Target condition being diagnosed
The rotator cuff is composed of the subscapularis, supraspinatus, infraspinatus and teres minor tendons; the long head of the biceps tendon also contributes to the cuff. The role of the rotator cuff is to stabilise the humeral head into the glenoid cavity, preventing the upward migration of the humeral head. The four muscles are recruited during different arm movements. The subscapularis is recruited in internal rotation, the supraspinatus in elevation, and the infraspinatus and teres minor in external rotation (Clark 1992; Favard 2007; Matsen 2008).
Rotator cuff tendinopathy can lead to progressive failure of the rotator cuff, typically progressing from partial to a full thickness tear of the supraspinatus tendon then extending into the infraspinatus tendon or the subscapularis tendon, or both. A review by Lewis 2009 concluded that the pathoaetiology of rotator cuff tears is multifactorial and that tears are correlated with a combination of extrinsic and intrinsic factors, but that more research is necessary to fully understand the aetiology of rotator cuff tears. The extrinsic factors (i.e. those external to the rotator cuff) can be divided into anatomical factors, such as the shape of the acromion (i.e. curved or hooked) and coracoacromial ligament, os acromiale and acromial spurs (Baring 2007; Bigliani 1991; Lewis 2009; Neer 1972; Neer 1983; Nho 2008), and environmental factors including aging, shoulder overuse, smoking, obesity and some metabolic disorders such as diabetes (Chen 2003; Galatz 2006; Harryman 2003; Lewis 2009; Nho 2008; Wendelboe 2004). The intrinsic factors include, among others, repetitive microtrauma, areas of hypoperfusion in the tendons, inflammation and cellular changes in the tendons such as disorganisation of the architecture of collagen (Biberthaler 2003; Levy 2008; Lewis 2009; Nirschl 1989; Rees 2008).
Shoulder pain is very common, with an incidence of 9.5 per 1000 patients in primary care in Cambridge, UK, where amongst them 85% presented with rotator cuff tendinopathy (Ostör 2005). Disorders of the rotator cuff tendons due to either wear or tear are among the most common causes of shoulder pain and disability. In Japan, the prevalence of rotator cuff tears is 20.7% in the general population and 36% in patients with shoulder pain (Yamamoto 2010). More than 4.5 million physician visits occurred and approximately 40,000 inpatient surgeries were performed for rotator cuff problems in the United States in 2002 (Oh 2007).
The diagnosis of rotator cuff tears is mainly based on the patient's history and physical examination. The value of physical examination of the shoulder has been addressed in another Cochrane review (Hanchard 2013). The clinical manifestations vary widely (Duckworth 1999; Matsen 2008). Acute, traumatic full thickness cuff tears may present with sudden onset of weakness during elevation of the arm after a trauma in which the arm has been forced to the side (like a fall with the arm out to the side or on catching a heavy falling object with the arm extended) (Matsen 2008). Chronic degenerative cuff defects may present with progressive pain and weakness, with concomitant loss of active motion. Pain in the lateral area of the shoulder is commonly present at night. Passive motion initially remains full until the pain limits active motion (Baring 2007; Matsen 2008). However, there are many people with degenerative rotator tears who are asymptomatic (Reilly 2006; Zanetti 2000).
Decisions about whether to order a diagnostic test include consideration of whether the results are likely to affect treatment. Plain radiographs of the shoulder may be useful to differentiate rotator tears from osteoarthritis of the glenohumeral or acromioclavicular joints and calcific tendonitis. Ultrasonography (US), magnetic resonance imaging (MRI) and magnetic resonance arthrography (MRA) are increasingly being used to detect rotator cuff tears, although who orders these tests may vary by setting. In some settings, these tests are mainly ordered by specialists but in other settings they are being ordered by primary care physicians or clinicians (Al-Shawi 2008; Miller 2008). In the context of specialist care, US or MRI, or both, are usually performed to determine the characteristics of the rotator cuff tears in order to plan surgery. In some settings, however, there has been a significant rise in the number of diagnostic US being performed in primary care. For example, in Australia there has been a more than fourfold increase, from 104,252 in the year 2000 to 2001 to 440,172 in 2008 to 2009 (Medicare Australia 2010). However, the utility of the test to affect treatment in primary care is unknown.
Tears of the rotator cuff can be classified in several ways: duration (acute or chronic), aetiology (traumatic or degenerative) or size (partial or full thickness). Radiologists often describe the size of tear in millimetres or centimetres or descriptively as small, medium, large or massive. All three factors (duration, aetiology and size) influence treatment decisions (Kuhn 2007).
Acute full thickness rotator cuff tears are uncommon and account for less than 10% of all rotator cuff tears. People with acute full thickness tears usually present with a history of acute trauma, such as a fall or dislocation, and immediate pain and weakness. Prompt surgical treatment, ideally within six weeks, is the recommended treatment (Rees 2008). For all other full thickness rotator cuff tears, surgical treatment is usually reserved for those who fail to improve after a period of conservative treatment, although the most effective surgical intervention and its timing remain uncertain (Coghlan 2009; Dunn 2005; Oh 2007; Rees 2008). For example, a delay in surgical repair of a large tear may allow the injured tendon to retract and the muscle to atrophy (Matsen 2008; Oh 2007). On the other hand, asymptomatic tears are common; these are chronic tears that normally do not compromise the function of the shoulder. A recent review reported the prevalence of full thickness tears in 2553 unselected cadavers as 30% (Rees 2008). Furthermore, the pathogenesis and progression to symptomatic tears remains unclear (Rees 2008). In addition, in contrast to acute full thickness tears, symptoms due to acute or chronic partial thickness cuff tears frequently improve with conservative interventions (Matava 2005; Matsen 2008).
While spontaneous healing of a partial thickness tear is unlikely in most cases, the explanation for the ‘cure’ with conservative treatment is due to the likely resolution of the accompanying inflammation over time and may also be related to the residual cuff muscles compensating for the mechanical deficiency of the torn cuff (Fukuda 1996; Fukuda 2003; Matava 2005; Matsen 2008). As with full thickness tears, no simple treatment algorithm for partial thickness rotator cuff tears exists. Surgical treatment, however, is normally indicated for people with persisting symptoms despite conservative treatment and in whom imaging suggests the presence of a partial thickness tear or tears. The ideal timing of surgical intervention also remains unclear (Fukuda 2003; Matava 2005). However, case series and anecdotal evidence suggest that satisfactory results are usually achieved with surgery provided there is a good blood supply to the tendon, contact between the torn ends, absence of retraction and adequate trophic quality of the muscle (Fukuda 2003).
Another recognised category of tears is massive complete tears, in which a large area of the humeral head is uncovered (Wolfgang 1974). Post 1983 defined a massive tear as greater than 5 cm. These tears, which are difficult to repair, are more commonly found in women over 65 years of age and are associated with advanced atrophy, degeneration and progressive fatty infiltration of the rotator cuff muscles (Dines 2007; Gerber 2000). Treatment options for these massive, retracted tears are limited as they are often deemed irreparable. In younger people, tendon transfers may be considered (Neri 2009).
