The accuracy of plain radiographs in diagnosing degenerate rotator cuff disease

Abstract

Background

A number of radiographic signs have been previously demonstrated to be associated with degenerative rotator cuff tears. An ability to predict the presence of a tear by radiography would permit the early commencement of appropriate treatment and the avoidance of unnecessary invasive investigations. The aim of the present study was to determine the accuracy of using radiographic signs to predict the presence of a cuff tear on arthroscopy.

Methods

Fifty consecutive patients who had undergone shoulder arthroscopy and had pre-operative plain radiographs were included. Pre-operative radiographs were reviewed by a consultant shoulder surgeon, a consultant radiologist and a senior clinical fellow for the following signs: acromial spur; subjective reduction of subacromial space; sourcil sign; acromial acetabularization; os acromiale; greater tuberosity cortical irregularity; greater tuberosity sclerosis; humeral head rounding; cyst; and reduction in acromiohumeral head distance.

Results

The presence of tuberosity sclerosis (p < 0.0001), tuberosity irregularities (p < 0.0001), tuberosity cyst (p = 0.004) and sourcil sign (p = 0.019) was associated with the presence of a rotator cuff tear. The combined sensitivity of prediction of tear by the observers following radiographic review was 91.7%, with a combined negative predictive value of 80%.

Conclusions

The assessment of radiographs by senior clinicians is a useful tool for confirming the absence of a rotator cuff tear.

Introduction

Rotator cuff tears are amongst the most common of shoulder pathologies presenting to the orthopaedic clinic. They have an estimated prevalence of 22%, increasing with age.¹ The primary imaging modalities for the diagnosis of cuff tears are ultrasonography (US) and magnetic resonance imaging (MRI). Recent meta-analyses have demonstrated these modalities to exhibit high sensitivities and specificities, particularly for full-thickness tears.^2,3 US is relatively inexpensive and quick to perform, although it is very operator dependent. MRI also provides a detailed evaluation of surrounding structures, although it is costly and less readily accessible.

Plain radiography is typically the initial imaging modality for patients presenting to a shoulder clinic. An ability to predict rotator cuff disease on initial radiography may enable the early commencement of appropriate treatment and confer greater patient satisfaction, as well as minimize the need for costly secondary imaging. A number of radiographic signs have been suggested to be associated with postero–superior cuff tears (supraspinatus and infraspinatus), including an acromial spur, a reduced subacromial space, sourcil sign, acromial acetabularization, greater tuberosity irregularity, humeral head sclerosis, humeral head rounding, soft tissue calcification, an os acromiale and a reduced acromio–humeral head distance.^4–16 Cortical irregularity of the greater tuberosity has been demonstrated previously, in both radiographic and US studies, to be an indicator of cuff tear (high specificity, modest sensitivity).^14,17 Evidence of a positive association of greater tuberosity sclerosis with cuff tear has proved more inconsistent. In a study of arthroscopically proven cuff tears, Pearsall et al.¹⁰ demonstrated a significant association. This is in contrast to an earlier study by Huang et al.⁹ However, this earlier study had used MRI to determine to presence of a cuff tear, which is less sensitive than arthroscopy.

The aim of the present study was to determine the accuracy of plain radiographic findings with respect to predicting the presence of a rotator cuff tear, in isolation or in combination, by comparison with the gold standard of arthroscopy. Thus, the null hypothesis was that there is no relationship between the presence of radiographic signs and a rotator cuff tear.

Materials and methods

Cases were identified retrospectively at the Royal Free Hospital, London. The study was approved by the trust research governance board. Fifty consecutive patients, who had reached skeletal maturity and had undergone shoulder arthroscopy between January 2011 and December 2013, had been randomly selected to avoid selection bias. Inclusion criteria included the presence of pre-operative shoulder radiographs and the use of shoulder arthroscopy. Exclusion criteria included patients with first-time shoulder dislocation or acute rotator cuff tear because radiographs would be expected to be normal in these cases. Eligible patients were then allocated a specific patient research assignment number to ensure observer blindness. Of the 50 patients selected for the present study, 42 patients had presumptive diagnosis of rotator cuff disease, four had presumptive diagnosis of adhesive capsulitis and four had recurrent instability.

Standard true anteroposterior and lateral views of the shoulder were reviewed by three assessors: a musculoskeletal radiologist, a senior consultant shoulder surgeon and a senior clinical fellow. The assessments were then repeated after a 3-week interval. The images were reviewed on a digital picture archiving and communication system (PACS). Eleven parameters were studied, which can be largely divided into acromion related pathologies (acromial spur, subjective reduction of subacromial space, sourcil sign, acromial acetabularization, os acromiale) and humeral head pathologies (greater tuberosity cortical irregularity, greater tuberosity sclerosis, humeral head rounding, the presence of a cyst, reduction in acromio–humeral head distance) (Figs 1 and 2). The study parameters were scored using dichotomous outcome (presence or absence). The acromio–humeral distance was also measured and classified into dichotomous outcome (normal or reduced) using 6-mm as a cut-off in accordance with Goutallier et al.¹⁶ Following radiographic assessment, the assessors also made a pragmatic diagnosis of the presence or absence of a cuff tear. Operating notes and arthroscopic photographs were reviewed by a separate senior clinical fellow to determine the presence of a cuff tear. For the present study, we defined a positive postero–superior cuff tear as a full-thickness tear or a > 50% thickness tear because these represent significant cuff pathology.¹⁸

Figure 1.

