Diagnostic Value of MRA and MRI for The Articular-Sided Partial-Thickness Rotator Cuff Tears: A Meta-Analysis

Background: This study aims to compare the diagnostic accuracy of magnetic resonance imaging (MRI) and MR arthrography (MRA) for the articular-sided partial-thickness rotator cuff tear (PTRCT). Methods: Three electronic databases, PubMed/Medline, Embase and Cochrane Library, were utilized to retrieve articles comparing the diagnostic value of MRA and MRI for detecting articular-sided PTRCTs. The pooled statistical indexes included sensitivity, specicity, positive/negative predictive value, diagnostic odds ratio (DOR) and the area under receiver operating characteristic curve (AUC). Results: Eleven studies involving 1703 patients and 1704 shoulders were included. The pooled sensitivity, specicity, DOR and AUC and their 95% CIs of MRA to diagnose articular-sided PTRCTs were 0.81 (95% CI, 0.65-0.90), 0.96 (95% CI, 0.91-0.98), 68.14 (95% CI, 33.20-139.84) and 0.96 (95% CI, 0.94-0.97), respectively. The pooled sensitivity, specicity, DOR and AUC and their 95% CIs of MRI were 0.78 (95% CI, 0.65-0.87) and 0.97 (95% CI, 0.84-0.99), 47.82 (95% CI, 8.29-275.89) and 0.89 (95% CI, 0.86-0.92), respectively. Conclusions: This meta-analysis reveals that MRA has a better diagnostic value than that of MRI for the diagnosis of articular-sided partial-thickness rotator cuff tears, but only small improvement of sensitivity. Considering the price and invasion of MRA, MRI is recommended as an initial examination to detect patients suspected with articular-side partial-thickness rotator cuff tears. priori as follows: 0.20 or less; poor agreement, 0.21 to 0.40; fair agreement, 0.41 to 0.60; moderate agreement, 0.61 to 0.80; substantial agreement, and 0.81 to 0.99; almost perfect agreement. Quantitative analyses in this study were conducted with the forest plots using STATA 12.0 Version (V. 12.0, StataCorp, College Station, TX). The whole process of searching, ltering, data extraction and quality assessment was implemented by two researchers (LFX & DJL) independently and repeatedly. For any discrepancy, a consensus was reached by discussion with an arbitrator (LLX). The pooled estimates indexes to assess the accuracy of MRA/MRI included sensitivity, specicity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR) and area under receiver operating characteristic curve (AUC). The threshold effect was tested by calculating the logarithm of sensitivity and logarithm of 1-specicity. Heterogeneity among the included studies was assessed using the I 2 statistic. The meta-regression and subgroup analyses were conducted to explore the available source of heterogeneity. Publication bias was performed using Deeks' Funnel Plot Asymmetry Test. Additionally, for the non-threshold effect, we performed meta-regression analysis and the patient sample size ( ≥ 100 or < 100), publication year (before or after 2014), magnetic eld strength (3.0-T or not), number of readers ( ≥ 2 or 1), Blinding (Yes or No) and QUADAS-2 score ( ≥ 10 or < 10), design of study (prospective or retrospective) as well as muscle tendon were used as covariates. but only small improvement of sensitivity. Considering the price and invasion of MRA, MRI is recommended as an initial examination to detect patients suspected with articular-side partial-thickness rotator cuff tears.