The indications for surgical treatment of rotator cuff tears have not been fully defined. A systematic review of surgical treatment for rotator cuff disease (including tears), which included 14 trials, was unable to draw firm conclusions about the effectiveness of surgery (Coghlan 2009). Nonetheless, the review suggested that there were no significant differences in outcomes between open or arthroscopic surgery and non-surgical treatment (Coghlan 2009). Many studies have demonstrated that the size of the tear is correlated to the final outcome; partial or small full thickness tears usually have a satisfactory surgical result (Bianchi 2005; Bryant 2002; Fotiadou 2008).
Currently, US, MRI or MRA are usually performed in patients contemplating surgery for rotator cuff tears to determine the characteristics of the tears. With the improvement of technology, the accuracy of these imaging tests is considered to have improved significantly over time, enabling useful assessment of the size and extent of the rotator cuff tear when planning surgery (Rees 2008).
US is a diagnostic imaging technique used to visualise deep structures of the body by recording the echoes of pulsed ultrasonic waves directed into the tissues and reflected by tissue planes to the transducer. These echoes are converted into 'pictures' of the tissues under examination. Seltzer 1979 was the first to describe ultrasonographic evaluation of rotator cuff diseases. US of the shoulder is utilised in secondary, tertiary and, increasingly, primary healthcare settings to evaluate the integrity of the rotator cuff. It consists of a non-invasive examination that has practically no adverse effects and allows dynamic visualisation of the tendons during movement of the shoulder (Al-Shawi 2008). However, operator dependence and a long learning curve are frequently considered to be its limitation (O'Connor 2005; Rutten 2006), principally in view of partial thickness tears for which Le Corroller 2008 described a high interobserver variability.
MRI uses a powerful magnetic field to align the hydrogen atoms of water and other molecules in the body. Pulses of radiofrequency are applied which excite the magnetised atoms. These movements of hydrogen atoms, which vary in different tissues, are captured and the signal can be manipulated to build up an image of the body (Witte 2003). The first article about the use of MRI in the shoulder was published in 1986 (Kneeland 1986). Since then, this technique has been widely used in secondary and tertiary healthcare practice. MRI is a non-invasive method of imaging that is unique in allowing high resolution images in multiple planes. It is a static examination that may be enhanced by an intra-articular injection of radiopaque dye (this is called magnetic resonance arthrography). The radiopaque dye acts as contrast material that helps to delineate intra-articular structures and outline abnormalities. MRA of the shoulder is also useful for assessing the rotator cuff integrity. In comparison with conventional MRI, MRA may improve diagnostic performance in detecting shoulder diseases; however, any potential benefit from MRA must be weighed against the invasiveness and additional discomfort caused by the procedure.
MRI and MRA have some absolute contraindications, such as the presence of intracerebral aneurysm clips, cardiac pacemakers, automatic defibrillators, biostimulators, implanted infusion devices, cochlear implants and metallic orbital foreign bodies (Witte 2003). They are also expensive and time consuming procedures. The strength of the magnet, the sequences used in the examinations and the person (e.g. consultant radiologist, musculoskeletal radiologist or trainee) interpreting and reporting the test may all affect the results.
Summary of diagnostic pathway
The evaluation of patients with suspected rotator cuff tear(s) should initiate with a full history of the patient's complaints and a thorough clinical examination of the shoulder. Decisions for using an imaging diagnostic test may be supported by whether the results are likely to affect treatment. For example, MRI, MRA or US might confirm a possible full thickness tear. The three index tests considered can also be used as triage tests in people suspected of having partial thickness tears. People whose tests were positive can be treated as having partial tears, while people with rotator cuff symptoms whose tests were negative can undergo further diagnostic procedures, such as diagnostic arthroscopy.
The reference tests for diagnosis of rotator cuff tears are invasive. The most common reference test is diagnostic arthroscopy. Arthroscopy is a minimally invasive surgical procedure that involves insertion of an arthroscope, a type of fibre-optic endoscope, into the joint through a small incision. This allows the surgeon to inspect and probe the articular (joint) and bursal side of the rotator cuff tendons, to assess accurately the rotator cuff insertion (footprint) and to perform a general examination of the shoulder joint in order to identify and treat other potential lesions (Dinnes 2003; Matava 2005). However, limitations associated with diagnostic shoulder arthroscopy include the need for anaesthesia, hospital admission and some interobserver variation in the classification of tears (Kuhn 2007).
Open surgery (including mini-open) has also been used as a reference test although it is more limited than arthroscopy because joint surface or inferior surface tears are difficult to access and identify using an open approach. Thus open surgery is less accurate than arthroscopy for detecting partial rotator cuff tears.
US, MRI and MRA are increasingly being used to assess the presence and size of rotator cuff tears to assist in planning surgical treatment. Improved techniques have resulted in increased reliance on these tests, in place of a separate diagnostic arthroscopy, although arthroscopic examination of the shoulder joint is still commonly performed as part of surgical treatment. US, MRI and MRA are operator and reader dependent. It is not known whether any one test is superior to either of the two others or whether performing US and MRI or US and MRA enhances their value (Swen 1999). It is also not known whether these diagnostic tests provide useful additional information compared with diagnostic arthroscopy, which is an accepted part of the surgical treatment. While, the units costs of MRI and MRA are greater than US, the cost-effectiveness of the three tests has not been determined.
We identified two relevant systematic reviews with meta-analyses that assessed diagnostic imaging tests for rotator cuff disease (De Jesus 2009; Dinnes 2003). The literature search in both reviews was restricted to English language only. Dinnes 2003 evaluated the diagnostic accuracy of clinical testing of US, MRI and MRA for detecting rotator cuff tears using both surgical and non-surgical tests as the reference standard. The authors included 38 studies that assessed the accuracy of US, 29 studies that assessed the accuracy of MRI and 6 studies that assessed the accuracy of MRA and concluded that US or MRI were equivalent for detecting full thickness rotator cuff tears, although MRI was more expensive and US may be better at detecting partial tears. The search date for the review was October 2001. A later review with a search date in September 2007, De Jesus 2009, conducted a meta-analysis comparing the diagnostic accuracy of US and MRI for rotator cuff tears using surgery as the reference standard. This systematic review included 65 studies but the appraisal of the methodological quality of the included studies was unclear or inadequate. De Jesus 2009 concluded that US is as accurate as MRI for both full and partial thickness rotator cuff tears and also suggested that US may be the most cost-effective imaging test for detecting rotator cuff tears.
Important technological improvements in US, MRI and MRA have been made since the search dates of both systematic reviews, and new studies evaluating US, MRI and MRA have been published. Our review involves an updated search for diagnostic accuracy studies for rotator cuff tears and will not be restricted to English language publications.
To compare the diagnostic test accuracy of magnetic resonance imaging (MRI), magnetic resonance arthrography (MRA) and ultrasonography (US) for detecting any rotator cuff tears (i.e. partial or full thickness) in people with shoulder pain for whom surgery is being considered.
We divided our objectives as follows.
