Physical therapists use several evaluation measures to identify the most important factors related to disability. However, the degree to which these evaluation components explain shoulder disability is not well known and that may detract clinicians from the best clinical reasoning.
ObjectiveTo determine how much evaluation components explain shoulder function.
MethodsEighty-one individuals with unilateral shoulder pain for at least four weeks and meeting clinical exam criteria to exclude cervical referred pain, adhesive capsulitis, and shoulder instability, participated in this study. Several typical clinical evaluation components were assessed as potential independent variables in a regression model using the Disabilities of the Arm, Shoulder, and Hand (DASH) score as a proxy to shoulder function. Two multivariate models were built to include (1) evaluation components from physical exam plus clinical history and (2) a model considering all previous variables and magnetic resonance imaging (MRI) data.
ResultsPain catastrophizing was the best variable in the model explaining at least 10% of the DASH variance. Sex and lower trapezius muscle strength explained considerably less of shoulder function. The MRI data did not improve the model performance.
ConclusionThe complexity of shoulder function is not independently explained by pathoanatomical abnormalities. Psychological aspects may explain more of shoulder function even when combined with physical components in some patients.
Patients with shoulder complaints are commonly assessed using magnetic resonance imaging (MRI) and patient self-reported measures such as pain intensity and duration of symptoms.1,2 Range of motion and muscle strength are also frequent physical measures used in the typical clinical evaluation.3–7 However, it is not clear how these clinical components combined explain shoulder function as no previous studies have comprehensively considered pain in combination with other evaluation components and imaging findings.
Previous studies have shown that pain intensity can negatively influence shoulder function, but the magnitude of this association has been variable across studies.6 Psychological aspects such as pain catastrophizing defined as a set of negative and exaggerated cognitive and emotional schema in response to actual or potential pain is also related to shoulder function.5,8 When combined with fear-avoidance, i.e., self-restriction of activities because of fear, pain catastrophizing explained up to 28% of shoulder function in a previous study.9 However, other studies reported that pain catastrophizing explained only 9%10 or even less11 of shoulder function as measured with the Shoulder Pain and Disability Index (SPADI) score.
While some recent studies have reported data regarding psychological outcomes, traditionally there is a strong influence from the biomedical model focusing on pathoanatomical findings as the cause of pain during clinical evaluation. It has been reported that large rotator cuff tears are associated with worse shoulder function, however, most of these studies used questionnaires like the Constant-Murley score with items that contribute greatly to the final score, thus even patients with mild physical deficits may exhibit low function scores.12 A systematic review6 identified studies describing a small association between muscle strength13 or range of motion14,15 with shoulder function while other studies have found significant associations but with varying magnitude of association.15–19
As previous studies did not explore shoulder function, taking into account several evaluation components together, more studies are needed to provide a better understanding of how a patient's history, clinical examination, and imaging findings explain shoulder function. The objective of this study was to determine how much some of the most typical evaluation components assessed during clinical consultation explain shoulder function as measured by the Disabilities of the Arm, Shoulder, and Hand (DASH) questionnaire.
MethodsParticipants and eligibility criteriaThis was a cross-sectional study in which participants were recruited using posts on local websites and printed flyers at the university and surrounding community from 2015 to 2016. All eligible participants had self-reported unilateral shoulder pain for at least four weeks. Individuals with bilateral pain, history of upper limb fractures or surgery, metallic implants in the head, thorax, or arms, shoulder instability, or history of recurrent shoulder dislocations, pseudoparalysis, or deficit in shoulder active range of motion, clinical signs of adhesive capsulitis20 or limitation in shoulder passive range of motion, self-reported neck pain, or fibromyalgia were excluded from the study.21 All individuals were evaluated by one physical therapist with five years of clinical experience treating patients with upper limb dysfunctions. This study was approved by the institutional review board of the Universidade Federal de São Carlos, São Carlos, SP, Brazil (protocol number 1.394.925) and all individuals signed a written consent before study enrollment. The strengthening the reporting of observational studies in epidemiology (STROBE) and the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) were followed to ensure the quality of the study.22,23
Self-reported upper extremity functionSelf-reported upper extremity disability was evaluated using the Brazilian version of the DASH questionnaire. The DASH is a self-reported questionnaire with 30 questions that assess the individual's ability to perform daily activities. Scores on the DASH can range from 0 to 100 with 0 best and 100 the worst possible score.24 The DASH is widely used to assess individuals with shoulder pain25–27 and demonstrates excellent reliability and responsiveness.28 Moreover, the DASH is a wide-ranging instrument and the most linked to International Classification of Functioning, Disability and Health categories in comparison to other common self-reported or composite patient-reported outcome measurements.29
Muscle strengthThe serratus anterior, lower trapezius, and infraspinatus muscle strength was evaluated. A detailed description and procedures for muscle strength testing can be found elsewhere.30,31 For familiarization, individuals performed 1 submaximal repetition of each test. Three 5-second repetitions of maximum isometric contractions for each test were averaged.32 The principal investigator gave standardized verbal encouragement to all individuals during muscle testing to facilitate maximal force production. Individuals repeated the test if any compensation with the trunk or legs occurred during the test performance.