Introduction
Partial-thickness rotator cuff tear (PTRCT) was rst described by Codman [1] and later divided into bursal-sided, articular-sided, and intratendinous tears according to the depth and location of the tear [2]. Partial tears are shown to have a variable rate of progression with 28-40% eventually becoming fullthickness tears [3][4][5]. It should be noted that articular-sided PTRCT was 2-3 times more common than bursal-sided tears [6,7].
Recent literature has reported inconsistencies in the correlation between rotator cuff (RC) lesions viewed with magnetic resonance imaging (MRI) and those found arthroscopically, especially regarding partial-thickness tears [8,9]. The improvements of arthroscopic diagnosis and imaging modalities and the cognition status of partial-thickness RCTs created a dramatic increase in the number of patients diagnosed with partial-thickness RCTs. Technological advances in arthroscopic shoulder surgery have made surgical management of rotator cuff tears much less invasive and thereby more cost effective [10,11]. Good outcomes can be achieved with arthroscopic debridement and selective acromioplasty among patients with articular-or bursal-sided PT tear of < 50% tendon thickness. Unfortunately, fewer studies have focused on articular-sided tears partial-thickness RCTs. Therefore, it is very important to distinguish articular-sided PTRCTs from all kinds of tears, especially from PTRCTs.
Shoulder plain lm and various physical examination tests have been shown to be insu cient at effectively diagnosing rotator cuff tears [12,13]. A retrospective study in a large homogeneous study group demonstrated that the arthroscopic transtendon repair of articular-sided partial-thickness RCTs is an effective and safe procedure that leads to signi cant improvement in pain and shoulder function, with high patients' satisfaction rate, while the complication rate is low [14].
Numerous diagnostic studies focused on the evaluation of performance of MRA, MRI or ultrasound (US) imaging for the diagnosis of full-or partial-thickness tears. A previous meta-analysis con rmed that 3-dimensional US is very effective and highly accurate to detect full-thickness RCTs, but may lack accuracy in the diagnosis of partial tears [15]. A recent network meta-analysis of diagnostic tests performed by our team reported that high-eld MRA had the highest diagnostic value, then low-eld MRA, followed by high-eld MRI, high-frequency ultrasound, low-eld MRI, and low-frequency ultrasound [16]. However, both of two meta-analyses were performed based on the full-, partial-or any tear. It is important to identify the subtype of partial-thickness tears [17,18]. Furthermore, another meta-analysis also conducted by our team indicated that MRA and MRI have similar diagnostic value for the diagnosis of bursal-sided partialthickness rotator cuff tears [19]. However, there is no study focused on the articular-sided PTRCTs.
To our knowledge, several studies have been performed to rigorously assess the diagnostic performance of MRA and MRI for the articular-sided PT-RCTs, but the results remains controversial. Therefore, the purpose of this meta-analysis was to assess the diagnostic accuracy of MRA and MRI for detecting articularsided PTRCTs based on all available scienti c published data.

Materials And Methods
This study was performed in accordance with the Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies (PRISMA-DTA) statement [20].

Data Sources and Search Strategy
Two independent reviewers searched PubMed, Embase and the Cochrance Library (including Epub Ahead of Print) for titles from data inception to January 1, 2019. The vocabulary and syntax were speci cally adapted according to the database. The combination of keywords and mesh terms were "diagnostic", "diagnosis", "rotator cuff", "supraspinatus", "infraspinatus", or "subscapularis" and "MRI", "magnetic resonance imaging", "MRA" or "magnetic resonance arthrography". The purpose, research question, and eligibility criteria for the search were determined a priori. Additionally, the reference lists of relevant articles and included studies were searched by hand for supplementary eligible records. Any disagreement was settled through the discussion of researchers until a consensus was reached.

Study Screening and Selection
The included studies should con rm all criteria listed as follows: 1) study design, diagnostic accuracy study; 2) population, patients with suspected rotator cuff tear; 3) MRA/MRI test was performed; 4) the nal diagnosis of articular-sided RCT was con rmed by predesigned gold standards; and 5) diagnostic data (number of false/true-positive [FP/TP] and false/true-negative [FN/TN] cases) could be extracted or calculated to construct a two-by-two contingency table.
Exclusion criteria were 1) animal studies or cadaver experiments; 2) studies in which articular-sided RCT could not be differentiated; or 3) commentary, letter, case-report, reviews or conference poster.
Two investigators performed blindly a systematic screening in duplicate. We rst removed redundant and unrelated records by screening titles and abstracts.
Then the full texts of remainders were downloaded to con rm their eligibility based on above criteria. The nal decision regarding inclusion was according to the full article using inclusion and exclusion criteria.