- To compare the diagnostic accuracy of US, MRI and MRA for diagnosing any rotator cuff tears (partial or full thickness)
- To compare the diagnostic accuracy of US, MRI and MRA for diagnosing full thickness rotator cuff tears (one or more tendons)
- To compare the diagnostic accuracy of US, MRI and MRA for diagnosing partial thickness rotator cuff tears
Investigation of sources of heterogeneity
We planned to investigate the following potential sources of heterogeneity:
- Type of tear: acute traumatic and chronic degenerative
- Type of reference standard: open (including mini-open) surgery or arthroscopy
Criteria for considering studies for this review
Types of studies
All diagnostic accuracy studies that compared one or more of the index tests with one or both of the reference tests in patients suspected of having a partial or full thickness rotator cuff tear were included. We only included results from full reports of prospective studies. Studies with an excessively long period of time (i.e. a year or longer) between the index and reference tests were excluded because there is evidence that rotator cuff tears can progress over time (Mall 2010; Melis 2010); however, the rate of progression is not clearly defined.
We included articles in English and languages for which a full translation could be obtained. Non-English articles where a full translation could not be obtained are cited in the Characteristics of studies awaiting classification but not included in the review.
For studies reported in multiple publications, we included only the most recent or complete report. References to the other publications were cited under the same study identifier.
We included people with shoulder pain suspected of having a rotator cuff tear for whom surgery was being considered. Studies that included healthy controls or participants who had been previously diagnosed with other specific shoulder pain (e.g. shoulder instability, osteoarthritis, rheumatoid arthritis, frozen shoulder, benign or malignant tumours or referred pain) were excluded. Studies that included participants with shoulder pain, but in which it was unclear if all the participants were suspected of having rotator cuff tears, were also excluded.
Studies that assessed the accuracy of US, MRI or MRA were included.
We included studies that evaluated the index tests for detection of at least one of three target conditions:
- presence of any rotator cuff tears (partial or full thickness);
- presence of full thickness tears;
- presence of partial thickness tears.
To standardise classification for this review, rotator cuff tears were dichotomised as absence or presence of any, full and partial thickness tears.
We required arthroscopy or open (including mini-open) surgery findings to be the reference standards.
Search methods for identification of studies
We searched relevant computerised databases for eligible diagnostic studies: MEDLINE (PubMed) (1966 to March 2011), EMBASE (Elsevier) (1980 to February 2011), LILACS (Bireme) (1982 to February 2011) and the Cochrane Register of Diagnostic Test Accuracy Studies (February 2011). We also searched DARE (Database of Abstracts of Reviews of Effects), the HTA Database (Health Technology Assessments Database) and the MEDION database (February 2011) for other related diagnostic test accuracy reviews, and we checked the reference lists of those reviews that were relevant for additional studies. We also searched the US Health Services Research Projects in Progress and the UK Clinical Research Network Portfolio Database for ongoing and recently completed studies. When possible, non-English articles were assessed through translation by a native speaker.
We used a sensitive search strategy as recommended by the Cochrane Collaboration for MEDLINE (PubMed), EMBASE (Elsevier) and LILACS (Bireme) (De Vet 2008). See Appendix 1 for the MEDLINE and EMBASE search strategies.
Searching other resources
We checked the reference lists of articles, reviews and textbooks for relevant primary diagnostic studies and systematic reviews. We handsearched abstracts of the British Elbow and Shoulder Society annual meetings (2005 to July 2011) and American Academy of Orthopaedic Surgeons annual meetings (2005 to July 2011). We also contacted experts in the field.
Data collection and analysis
We used the methods suggested in the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy (Deeks 2009).
Selection of studies
Two review authors (ML and RJ) independently screened the titles and abstracts of retrieved records to identify potentially relevant studies for inclusion. Duplicates were removed and the remaining references were examined. Studies which clearly did not meet the inclusion criteria were excluded, and copies of the full text of potentially relevant references were obtained. ML and RJ independently assessed full text reports and determined inclusion or exclusion of the studies. Any uncertainties or disagreements were resolved by discussion and, when necessary, by adjudication from a third author (RB).
Data extraction and management
Two review authors (ML and RJ) independently collected the available data using a piloted data extraction form without masking of study authors or other identifying information. A third review author (RB) was consulted for resolution of any disagreements. When necessary, we sent requests to study authors for additional information or data. Diagnostic accuracy studies that reported insufficient data for construction of two-by-two tables were excluded from the review.
We retrieved the following data.
- General information: title, journal, year, publication status, country of study, period of study, primary objective and study design (i.e. prospective versus retrospective and consecutive versus non-consecutive).
- Sample size: number of participants meeting the criteria and total number screened.
- Baseline characteristics: baseline diagnosis, age, sex, dominant arm, nature of onset (e.g. traumatic or non-traumatic), duration of symptoms, prior treatment, inclusion and exclusion criteria.
- Target condition as reported.
- Index test: description of technique, criteria for positive result, timing of test and expertise of the clinician or technician performing the test.
- Reference standard test: description of technique, criteria for positive result, time from index to reference test and expertise of the clinician or technician performing the test.
- Adverse effects or complications due to index test(s) and reference standard test(s).
- Number of true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN). We extracted data for operational definitions for category of tear (e.g. partial, full or any thickness tears). Multiple outcome categories are often reported for rotator cuff tears: partial thickness tear, full thickness tear and no tears (i.e. three-by-three tables). Currently available methods for evaluating diagnostic tests rely on dichotomised disease status. Therefore, for the assessment of each target condition, we dichotomised rotator cuff tears using a strategy based on the options for treatment. To create two-by-two tables for partial thickness tears, data for full thickness tears were included with those for no tears. We did not exclude data for any category. We included data for partial thickness tears with those for full thickness tears to create two-by-two tables for any tears.
Assessment of methodological quality
The methodological quality of the included studies was assessed independently by two review authors (ML and RJ) and disagreement on study quality was resolved by a third review author (RB). At the same time as data extraction, the methodological quality of selected studies was assessed using a modified version of the QUADAS checklist (Whiting 2003), following the guidelines provided in Chapter 9 of the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy (Reitsma 2009). Appendix 2 explains how we applied the QUADAS items for assessing the included studies.
Statistical analysis and data synthesis
Our unit of analysis was the shoulder. For each test and target condition, estimates of sensitivity and specificity from each study were plotted in receiver operating characteristic (ROC) space and forest plots for visual examination of variation in test accuracy. Where adequate data were available, we conducted meta-analyses using the bivariate model (Chu 2006; Reitsma 2005). In the bivariate model, the logit-transformed sensitivities and specificities, and the correlation between them across studies are modelled directly. The model accounts for sampling variability within studies and also accounts for between study variability through the inclusion of random-effects. In preliminary meta-analyses for each target condition, we fitted the bivariate model separately for each test. We examined the variance of the random-effects parameters to consider the magnitude of heterogeneity and to judge whether there were differences in heterogeneity in sensitivities and specificities between tests, before comparing the tests in a single model for formal assessment of comparative accuracy.