Pathoanatomical findingsIndividuals underwent a standardized MRI examination scheduled after the physical examination with a gradient-echo in T2 and spin-echo sequences in T1, T2, and proton density to determine the presence of structural abnormalities. All scans included slices with 3.5 to 4.0 mm of thickness in sagittal, coronal, and axial planes without contrast material. A 1.5 Tesla-MRI device (Magnetom Essenza, Siemens®) with a dedicated shoulder array coil was used. MRI scans were interpreted by a board-certified orthopedic surgeon with 12 years of shoulder specialized experience after fellowship training. Pathoanatomical findings were grouped into inflammatory signs such as subacromial increased fluid or acromioclavicular joint osteoarthritis and rotator cuff tears such as partial- or full-thickness tear, defined as the presence of discontinuity of the tendon along its superior (bursal) or inferior (articular or undersurface) surface with an extraarticular fluid-filled gap (T2-weighted) and the presence of discontinuity of the tendon with a fluid-filled gap that extends from the articular to bursal surface observed mainly in T2-weightned sequences, respectively.33–35
Pain catastrophizing and demographicsPain catastrophizing was assessed with the Brazilian version of the Pain Catastrophizing Scale (PCS). The PCS contains 13 statements related to thoughts and feelings that represent pain catastrophizing and its underlying constructs such as pain magnification, helplessness, and rumination.36 This scale ranges from 0 to 52 with 0 as the best and 52 as the worst possible score. The PCS exhibits adequate construct validity and reliability. Age, sex, duration of the symptoms, body mass index, and presence of scapular dyskinesis using the scapular dyskinesis test37,38 were recorded for all participants.
Statistical analysisA linear regression analysis was used to verify how much evaluation components explain shoulder function. First, we grouped typical evaluation components in eight categories: a) demographics, b) range of motion, c) pain, d) pain catastrophizing, e) exposure, f) muscle strength, g) special tests and scapular dyskinesis, and h) MRI. Second, we verified the correlation between each evaluation component within all categories and the DASH. Third, the most correlated clinical component from each category was selected to be tested in the regression model. (Supplementary material). All clinical components tested as potential independent variables showed some individual contribution to shoulder function in previous studies.6,16,39–42 The dependent variable was the DASH questionnaire.
Multicollinearity was verified before running the regression analysis by identifying highly correlated explanatory variables (r ≥ 0.7). Multicollinearity between categorical explanatory variables was assessed by Phi and Cramer's V statistics and Chi-square test for independence. Collinearity was identified as a problem in the model by verifying condition index values greater than 30 and variance inflation factor (VIF) greater than 10 in the collinearity diagnosis table.43 Linearity and outliers were also evaluated by looking at bivariate scatterplots and partial plots for each potential independent variable and the dependent variable. After these steps, serratus anterior strength and variables from pain category were considered unfit to the model due to a non-linear relationship to the dependent variable and the presence of important outliers. Lastly, the body mass index presented a very high condition index value and we decided to remove this variable from the model.44,45
Two models were tested. In the first model, only variables related to clinical history and physical examination were included as independent variables. In the second model, all variables from the first model and MRI variables were used as independent variables. The performance between the two models was compared using the adjusted total explained variance (adjusted-R2) and Akaike Information Criterion (AIC). The model with the highest adjusted-R2 and the smallest AIC was deemed to best explain the DASH variance.46–48 The estimated sample size for adequate power in multivariate models may vary between 8 and 10 individuals per predictor or explanatory variable.49,50 IBM SPSS Statistics, version 23 (IBM Corp, Armonk, NY) was used to perform all statistical analyses.
ResultsEighty-one individuals completed the study. Four individuals did not complete the MRI examination due to claustrophobia. Most participants were not involved in sports practice or physically demanding jobs and did not exhibit a well-distributed duration of symptoms. Most participants reported low pain intensity during arm elevation in the assessment day, but the average pain during the week was high. Therefore, no common gesture or maneuver was reported among the subjects, which is representative of the variability in the clinical setting. Additional participants' characteristics are presented in Table 1.