An anteroposterior radiograph of a right shoulder demonstrating acromial sclerosis (black arrow).

Figure 2.

An axial radiograph of a right shoulder demonstrating greater tuberosity cortical irregularity (black arrow).

Statistical analysis

The data were analyzed using the SPSS, version 22 (IBM Corp., Armonk, NY, USA). Intra- and inter-observer variability was expressed using Cohen’s kappa coefficient. The predictions from all three observers were reported individually and the average prediction calculated. The sensitivity, specificity and prediction values for the average prediction were determined. Chi-squared tests were carried out to investigate the correlation between each parameter with the status of the cuff. Because these parameters often co-exist in one radiograph, relationships may not be evident in one-dimensional analysis. Hence, we carried out dimension reduction analysis to examine the distribution and grouping of these 11 covariates with cuff tear. The variables that were closely related were then examined further with categorical regression analysis and optimal scaling.

RESULTS

Baseline characteristics

The mean (SD) age of the study cohort was 51.8 (16.6) years (range 17 years to 81 years). Of the 50 patients, 24 (48%) had a full-thickness tear or a tear greater than 50%; two had a partial tear of < 50% (4%) and 24 had no tear (48%). The mean (SD) age was 57.9 (14.5) years (range 33 years to 81 years) for those patients with a full-thickness tear. The mean age of patients without a tear was 46.2 years (range 17 years to 71 years).

Radiographic parameters

One hundred percent (100%; 24/24) of patients with a > 50% postero–superior cuff tear had three or more radiographic signs compared to 23% (6/26) of those with a partial or no tear. Of the six patients who had more than three radiological signs, one had adhesive capsulitis, whereas the other five had combination of at least two of the following pathologies: partial articular side cuff tear of less than 50%, long head of biceps pathology or a subscapularis tear.

The chi-squared test demonstrated tuberosity sclerosis, tuberosity irregularities, tuberosity cyst and sourcil sign to be significantly associated with the presence of a rotator cuff tear (Table 1).

Table 1.

Correlation of cuff tear with radiographic parameters.

Cross-tabulation pairing	Pearson’s chi-squared significance
Cuff tear – spur	p = 0.616
Cuff tear – subacromial space reduction	p = 0.048
Cuff tear – Sourcil sign	p = 0.019
Cuff tear – acetabularization of acromion	Not computed because no radiograph had this sign
Cuff tear – tuberosity sclerosis	p < 0.0001
Cuff tear – tuberosity irregularities	p < 0.0001
Cuff tear – tuberosity cyst	p = 0.004
Cuff tear – rounding of humeral head	p = 0.048
Cuff tear – calcification	p = 0.917
Cuff tear – os-acromiale	p = 0.332
Tear versus reduced acromio–humeral distance	p = 0.026

We examined the relationship of the covariates with principal component analysis and confirmed that these four factors were commonly associated with cuff tear. The Kaiser–Meyer–Olkin measure of sample adequacy was 0.72 and Bartlett’s test of sphericity was significant (p < 0.0001). The rotated component matrix showed that humeral side radiological signs commonly co-exist with cuff pathology. The principal components that are grouped together with cuff pathology were tuberosity sclerosis, tuberosity irregularities and tuberosity cyst. The only acromion side radiological sign that was weakly related to cuff tear was the sourcil’s sign. This is illustrated in Fig. 3, which shows the grouping of cuff tear with tuberosity sclerosis, tuberosity irregularities, tuberosity cyst and sourcil sign.

Figure 3.

Graph showing grouping of cuff tear with tuberosity sclerosis, tuberosity irregularities, tuberosity cyst and sourcil sign (black dots). Key: surgtear = tear confirmed on arthroscopy; cystaverage = presence of cyst; scleraverage = presence of sclerosis; irregaverage = irregularities on greater tuberosities; sourcilaverage = sourcil sign.

Using the covariates identified from the principal component analysis, we carried out regression analysis to confirm the predictability of these covariates. Categorical regression analysis showed an r² value of 0.44. The model also confirmed that tuberosity irregularity has the biggest effect on predicting cuff tear (standardized co-efficient, β = 0.41), followed by tuberosity sclerosis (0.145) and tuberosity cyst (0.151) and, to a lesser extent, sourcil sign (0.043).