Comparative accuracy studies are scarce (Takwoingi 2013). Therefore, whenever possible, we included all studies of US, MRI and MRA (i.e. an indirect comparison) in the main comparative meta-analysis for each target condition. Due to few studies of MRA and considerable heterogeneity in study results, we only performed pairwise comparisons of MRI and US for detection of partial thickness tears and any tears but compared the three tests for detection of full thickness tears. We compared test accuracy by adding covariate terms for test type to the parameters of the bivariate model to determine which test was superior in terms of sensitivity or specificity or both. The variance coefficients from the preliminary meta-analysis and summary ROC plot for each test indicated differences in heterogeneity between tests and so we extended the bivariate model to allow the variances of the random-effects to vary with test type. We assessed the statistical significance of the difference in sensitivity or specificity between tests by using likelihood ratio tests comparing models with and without the covariate terms in the bivariate model. The summary sensitivities and specificities (i.e. average operating points) were plotted on summary ROC plots with corresponding 95% confidence regions. Summary positive and negative likelihood ratios were derived from functions of the bivariate model parameters, with 95% confidence intervals computed using the delta method.
Indirect comparisons of tests are not ideal and are susceptible to bias because other factors, such as participant and study design characteristics, may confound differences between tests. Thus in secondary analyses, we restricted the test comparisons to only studies that evaluated the tests in the same population. Because the studies were few, we were unable to perform meta-analyses but used linked summary ROC plots where estimates for each of the two tests from each study are joined by a line to illustrate the results. Furthermore, for each target condition, we quantified the difference in sensitivities and specificities between pairs of tests by computing differences in these proportions together with the corresponding 95% CI. Thus we visually and numerically demonstrated the change and consistency of the direction of the change in test performance between the tests. We used the xtmelogit command in Stata version 11.2 (StataCorp, College Station, Texas) to fit the bivariate models.
Investigations of heterogeneity
Heterogeneity was investigated in the first instance through visual examination of forest plots and summary ROC plots. The type of tear and type of reference standard reported in each study were presented on forest plots along with the estimates of sensitivity and specificity. In exploratory analyses, we ordered studies on the forest plots by each of the two covariates in turn and also by sensitivity or specificity to examine the pattern of variation between studies. If there were sufficient data we planned to formally investigate heterogeneity by adding covariates to the bivariate model for each potential source of heterogeneity.
If there were sufficient studies, we performed sensitivity analyses by comparing results based on all studies with results of subsets of studies that complied (scored 'Yes') with the following methodological quality items of the QUADAS checklist (Whiting 2003).
- Representative spectrum
- Acceptable reference standard
- Acceptable delay between tests
- Index test results blinded
- Reference standard results blinded
We also investigated the effect of unit of analysis by excluding studies that included both shoulders for any individual.
Results of the search
The search strategy identified 3169 references and the handsearch identified an additional three records (Figure 1). Of these, 2902 were excluded by initial screening of reference titles and abstracts. There were 974 duplicates and 1926 were either not relevant or did not meet the inclusion criteria. We were unable to obtain full text articles for two studies because they were not available from libraries or vendors.
|Figure 1. Study flow diagram|
Of the 270 potentially eligible studies that were remaining and for which full reports were obtained (192 were reported in English and 78 in a non-English language), 20 studies met our inclusion criteria and were included in the review. Three of the included studies had additional published data. Two hundred and eighteen studies did not meet our inclusion criteria and were excluded (see Characteristics of excluded studies) and four reported on the same population or a subset of an already excluded study. At the time of publication, we are still awaiting translation of 25 non-English articles that are potentially relevant based upon their title and abstract; these are listed in Studies awaiting classification. Data from these studies will be added in future updates of this review if the studies are found to be eligible for inclusion.
Among the 20 included studies, six (Iannotti 2005; Kang 2009; Martin-Hervas 2001; Sipola 2010; Swen 1999; Teefey 2004) evaluated the accuracy of two different tests against the reference standard(s). See the Characteristics of included studies for details of the individual studies.
Methodological quality of included studies
The methodological quality of the 20 included studies was judged to be low or unclear for most categories and is summarised in Figure 2. The quality assessment results for the individual studies can be found in Figure 3 and details are given in the Characteristics of included studies.
|Figure 2. Methodological quality graph: review authors' judgements about each methodological quality item presented as percentages across all included studies|
|Figure 3. Methodological quality summary: review authors' judgements about each methodological quality item for each included study|
The spectrum of participants (item 1) was judged to be representative in only 6 (30%) of the 20 studies. To be judged representative, studies had to be prospective with consecutive recruitment. The setting had to be secondary or tertiary care and the patients had to present with shoulder pain caused by a suspected rotator cuff tear for which surgery was being considered for treatment. Only half of the studies included an appropriate reference standard (item 2) and avoided partial verification (item 4). The majority (more than 50%) of studies poorly described the following QUADAS items: time period between reference standard and index test (item 3), differential verification bias (item 5), reference standard results blinded (item 8), relevant clinical information (item 9), and learning curve and training reported for both the index and reference standard readers (items 12 and 13) (see Appendix 2 for further explanation of these items). The remaining QUADAS items were well described in 50% to 75% of the included studies: index test results blinded (item 7), un-interpretable results reported (item 10), withdrawals explained (item 11) and index test criteria for a positive result (item 14). Criteria for test positivity was reported by 15 studies and varied between studies; the criteria are presented in detail in the Characteristics of included studies. As we anticipated in our protocol, the answer for 'incorporation avoided' (item 6) was 'Yes' (no bias) for all included studies.
Our meta-analyses were based on indirect comparisons because meta-analyses of studies that directly compared tests were not possible; there were only six comparative studies. No study directly compared MRA and MRI, or all three modalities in the same patients for any of the target conditions.The summary sensitivities and specificities for the tests are shown in Table 1 for each target condition. For MRA, meta-analysis was performed only for studies evaluating detection of full thickness tears due to the few studies and the degree of heterogeneity observed on summary ROC plots for studies evaluating partial thickness tears and any rotator cuff tears.
Two studies (Iannotti 2005; Milosavljevic 2005) included both shoulders of one and five patients respectively. The remaining studies reported the same number of patients and shoulders, with the exception of Milosavljevic 2005 where this information was missing.
Detection of any rotator cuff tears
Figure 4 shows the forest plots of the sensitivity and specificity estimates for MRI, US and MRA for the 17 studies that assessed any rotator cuff tears.
|Figure 4. Accuracy of MRA, MRI and US for detecting any rotator cuff tears (forest plot)|
Six studies, based on 347 shoulders from 346 patients, assessed the diagnostic accuracy of MRI. The median study size was 55 (range 30 to 99), and the median prevalence of any rotator cuff tear was 73% (range 50% to 96%). The sensitivity of MRI reported in the studies ranged from 91% to 100%, and specificity from 67% to 100%. The summary estimates for the sensitivity and specificity of MRI were 98% (95% CI 92% to 99%) and 79% (95% CI 68% to 87%) respectively. The positive and negative likelihood ratios were 5 (95% CI 2 to 10) and 0.03 (95% CI 0.01 to 0.11) respectively.