Participants characteristics (n = 81).
The first model without MRI data explained 37% of the DASH variance. Only PCS, lower trapezius strength, and sex showed statistical significance to the model explaining 12%, 4%, and 4% of the DASH variance, respectively. Sex and lower trapezius strength were inversely related to the DASH (Table 2). The independent variables from the MRI in the final model did not substantively change the model R2 (35%) and the AIC was smaller for the first model suggesting that the first model had the best performance (Table 2).
Multiple regression.
AIC, Akaike Information Criterion; CI, confidence interval; Std, Standard; Part2, squared semipartial correlation; *, statistical significance.
Our results indicated that PCS explained at least 10% of the DASH variance. In other words, a PCS score variation of 30 points, which classifies an individual as a “catastrophizer”,51 may influence up to 14 points change on DASH. Sex was the second most important independent variable in the model. In general, women exhibited approximately 9 points less than men on DASH, but there is low confidence in this result because of the wide confidence interval (Table 2). The lower trapezius strength and sex were less important to the model explaining less of the DASH variance. In summary, the higher the pain catastrophizing, the worse the upper-extremity disability with little to no influence from sex or muscle strength, and that was not different in shoulders with signs of inflammation or in the presence of rotator cuff tears of any severity. We believe that PCS was the most important variable in the model because the DASH questionnaire incorporates more activities and body functions in comparison to other instruments.
The observed relationship between psychological variables and shoulder function is inconsistent.9,10,52,53 Kromer et al.10 reported no relationship between pain catastrophizing and shoulder function but that may be due to the low PCS scores (median = 9) of the sample. Consistent with our results, Coronado et al.9 used an hierarchical multivariate analysis running several models and observed that higher levels of pain catastrophizing indicated worse shoulder function. Differences between our studies regarding how much of shoulder function is explained by pain catastrophizing may be due to how multivariate models were built (entry versus hierarchical). Also, the performance and accuracy of regression models may differ depending on the sample characteristics. Other studies exhibited a poor model performance52 or a small contribution from psychological variables to shoulder function.53
The two MRI variables used in our study were grouped into two of the most prevalent categories clinicians use in practice. Usually, when focusing on pathoanatomical abnormalities in a typical clinical setting, clinicians deal mostly with inflammatory conditions or rotator cuff tears.54,55 Therefore, we chose to group MRI data because of the high variability when analyzing unique tendon prevalence and tear types. Although that approach may conceal these details, grouping tendon data would most likely increase the contribution of MRI variables to the model and that did not happen. We and others have provided information suggesting MRI may be overused in the early management of patients with non-specific shoulder pain.12,56–59 Imaging abnormalities are still important to clinical decision-making and prognosis in patients with shoulder pain as long as this information is judiciously considered with other components of clinical evaluation.
This study is not without limitations. We performed a pre-selection of independent variables instead of running a full model with all variables at the same time. Pre-selecting variables is one of the most popular practices used in regression analysis studies, but that might let relevant variables out of the model and compromise the external validity of the results. Variables such as employment status, education level, smoking habit, and additional yellow flags such as kinesiophobia and anxiety were not assessed. Scapular dyskinesis has been considered related to shoulder function and pain but there is conflicting evidence.60–62 We believe the high variability for scapular dyskinesis that is typically observed in clinical evaluation limited the contribution of this component in the multivariate model, which also explains the inconsistency in the literature. The level of disability of the participants was lower than expected, but we believe the data are still representative as all individuals reported important average pain intensity in their last week. The association between pain and shoulder function may not follow a linear trend depending on the patient's characteristic and other clinical components interaction. Despite the study's limitations, a substantial portion of the DASH variance was explained by the retained variables, especially by pain catastrophizing. That highlights how other important components of the shoulder clinical evaluation may be overlooked in patients presenting non-specific shoulder pain in the general setting.
ConclusionPain catastrophizing was the clinical component that mostly explained shoulder function in our multivariate model. Sex and lower trapezius muscle strength exhibited marginal and inconsistent influence in the model. Shoulder inflammatory status and the presence of rotator cuff tear did not explain shoulder function. Our results suggested that a more holistic evaluation approach may be beneficial to understand how the various aspects of clinical presentation explain upper-extremity disability.
This research was conducted with financial support from the National Council for Scientific and Technological Development (CNPq, process number 302789/2017-0) and financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001.
