but may instead result from the cumulative effect of standard rehabilitation and, in selected cases, targeted interventions.^{8, 32} Overall, these findings confirm that HSP is an early and dynamic complication of stroke, with its key determinants largely consolidating within the first trimester.

Among all clinical and ultrasound variables, passive ROM was the only factor consistently associated with shoulder pain across all timepoints (T0-T4). This finding highlights the central role of passive mobility in the pain experience after stroke, while also requiring careful interpretation of its underlying mechanisms. Because passive ROM was expressed on a 0-100% scale, the regression coefficient reflects the effect of minimal changes in mobility; when applied to clinically realistic ROM losses, the associated increase in pain becomes substantial. Thus, despite the small per-unit effect size, reduced passive mobility consistently translated into meaningful pain amplification throughout the post-stroke course.

In principle, reduced passive ROM may reflect muscle contracture or capsular restriction, conditions known to generate pain when shortened tissues are stretched. Such structural mechanisms could plausibly contribute to pain from T1 onward, when soft-tissue alterations have had sufficient time to develop. However, reduced passive ROM was already associated with pain at T0, within the first 72 hours after stroke, when true muscle shortening or capsular fibrosis is highly unlikely. This early association strongly suggests a pain-dependent measurement effect, whereby passive mobilization is prematurely interrupted because the examiner must stop the movement when discomfort arises. Importantly, this pain-related mechanism may also persist at later timepoints. Consequently, even beyond the acute phase, reduced passive ROM cannot be reliably interpreted as a purely causal determinant of pain, as it may still, at least in part, represent a consequence rather than a cause of shoulder pain.

When examining the determinants of HSP across the different phases of recovery, a coherent temporal sequence of mechanisms emerges. In the acute phase, within the first 72 hours after stroke, HSP was already independently associated with glenohumeral subluxation and neurological severity, as measured by the NIHSS. Acute stroke immediately produces paresis and muscle hypotonia, which represent the negative signs of the upper motor neuron syndrome and precede the later development of positive phenomena such as spasticity, co-contractions, and muscle spasms.³³ This early flaccid paresis markedly reduces muscular support around the shoulder, promoting glenohumeral subluxation.^34-36 The resulting traction and stretching of periarticular soft tissues provide a plausible explanation for the pain observed, in line with previous observations.^{18, 37} The independent association between pain and stroke severity further supports this interpretation, since more severe strokes typically produce more pronounced paresis and hypotonia, increasing susceptibility to glenohumeral subluxation.³⁵

At one month, HSP remains associated with glenohumeral subluxation, and adhesive capsulitis emerges as an additional contributor. At this stage, reduced joint use and prolonged immobilization play a central role in driving capsular involvement. Limited shoulder movement promotes synovial inflammation, collagen deposition, and progressive capsular stiffening, creating the biological substrate for capsular pain.³⁸ Ultrasonographic studies further support the link between reduced joint use, altered biomechanics, and early capsular restriction.³ Therefore, the one-month stage reflects the coexistence of mechanical pain due to glenohumeral instability and immobility-driven capsular changes.

At three months, HSP reached its peak, consistent with most longitudinal studies.^{24, 29} At this stage, glenohumeral subluxation and adhesive capsulitis remained independently associated with pain, indicating the persistence of both mechanical instability and capsular inflammation as major contributors.^{39, 40} Furthermore, tendon tears also emerged as an independent predictor. The temporal emergence of tendon-related pain is physiologically plausible. During the first three months after stroke, patients typically undergo substantial motor recovery and progressively increase their level of activity during intensive rehabilitation. In individuals with pre-existing degenerative changes and/or impaired scapulohumeral biomechanics, this sharp rise in activity may increase mechanical load on the rotator cuff and the LHBT, thereby converting subclinical tendinopathies into symptomatic tears or unmasking previously asymptomatic lesions. This interpretation is supported by several ultrasonographic and MRI studies showing that tendinopathies and tendon tears are frequently observed in patients with shoulder pain during the subacute and chronic phases.^{3, 39, 41}

The fact that tendon tears became significant only in the multivariable model suggests a suppression effect. At this stage, tendon pathology strongly covariates with passive ROM limitation and adhesive capsulitis, both of which are closely linked to pain. Once these factors are accounted for, the independent contribution of tendon tears to pain becomes apparent, consistent with the increased mechanical stress imposed on vulnerable tendons by capsular stiffness and painful motion restriction.