Thirteen studies assessed the accuracy of US to detect any rotator cuff tears. The studies included a total of 854 shoulders from 848 patients with a median study size of 50 (range 24 to 190). The prevalence of any rotator cuff tears in the US studies was 80% (range 34% to 92%), and the sensitivities ranged from 33% to 100%, specificities from 45% to 100%. The summary sensitivity and specificity of US were 91% (95% CI 83% to 95%) and 85% (95% CI 74% to 92%) respectively. The positive and negative likelihood ratios were 6 (95% CI 3 to 12) and 0.11 (95% CI 0.05 to 0.22) respectively.
Three studies, based on 183 shoulders from 183 participants, assessed the accuracy of MRA for detection of any rotator cuff tears. The median study size was 58 (range 50 to 75), and the median prevalence was 85% (range 62% to 90%). The sensitivity of MRA ranged from 72% to 100%, and specificity from 5% to 80%. Meta-analysis was not performed but study specific estimates of sensitivity and specificity were plotted in ROC space with 95% CI in Figure 5.
|Figure 5. Study estimates of sensitivity and specificity with 95% confidence intervals plotted in ROC space for MRA for the detection of any rotator cuff tears|
Comparison of MRI and US for detection of any rotator cuff tears
Using the 11 studies that evaluated the accuracy of either MRI or US for detection of any rotator cuff tears, neither test was found to be superior in terms of sensitivity or specificity. Although the sensitivity of MRI was 7% higher than that of US and the specificity of MRI was 6% lower than that of US (Figure 6; Table 1), there was no statistically significant difference between the two tests (P = 0.13). In the analysis restricted to the three studies (231 shoulders from 230 patients) that performed head-to-head comparisons of MRI and US within the same patients ( Table 2, see Appendix 3 for additional figure), two studies reported higher sensitivity and specificity for MRI compared to US while the other study reported higher sensitivity and specificity for US compared to MRI. For head-to-head comparisons of MRA and US, there were only two studies (127 shoulders from 127 patients). Both studies reported higher sensitivity for MRA compared to US but the estimates of specificity were conflicting ( Table 3).
|Figure 6. Study estimates of sensitivity and specificity, and summary points with 95% confidence regions plotted in ROC space for MRI and US for the detection of any rotator cuff tears|
Detection of full thickness rotator cuff tears
The estimates of sensitivity and specificity for the 14 studies that evaluated either MRI, US or MRA for the detection of full thickness rotator cuff tears are shown in Figure 7.
|Figure 7. Accuracy of MRA, MRI and US for detecting full thickness rotator cuff tears (forest plot)|
Seven studies, based on 368 shoulders from 367 patients, assessed the diagnostic accuracy of MRI. The median study size was 48 (range 21 to 99), and the median prevalence of full thickness rotator cuff tear was 62% (range 37% to 69%). The sensitivities ranged from 77% to 100%, and specificities ranged from 68% to 100%. The summary sensitivity and specificity of MRI were 94% (95% CI 85% to 98%) and 93% (95% CI 83% to 97%) respectively. The positive and negative likelihood ratios were 13 (95% CI 6 to 29) and 0.06 (95% CI 0.02 to 0.16) respectively.
Ten studies (729 shoulders from 723 patients) assessed the accuracy of US to detect full thickness tears. The median study size was 66 (range 21 to 190), and the median prevalence was 48% (range 29% to 80%). Sensitivities ranged from 58% to 100%. Specificities ranged from 53% to 100%. The summary sensitivity and specificity of US were 92% (95% CI 82% to 96%) and 93% (95% CI 81% to 97%) respectively. The positive and negative likelihood ratios were 12 (95% CI 5 to 34) and 0.09 (95% CI 0.04 to 0.20) respectively.
Three studies (the same studies that assessed any rotator cuff tears) assessed the accuracy of MRA to detect full thickness tears with sensitivities ranging from 88% to 100% and specificities ranging from 90% to 94%. The median prevalence was 76% (range 17% to 80%). The summary sensitivity and specificity of MRA were 94% (95% CI 80% to 98%) and 92% (95% CI 83% to 97%) respectively. The positive and negative likelihood ratios were 12 (95% CI 5 to 30) and 0.06 (95% CI 0.02 to 0.23) respectively.
Comparison of MRI, MRA and US for detection of full thickness rotator cuff tears
Based on the 14 studies that assessed the accuracy of MRI, MRA or US for detection of full thickness rotator cuff tears, the summary sensitivities and specificities of MRI, MRA and US were found to be very similar (Figure 8; Table 1). There was no statistically significant difference in sensitivity or specificity (P = 0.7). Four studies (252 shoulders from 251 patients) directly compared MRI and US ( Table 2, see Appendix 3 for additional figure) within the same patients, with no consistency among the studies as to which test was superior in terms of either sensitivity or specificity. Two studies (127 shoulders from 127 patients) directly compared MRA and US ( Table 3). Both studies reported higher sensitivity for MRA compared to US. One of the two studies also reported a higher specificity while the other study reported no difference.
|Figure 8. Study estimates of sensitivity and specificity, and summary points with 95% confidence regions plotted in ROC space for MRA, MRI and US for the detection of full thickness rotator cuff tears|
Detection of partial thickness rotator cuff tears
Figure 9 shows the estimates of sensitivity and specificity for the 13 studies that evaluated either MRI, MRA or US for the detection of partial rotator cuff tears.
|Figure 9. Accuracy of MRI, US and MRA for detecting partial thickness rotator cuff tears (forest plot)|
All six studies (347 shoulders from 346 participants) that assessed the accuracy of MRI for the detection of any rotator cuff tears also assessed partial thickness tears. The median prevalence of partial thickness tears was 20% (range 3% to 37%). Sensitivities ranged from 50% to 100% and specificities ranged from 75% to 98%. The summary sensitivity and specificity of MRI were 74% (95% CI 59% to 85%) and 93% (95% CI 84% to 97%) respectively. The positive and negative likelihood ratios were 10 (95% CI 4 to 26) and 0.28 (95% CI 0.17 to 0.48) respectively.
Eight studies (660 shoulders from 654 participants) assessed the accuracy of US to detect partial thickness tears with sensitivities ranging from 13% to 100% and specificities ranging from 68% to 100%. The median prevalence was 14% (range 5% to 37%). The summary sensitivity and specificity of US were 52% (95% CI 33% to 70%) and 93% (95% CI 85% to 97%) respectively. The positive and negative likelihood ratios were 8 (95% CI 3 to 19) and 0.52 (95% CI 0.33 to 0.80) respectively.