Reliability of prediction

The sensitivity, specificity, positive predictive value and negative predictive value of the prediction by the fellow were 62.5%, 50%, 53.6% and 59%, respectively (Table 2). The sensitivity, specificity positive predictive value and negative predictive value of the prediction by the MSK radiologist were 87.1%, 48.1%, 61.6% and 83.5%, respectively. The sensitivity, specificity positive predictive value and negative predictive value of the prediction by the shoulder surgeon were 81.3%, 40.4%, 57.4% and 70%, respectively.

Table 2.

Accuracy and reliability of prediction of cuff tear.

Assessor	Sensitivity	Specificity	Positive predictive value	Negative predictive value	Intra-observer reliability (Cohen’s kappa; SE)	Inter-observer reliability (Fleiss’ kappa)
Fellow	62.5%	50%	53.6%	59%		0.566 (all 3); 0.75 (radiologist versus consultant surgeon)
Musculoskeletal radiologist	87.1%	48.1%	61.6%	83.5%	0.716 (0.1)
Shoulder surgeon	81.3%	40.4%	57.4%	70%	0.714 (0.1)
Consensus prediction	91.7%	30.8%	55%	80%

The prediction was repeated twice by each assessor (except the fellow). We derived an average prediction by using the consensus of prediction for each patient from all three assessors. The average prediction was more reliable. The sensitivity, specificity, positive predictive value and negative predictive value were 91.7%, 30.8%, 55% and 80%, respectively. Repeat analysis using the cohort with higher prevalence (age > 30 years) did not change the reliability. The consensus prediction had a high sensitivity and high negative predictive value (i.e. good at ruling out cuff tear).

The intra-observer reliability was 0.7 (Cohen’s kappa; standardized error 0.1) and the inter-observer error between all three assessors was 0.566 (Fleiss’ kappa). The inter-observer reliability between the MSK radiologist and senor consultant was 0.75 (Fleiss’ kappa).

Discussion

The present study has demonstrated plain radiography to be useful in detecting and ruling out a rotator cuff tear, particularly when reviewed by a senior orthopaedic surgeon or musculoskeletal radiologist (combined sensitivity of 91.7% and negative predictive value of 80%). The specificities and positive predictive value are less satisfactory. This contrasts with the relatively high sensitivities exhibited by both US and MRI.^2,3 However the reliability of the prediction would undoubtedly have been enhanced when taken into context with history and clinical examination.

The results of the present study also showed that patients with more than three abnormal radiological signs are likely to have a cuff tear. The abnormal radiological signs on the humeral side are more predictive of cuff tear. These include sclerosis of the tuberosity, irregularities of the tuberosity and the presence of cystic changes on the tuberosity. The presence of sourcil sign is a weak predictor, although this may be the result of a type II error because sourcil sign is rare in our cohort of mainly middle age patients (mean age 51.8 years). The findings from our study will help the clinicians to target further investigation and treatment to the ‘at risk’ group and save cost for not investigating the ‘normal’ group. In addition, the ability to predict the diagnosis and response in the clinic will also improve patient satisfaction, particularly in clinics that do not provide a one-stop scanning service.

Several plain radiographic findings have been previously suggested to reflect the presence of a rotator cuff tear,^4–16 although their utility in the diagnosis of tears remains controversial. In 2000, Bonsell et al.¹⁹ observed a number of radiographic changes (including acromial cysts, acromial sclerosis and acromioclavicular joint narrowing) that were associated with increasing age in a study of asymptomatic individuals. However, the subjects in their study¹⁹ were not otherwise evaluated for rotator cuff pathology (tears are asymptomatic in approximately two-thirds of individuals).¹

Several studies have demonstrated a reduced acromio-humeral head distance to be associated with a rotator cuff tear.^13,16 Indeed, Saupe et al.¹³ demonstrated a significant negative correlation between acromio–humeral head distance and the size of the tear. This sign was not found to be significant in the present study. This may be because it is a relatively late sign of rotator cuff disease and often reflects cuff decompensation with progression to cuff tear arthropathy. These patients may be less likely to undergo arthroscopic treatment. It has been shown that a short acromio–humeral distance represents a reduced likelihood of reparative success.²⁰ Hence, it may not be useful in the management of a younger cohort with a reparable cuff tear.

The strengths of the present study include the blinded and repeated radiographic analysis that was performed. The study included 11 previously described radiographic findings, giving a more comprehensive assessment than previous studies. Thus, it could identify a combination that confers an increased risk of cuff tear. It also used arthroscopic determination of rotator cuff status, which is the gold standard for diagnosis.

The weaknesses of the present study include a relatively small sample size, with all patients requiring arthroscopy, which may cause a bias towards symptomatic disease and the possible presence of other shoulder pathologies, as well as patient suitability for arthroscopic intervention.

Conclusions

In conclusion, the assessment of plain radiography of the shoulder by senior practitioners is useful for confirming the absence of a rotator cuff tear. This supports the modality’s ongoing status as the first-line investigation for shoulder pathology. When combined with a careful history and clinical examination, it will help guide the use of further radiological investigations, obviating unnecessary waiting times and tests for many without cuff tears.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Ethical Review and Patient Consent

Not required for this article.