Spasticity was associated with shoulder pain in the univariable analysis. However, this association should be interpreted with caution. Previous evidence indicates that spasticity per se is not inherently painful; rather, pain emerges when increased muscle tone is accompanied by muscle shortening and reduced extensibility, reflecting the development of early contracture.⁴² This conceptual framework may help explain the loss of significance of spasticity in the multivariable model. At T2, passive ROM likely reflects, at least in part, the presence of soft-tissue shortening and early contracture. Once passive ROM is included in the model, it may capture the structural component responsible for pain, thereby absorbing the effect attributed to spasticity in unadjusted analyses.

Younger age was independently associated with higher pain intensity, despite not being associated with pain in the univariable analysis. This finding is consistent with previous reports and may reflect greater perceptual sensitivity and more responsive nociceptive afferents in younger individuals.¹⁷ In the presence of structural or biomechanical abnormalities, these factors could contribute to an amplification of pain perception, making age-related differences more apparent only after adjustment for concurrent mechanical and structural drivers.

At six months, the HSP profile shifts toward a predominantly tendon-driven pattern. In the multivariable model, tendon tears emerged as the main independent determinant of pain, while adhesive capsulitis and glenohumeral subluxation were no longer associated. The disappearance of the association with adhesive capsulitis likely reflects treatment effects rather than spontaneous resolution. In this cohort, most cases identified at three months underwent ultrasound-guided hydrodistension, a procedure associated with improvements in pain and shoulder ROM.^{43, 44} Similarly, the lack of association between glenohumeral subluxation and pain should be interpreted considering targeted management strategies, including rehabilitation, shoulder orthoses, and local injection therapies.^{9, 39, 45, 46} Hemorrhagic stroke also showed an independent association at this stage, a finding restricted to this timepoint and without a clear mechanistic explanation, warranting cautious interpretation.

At twelve months, tendon tears again emerged as a key independent predictor of HSP, consistent with the pattern already observed at three and six months. This reinforces the interpretation that tendon pathology represents the most stable and persistent driver of pain in the chronic phase.^{3, 39} Glenohumeral subluxation also reappeared as an independent predictor at twelve months. Its late recurrence cannot be conclusively interpreted based on the present data. The borderline association with female sex aligns with previous findings in upper-limb spasticity, where women reported higher pain intensity.⁴² Overall, the one-year profile indicates that tendon pathology becomes the predominant determinant of residual shoulder pain intensity.

Integrated perspective and preventive approach

HSP develops predominantly within the first three months after stroke, representing a critical window during which most new cases arise. Across this interval, our data reveal a clear temporal sequence of mechanisms: early mechanical pain related to glenohumeral subluxation, capsular pain associated with the onset of adhesive capsulitis, and subsequent tendon-related pain. This temporal profile supports a targeted and phase-specific approach to prevention.

In the acute phase, when the patient is admitted to the stroke unit, prevention should focus on minimizing mechanical traction on the paretic shoulder through careful handling and appropriate support to limit humeral head descent. At the same time, regular passive mobilization should be initiated to reduce the risk of capsular stiffening. These measures should continue into the subacute phase, when patients are transferred to intensive rehabilitation, where systematic shoulder ultrasound becomes essential for detecting early capsular involvement and pre-existing or emerging tendon abnormalities. If capsulitis develops within the first three months, physiotherapeutic mobilization may be complemented by capsular hydrodistension. Because tendon tears correlate with pain from the third month onward and are typically preceded by tendinopathy, early identification and treatment of tendon pathology, potentially including focused shockwave therapy, may help prevent progression to rupture. Moreover, in patients with tendon abnormalities, therapeutic exercise should be carefully supervised to preserve physiological scapulohumeral rhythm and avoid overload.