Four studies, based on 233 shoulders from 233 participants, assessed the accuracy of MRA to detect partial thickness tears with sensitivities ranging from 62% to 80% and specificities ranging from 47% to 100%. The median prevalence was 27% (range 9% to 54%). Meta-analysis was not performed but study specific estimates of sensitivity and specificity were plotted in ROC space with 95% CI in Figure 10.
|Figure 10. Study estimates of sensitivity and specificity with 95% confidence intervals plotted in ROC space for MRA for the detection of partial thickness rotator cuff tears|
Comparison of MRI and US for detection of partial thickness rotator cuff tears
The diagnostic accuracy of MRI and/or US for detecting partial thickness tears was assessed in 11 studies. There was no statistically significant difference in sensitivity or specificity (P = 1.0) (see Table 1). The individual study estimates of sensitivity and specificity, with summary points and 95% confidence regions, for each test are shown in ROC space (Figure 11). The sensitivities for MRI and US were generally lower for detecting partial thickness tears than for detecting any or full thickness rotator cuff tears. The sensitivity of US for detecting partial thickness tears was only 52% (95% CI 33% to 70%).
|Figure 11. Study estimates of sensitivity and specificity, and summary points with 95% confidence regions plotted in ROC space for MRI and US for the detection of partial thickness rotator cuff tears|
The sensitivities and specificities of the three studies that directly compared MRI and US are shown in a ROC space (see Appendix 3 for figure) and differences between the sensitivities and specificities of the tests are presented for each study in Table 2. Two of the studies reported better sensitivity for MRI than US while all three studies reported better specificity for MRI compared to US. Two studies directly compared MRA and US for detection of any rotator cuff tears. Both studies reported better sensitivity and specificity for MRI compared with those of US ( Table 3). The same studies also assessed partial thickness tears.
Detection of any subscapularis tendon tears
One study, Mohtadi 2004, assessed the accuracy of MRA for detection of any subscapularis tendon tears, and included 58 shoulders from 58 participants. The study had a prevalence of 33% for subscapularis tendon tears. The sensitivity and specificity of MRA were 79% (95% CI 54% to 94%) and 72% (95% CI 55% to 85%) respectively.
Investigation of heterogeneity
The type of tear and the reference standard used in each study are shown by forest plots for each target condition in Figure 4, Figure 7 and Figure 9. The studies on each plot were ordered according to sensitivity and specificity to demonstrate any pattern in the observed estimates of test accuracy. Based on these descriptive analyses and the magnitude of the variances of the random-effects parameters, we observed greater variability in sensitivity and specificity across studies of US than across studies of MRI or MRA. We were unable to formally investigate potential sources of heterogeneity because the number of studies available for each test was either inadequate or the same value of a covariate was reported by most studies.
There were few studies of MRI and MRA, and so we could not perform sensitivity analyses for these tests. We performed sensitivity analyses for US for each of the target conditions. We were only able to investigate the impact of two (acceptable reference standard and index test results blinded) of the five quality items we had specified because few studies scored 'Yes' on the other three items (representative spectrum, acceptable delay between tests, and reference test results blinded). There were small differences in sensitivity and/or specificity (Appendix 4). The largest difference was observed between the summary sensitivity of US for detecting partial thickness tears based of all studies (52%, 95% CI 33% to 70%) and the summary sensitivity (62%, 45% to 77%) based on only studies where the reference standard was acceptable. However, the confidence intervals were comparable and the specificities were similar. The exclusion of studies that did not meet either criteria made no difference to our findings. Two studies included both shoulders for six participants and one study did not report the number of participants so it is unclear whether more than one shoulder was included per participant. We investigated the impact of the unit of analysis on the findings for MRI and US by excluding the three studies, thus assuming the individual as the unit of analysis; the results were found to be consistent with the main analyses based on shoulders.
Summary of findings
Summary of main results
This review summarised the evidence for the diagnostic accuracy of MRI, MRA and US for detecting rotator cuff tears in people with shoulder pain who were suspected of having a rotator cuff tear and for whom surgery was being considered. These imaging tests are usually carried out to determine the characteristics of the rotator cuff tear in order to plan surgery. We included only prospective accuracy studies that evaluated at least one of the tests. We identified 20 studies (1147 shoulders, 1141 participants), of which six evaluated the accuracy of two of the tests within the same participants (paired comparison).
We found no evidence to suggest differences in the sensitivities and specificities of MRI and US for detecting any rotator cuff tears or partial thickness tears. Similarly, we found no evidence to suggest differences in the sensitivities and specificities of MRI, MRA and US for detecting full thickness tears. The estimates were very similar and the tests demonstrated good discriminatory ability for detecting full thickness tears, with sensitivities and specificities of 92% and above. MRI and US had lower sensitivity for partial thickness tears than for any rotator cuff tears or full thickness tears, with US having a sensitivity of only 52% (95% CI 33% to 70%); this indicates that US may be only marginally better than chance in excluding a partial thickness tear. The specificities of the three tests were generally good except for detection of any rotator cuff tears. The estimates of sensitivity and specificity for any rotator cuff tears suggest that in a population of 100 people with shoulder pain suspected of having a rotator cuff tear and for whom surgery is being considered, if the prevalence was 80%, investigation with MRI may miss two cases (2/80, 3%), while investigation with US may miss seven cases (7/80, 9%). Among patients without a rotator cuff tear (20 out of 100), four patients tested using MRI may have a rotator cuff tear wrongly detected (4/20, 20%) and may undergo unnecessary surgery. A similar number (3/20, 15%) may be over-treated if US is used. The summary of all results are provided in Summary of findings.
It is important to emphasise that our review specifically addressed imaging of the rotator cuff by MRI, MRA or US in people with shoulder pain suspected of having a rotator cuff tear and for whom surgery is being considered, and therefore our results are not generalisable to people who present with shoulder pain in primary care where the prevalence of rotator cuff tears may be lower but importantly the prevalence of asymptomatic tears or people with shoulder pain not contemplating surgery could be much higher. Asymptomatic changes in the rotator cuff are common and increase with age and many observed abnormalities might not require specific treatment (Awerbuch 2008). Despite studies continuing to show that primary care practitioners display an over-reliance upon early imaging for shoulder pain (Buchbinder 2013; Johal 2008; Patel 2011), at the present time, guidelines for the management of shoulder pain in primary care do not advocate imaging for shoulder pain unless there is a suggestion of serious pathology (Bussières 2007; Geraets 2009).
The unit of analysis used in evaluating the diagnostic accuracy of a test is likely to have an impact on the estimates of sensitivity and specificity of the test. Our unit of analysis was the shoulder. However, only six out of 1080 participants had both shoulders included in 19 of the 20 included studies; it was unclear in one study (Martin-Hervas 2001) whether the number of shoulders was the same as the number of participants. With the exception of Iannotti 2005 and Milosavljevic 2005, the studies reported the same number of participants and shoulders. Both Iannotti 2005 and Martin-Hervas 2001 compared the accuracy of MRI and US while Milosavljevic 2005 evaluated only US. In sensitivity analyses, we examined the impact of the unit of analysis by excluding the two studies that included both shoulders for any participant and the one study where it was unclear if the number of shoulders was the same as the number of participants. Overall, findings from the sensitivity analyses were consistent with findings from the main analyses.
Strengths and weaknesses of the review
This review was planned and conducted following criteria and methods set out in a published protocol (Lenza 2011). Our results were based on a comprehensive and sensitive literature search that aimed to identify all published studies. We used wide search terms and several electronic databases, not limited by language, and we excluded search filters for diagnostic terms, as they have limited utility (De Vet 2008). Other strengths of this review are our quality assessment of studies and our synthesis of studies with similar methodological features into a meta-analytic summary based on recommended methods. To increase the applicability and reliability of the summary findings, we included only prospective studies that investigated people with shoulder pain due to a suspected rotator cuff tear and for whom surgery was being considered. We excluded retrospective studies because of their potential for high risk of spectrum and verification bias (Bossuyt 2003; Van der Schouw 1995).
Our review has some limitations. Our findings were based on small studies with poor reporting of participant characteristics and study design. Most of the QUADAS items were scored as unclear for many studies. For example, only 25% of the included studies reported the time interval between the index tests and the reference standard. For some analyses, we observed considerable heterogeneity in sensitivity and/or specificity, which may be due to several factors including variation in the criteria for a positive diagnostic test for both the index tests and the reference standard, technical details of the tests, variation in population, and variation in operator or reader experience. The three diagnostic tests are known to be operator and reader dependent which may account for some of the observed variation between studies, especially for studies of US which were found to be very heterogeneous. We could not formally investigate potential sources of heterogeneity due to the number of studies available for each test or because most studies reported the same covariate value. Our comparative meta-analyses were based mainly on non-comparative studies because only a small number of studies made direct comparisons between the tests. Consequently, it is possible that observed differences between tests may be confounded by differences in participant and study design characteristics. It is unclear to what extent these limitations influenced our findings.
An important weakness of this review is that due to resource limitations, 25 potentially eligible studies published in non-English languages are still awaiting translation. Good quality translation will be required to reliably extract data from these papers due to the complexity of diagnostic accuracy studies.The studies contain more than 2900 participants that could potentially provide data for analyses and they will be considered for inclusion in a future update of the review.
Comparison with existing reviews
We identified six previous systematic reviews of imaging tests to detect rotator cuff tears (De Jesus 2009; Dinnes 2003; Kelly 2009; Ottenheijm 2010; Shahabpour 2008; Smith 2012). Our review limited inclusion to prospective studies whereas the other systematic reviews allowed the inclusion of retrospective studies. Our literature search failed to identify a study (Ruiz Santiago 2000) which was included in the review by Smith 2012. However, this study would not have been eligible for inclusion in our review because arthrography or arthrographic computed tomography was also used as an index test.
Previous reviews reported similar results. De Jesus 2009 compared US with MRI for detecting rotator cuff tears using surgery as the reference standard. De Jesus 2009 included 65 studies and concluded that US was as accurate as MRI for diagnosing both full and partial thickness rotator cuff tears. Dinnes 2003 assessed the diagnostic accuracy of clinical testing, US and MRI for detecting rotator cuff tears using surgical and non-surgical tests as the reference standard (results also reported in Kelly 2009). Dinnes 2003 concluded that US and MRI were equivalent for detecting full thickness rotator cuff tears, and that MRI may be better at detecting partial thickness tears than US. Shahabpour 2008 also concluded that US and MRI were equivalent for detecting full thickness rotator cuff tears. However, in contrast Shahabpour 2008 concluded that MRA and US may be more accurate at detecting partial thickness tears than MRI. We did not pool MRA studies for detection of partial thickness tears. While our results suggested that MRI may be more sensitive than US, the difference was not statistically significant.
Ottenheijm 2010 assessed the accuracy of US for detecting subacromial diseases in patients presenting in primary and secondary care settings (search date 2001 to June 2010). This systematic review included 23 studies and reported pooled sensitivity and specificity values that were comparable with our results for detecting full thickness tears. Ottenheijm 2010 reported a sensitivity of 95% for detecting full thickness tears compared to 92% (95% CI 82% to 96%) in our systematic review and a specificity of 93% compared with 93% (95% CI 81% to 97%) in our systematic review. However, for detection of partial thickness tears, Ottenheijm 2010 reported a much higher pooled sensitivity of 72% compared with our finding of 52% (95% CI: 33% to 70%). Smith 2012, which included both retrospective and prospective studies, assessed the diagnostic accuracy of MRI and identified 44 studies published up to May 2011. This systematic review reported pooled sensitivity and specificity values that were similar to our results for detecting full thickness tears and partial thickness tears. Smith 2012 reported a pooled sensitivity of 91% (95% CI 86% to 94%) for detecting full thickness tears which was comparable to our result of 94% (95% CI 85% to 98%). Smith 2012 reported a pooled specificity of 97% (95% CI: 96% to 98%) for detecting full thickness tears which is similar to our specificity of 93% (95% CI 83% to 97%). Smith 2012 reported a pooled sensitivity of 80% (95% CI 79% to 84%) for detecting partial thickness tears which is comparable to our sensitivity of 74% (95% CI 59% to 85%); and a pooled specificity of 95% (95% CI 94% to 97%) which is similar to our specificity of 93% (95% CI 84% to 97%). Overall, the results are generally consistent across the different reviews even though there were differences in inclusion criteria and review methods. Despite our study being the most up-to-date published systematic review, we included a much smaller number of studies (20 studies) than some of the previous reviews because we restricted our analyses to only prospective studies thus reducing the risk of spectrum and verification bias.
Applicability of findings to the review question
The applicability of our findings is limited because only 30% of the included studies reported an adequately representative spectrum of consecutive patients from secondary or tertiary care. Furthermore, partial verification was avoided in only 50% of the studies. MRI, MRA and US may have similar accuracy for detecting full thickness rotator cuff tears. The sensitivity of both MRI and US for partial thickness rotator cuff tears appeared to be much lower than their sensitivity for any rotator cuff tears or for full thickness tears. While the difference in sensitivity between MRI and US for detecting partial thickness tears was not statistically significant, US showed a much lower sensitivity (52%) than MRI (74%). A sensitivity of 52% suggests that US may not be any better than chance for detecting partial thickness rotator cuff tears. The specificities of the three tests were generally high except for the detection of any rotator cuff tears.
In many countries, US is less time consuming and less expensive and more readily available in secondary and tertiary care than MRI or MRA. Despite MRI and MRA being comparable for detection of full thickness rotator cuff tears, the choice of test may depend upon cost and availability. As the scope of this review was to limited to test accuracy, we were not able to determine if applying any imaging test prior to surgery results in different surgical interventions or benefits in terms of pain relief and shoulder function following surgery.
Implications for practice
The diagnostic performance of MRI and US depends on the extent (i.e. partial or full thickness) of rotator cuff tears. Our findings suggest that MRI, US and MRA have good diagnostic accuracy and any of these tests could equally be used for detection of full thickness tears in people with shoulder pain for whom surgery is being considered. MRI and US also have good sensitivity for detecting any rotator cuff tears but poor sensitivity for detection of partial thickness tears. The validity and generalisability of our findings are limited because they were based on small, heterogeneous, non-comparative studies with methodological flaws.
Implications for research
There is a lack of good quality prospective cohort studies that directly compare the accuracy of MRI, MRA and US shoulder imaging tests for people in secondary and tertiary care, with suspected rotator cuff tears, for whom surgery is being considered. Consequently, further studies are needed in order to evaluate the comparative accuracy of these imaging tests in such circumstances. Future studies should use a blinded design and should limit the amount of time between the index and reference tests as much as possible because there is evidence that rotator cuff tears can progress over time. We suggest that arthroscopy be used as the reference standard test because it is accurate for assessing the articular and bursal side of the rotator cuff. The results of the index test(s) and reference standard should be interpreted by experienced operators.