Compared with current guidelines, which focus mainly on safe handling, correct positioning, and gentle passive mobilization,⁴⁷ this framework adds a mechanism-based and temporally structured perspective. By aligning preventive actions with the sequential emergence of mechanical, capsular, and tendon-related mechanisms, and by integrating early ultrasound screening and targeted interventions, it offers a more proactive and pathophysiologically informed strategy for reducing the onset and chronicity of post-stroke shoulder pain.

Limitations of the study

This study has several limitations. Firstly, the observational design does not permit the establishment of causal relationships between the analyzed factors and post-stroke shoulder pain.

Second, there was participant attrition during the follow-up period, with 22 patients (approximately 29%) failing to complete the 12-month follow-up. This reduction in sample size at subsequent time points may have limited the statistical power to identify weaker associations.

Furthermore, although ultrasound assessments were performed according to a standardized protocol, a certain degree of operator dependency could not be completely ruled out.

Finally, as this was an observational study conducted in a real-world clinical setting, patients received usual care, including standard rehabilitation programmers and, when clinically indicated, pharmacological or injection therapies. While this approach increases the external validity of the results, it also implies that such interventions may have influenced the incidence and intensity of shoulder pain over time without their effects being systematically controlled.

Conclusions

HSP follows a distinct temporal evolution in which early mechanical instability and capsular changes progressively give way to tendon involvement, with tendon tears emerging as the most persistent correlation of pain in the chronic phase. By integrating longitudinal clinical and ultrasound assessments from the acute to the chronic stage, this study provides novel evidence that phase-specific mechanisms underlie HSP and that preventive and therapeutic approaches should be tailored accordingly over time.

Supplementary Digital Material 1

Supplementary Table I

Number of dropouts and additional treatments for hemiplegic shoulder pain across follow-up intervals. The table reports the number of patients lost to follow-up in each time window and the number of patients who received additional treatments for shoulder pain, including pain medication, local injections, and antispastic treatment, according to clinical indication. Pain medication includes analgesics and/or anti-inflammatory drugs prescribed for shoulder pain. Local injection refers to ultrasound-guided capsular hydrodistension or other infiltrative treatments performed for hemiplegic shoulder pain. Antispastic treatment includes oral antispastic medication or botulinum toxin injections targeting upper-limb muscles

Supplementary Digital Material 2

Supplementary Figure 1

Longitudinal 12–3 MHz US scans, obtained with the probe aligned parallel to the humerus in the axillary pouch. A normal inferior capsule (arrowheads) extending over the humeral head (HH) and draping along the humeral neck (HN); B evident thickening of the inferior capsule (arrowheads).

Supplementary Figure 2

Longitudinal 12-3 MHz US scan along the supraspinatus tendon shows increased thickening of the bursa walls (arrows) associated with hyperechoic content, consistent with chronic bursitis.

Supplementary Figure 3

Longitudinal 12–3 MHz US scan, shows a partial tear through the entire thickness of the pre-insertional portion of the supraspinatus tendon (arrows). Note the loss of the normal convexity of the peribursal fat at the site of the tear.

Supplementary Figure 4

Longitudinal 12-3 MHz US scan over the posterior shoulder demonstrates a hypoechoic effusion (arrow), distending the posterior gleno-humeral recess. This recess is located between the humeral head and the posterior aspect of bony glenoid, deep to the infraspinatus tendon and muscle.

Supplementary Figure 5

Longitudinal 12–3 MHz US scans, showing the Acromion–Greater Tuberosity Distance (AGTD) measurement (between cursors). Panel A demonstrates a clear increase in the acromion–greater tuberosity distance; panel B, normal distance.