We thank Dr Helen Handoll and Mr Jonathan Rees; and the peer reviewers and contact editors of the Cochrane Diagnostic Test Accuracy Working Group (Dr Rob Scholten, Prof Danielle van der Windt) for helpful feedback at editorial review. We would also like to thank Mrs Lindsey Elstub and Dr Joanne Elliott for their assistance in preparing the review.
We also thank Dr Danielle van der Windt for her useful feedback on the inclusion/exclusion criteria and Robin Christensen for his contribution to the protocol.
- Top of page
- Authors' conclusions
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Differences between protocol and review
- Index terms
Presented below are all the data for all of the tests entered into the review.
Appendix 1. Search strategies
((Ultrasonography [mh] OR ultrasound [tw] OR ultrasonograph* [tw] OR sonograp*[tw] OR us [sh]) OR (Magnetic Resonance Imaging [mh] OR MR imag*[tw] OR magnetic resonance imag* [tw] OR MRI [tw])) AND (Rotator Cuff [mh] OR rotator cuff* [tw] OR musculotendinous cuff* [tw] OR subscapularis [tw] OR supraspinatus [tw] OR infraspinatus OR teres minor [tw]) AND (Rupture [mh:noexp] OR tear* [tw] OR torn [tw] OR thickness [tw] OR lesion* [tw] OR ruptur* [tw] OR injur* [tw])
Total references = 1551
1 'echography'/de AND [embase]/lim (124208)
2 ultrasound:ab,ti OR ultrasonograph*:ab,ti OR sonograp*:ab,ti AND [embase]/lim (192495)
3 #1 OR #2 (242499)
4 'nuclear magnetic resonance imaging'/de AND [embase]/lim (277184)
5 (('magnetic resonance' OR mr) NEAR/3 imag*):ab,ti AND [embase]/lim (130882)
6 mri:ab,ti AND [embase]/lim (108797)
7 #4 OR #5 OR #6 (311974)
8 'rotator cuff injury'/de OR 'rotator cuff rupture'/de AND [embase]/lim (3561)
9 'rotator cuff'/de AND [embase]/lim (1850)
10 'rotator cuff':ab,ti OR 'musculotendinous cuff':ab,ti OR subscapularis:ab,ti OR supraspinatus:ab,ti OR infraspinatus:ab,ti OR 'teres minor':ab,ti AND [embase]/lim (5679)
11 #9 OR #10 (6120)
12 'rupture'/de AND [embase]/lim (3798)
13 tear*:ab,ti OR torn:ab,ti OR thickness:ab,ti OR lesion*:ab,ti OR ruptur*:ab,ti OR injur*:ab,ti AND [embase]/lim (1001852)
14 #12 OR #13 (1002130)
15 #11 AND #14 (3615)
16 #8 OR #15 (4908)
17 #3 OR #7 (526691)
18 #16 AND #17 (1572)
(Mh Ultrasonography OR Tw ultrasound OR Tw ultrasonograph$ OR Tw Sonograp$) OR (Mh Magnetic Resonance Imaging OR (Tw magnetic AND Tw resonance AND Tw imag$) OR Tw MRI) [Words] and Mh Rotator Cuff OR (Tw rotator AND Tw cuff) OR (Tw musculotendinous AND Tw cuff) OR Tw subscapularis OR Tw supraspinatus OR Tw infraspinatus OR (Tw teres AND Tw minor) [Words] and Mh Rupture OR Tw tear$ OR Tw torn OR Tw thickness OR Tw lesion$ OR Tw rupture$ OR Tw injur$ [Words]
Total references = 30
Appendix 2. Assessment of methodological quality: QUADAS and additional items
Appendix 3. Additional figures
Summary ROC plot of within study comparisons of MRI and US for detection of any rotator cuff tears (Figure 12)
|Figure 12. Paired comparison of MRI and US for detection of any rotator cuff tears. Connectling lines link study estimates of sensitivity and specificity for both tests in each study|
Summary ROC plot of within study comparisons of MRI and US for detection of full thickness rotator cuff tears (Figure 13)
|Figure 13. Paired comparison of MRI and US for detection of full thickness rotator cuff tears. Connectling lines link study estimates of sensitivity and specificity for both tests in each study|
Summary ROC plot of within study comparisons of MRI and US for detection of partial thickness rotator cuff tears (Figure 14)
|Figure 14. Paired comparison of MRI and US for detection of partial thickness rotator cuff tears. Connectling lines link study estimates of sensitivity and specificity for both tests in each study|
Appendix 4. Sensitivity analyses for US studies for detection of rotator cuff tears (any, partial or full thickness)
Sensitivity analyses performed by excluding studies that scored 'Unclear' or 'No' for each of the two QUADAS criteria listed in the table.
Last assessed as up-to-date: 22 August 2011.
Contributions of authors
All authors contributed to the development of the review and commented on and approved the final version. The guarantor of this review is Mario Lenza.
Declarations of interest
Sources of support
- Universidade Federal de São Paulo, Brazil.
- The Parker Institute, Denmark.The Parker Institute: Musculoskeletal Statistics Unit is supported by grants from The Oak Foundation
- Teesside University, UK.
- Department of Epidemiology and Preventive Medicine, School of Public Health and Preventive Medicine, Monash University, Australia.In kind support
- Cabrini Institute, Cabrini Hospital, Malvern, Victoria, Australia.In kind support
- No sources of support supplied
Differences between protocol and review
- We only included studies of participants suspected of having rotator cuff tears. Studies that reported results of people who had been previously diagnosed with, or suspected of having, other specific shoulder diagnoses were excluded. If it was unclear whether or not all participants were suspected of having rotator cuff tears, we also excluded these studies
- Inasmuch as there is no set time point beyond which it is known that rotator cuff tears progress, we accepted studies in which the time between the index test and the reference standard test was up to a year (rather than six months as specified in the protocol).
- We included the MEDION database in our search strategy.
- We restricted our analyses to prospective studies and excluded retrospective studies because of the high risk of spectrum and verification biases in these studies.
- We made an amendment in the assessment of methodological quality – item seven (index test results blinded). We removed “if the study was retrospective” as a reason to say No because we included only prospective studies.
- We made an amendment in the assessment of methodological quality - item eight (reference standard results blinded). We excluded “if the study was retrospective” as a reason to say No because it was covered by the first part of the sentence.
- We included in the assessment of methodological quality table an additional generic quality item assessing whether or not the criteria for a positive index test result was reported.
- We used the bivariate model for meta-analysis instead of the hierarchical summary ROC (HSROC) model. Given the available information, we assumed a common threshold was applicable but with heterogeneity around this common threshold due to variation in interpretation in practice. Therefore we consider the bivariate model and the estimation of summary points (with 95% confidence regions) appropriate for summarising the results of the review.
- We conducted sensitivity analyses to examine the effect of unit of analysis.
Medical Subject Headings (MeSH)
MeSH check words
* Indicates the major publication for the study