Data Extraction
Following information were collected from all included articles into a pre-designed Microsoft Excel spreadsheet (Version 2010, Microsoft, Redmond, WA, USA) by two investigators blindly and repeatedly: the rst author's surname, publication time of original study, sources of origin, participant characteristics (number, age and gender), design of study, gold standard, time from diagnostic test to gold standard, blinding, no. of readers, the experience of readers, clinical ndings of patients, technical parameters of MRA/MRI including the administration of contrast agent (indirect or direct), vendor, model, magnetic eld strength, various sequences, slice thickness, analyzed image plane. Data extraction from all included studies was completed in tandem by two independent reviewers. The spreadsheets were combined and each reviewer checked a random selection of the other's entries for quality control. Any discrepancies were resolved by consensus decision.

Quality Assessment
Two investigators performed blindly the methodological quality of the included studies in duplicate using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool [21].

Statistical Analysis
Inter-reviewer agreement was calculated at each stage of searching, screening, and the quality assessment of the included studies with a Kappa (κ) statistic.
Agreement was categorized a priori as follows: 0.20 or less; poor agreement, 0.21 to 0.40; fair agreement, 0.41 to 0.60; moderate agreement, 0.61 to 0.80; substantial agreement, and 0.81 to 0.99; almost perfect agreement. Quantitative analyses in this study were conducted with the forest plots using STATA 12.0 Version (V. 12.0, StataCorp, College Station, TX). The whole process of searching, ltering, data extraction and quality assessment was implemented by two researchers (LFX & DJL) independently and repeatedly. For any discrepancy, a consensus was reached by discussion with an arbitrator (LLX). The pooled estimates indexes to assess the accuracy of MRA/MRI included sensitivity, speci city, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR) and area under receiver operating characteristic curve (AUC). The threshold effect was tested by calculating the logarithm of sensitivity and logarithm of 1-speci city. Heterogeneity among the included studies was assessed using the I 2 statistic. The meta-regression and subgroup analyses were conducted to explore the available source of heterogeneity. Publication bias was performed using Deeks' Funnel Plot Asymmetry Test.
Additionally, for the non-threshold effect, we performed meta-regression analysis and the patient sample size (≥ 100 or < 100), publication year (before or after 2014), magnetic eld strength (3.0-T or not), number of readers (≥ 2 or 1), Blinding (Yes or No) and QUADAS-2 score (≥ 10 or < 10), design of study (prospective or retrospective) as well as muscle tendon were used as covariates.

Studies retrieved and Characteristics
In total, 2192 articles were identi ed by searching three databases and removing duplicates. The screening of the reference lists of these and other relevant articles yielded 4 additional studies. After screening remaining titles and abstracts, and identifying related full-text, nally, 11 studies [18,[22][23][24][25][26][27][28][29][30][31] published during the period from 2007 to 2018 remained for quantitative analysis. The selection processes for the eligible studies are summarized in Figure 1.
It has become a popular topic whether contrast agent is used in the application of MRI to diagnosis RCTs. A network meta-analysis of 144 diagnostic studies involving 14059 patients (14212 shoulders) demonstrated that for the detection of full-thickness tears, partial-thickness tears, or any tear, MRA had the highest sensitivity, speci city, and superiority index in three common-used imaging modalities (MRA, MRI and US), which revealed that high-eld MRA had the highest diagnostic value for detecting any tear, followed by low-eld MRA, high-eld MRI, high-frequency US, low-eld MRI, and low-frequency US [16]. In addition, another meta-analysis included 14 studies involving 1216 patients with labral lesions, revealed that MRA had the highest sensitivity and speci city compared with those of MRI and CTA, which indicated that MRA was suggested for use in patients with chronic shoulder symptoms or a pathologic abnormality [32]. MRA has a good diagnostic performance mainly depending on the objective evidence of the leakage of contrast agent, accompanied by a good anatomic resolution and subtle defects depicted by contrast agents [33,34]. Nonetheless, MRA, as an invasive diagnosis modality, has many disadvantages and limitations, including a longer examination time, increased risk of infection and adverse complication [35,36].
MRI, as a noninvasive and reproducible diagnosis image, was recommended for the diagnosis of patient with suspected rotator cuff injuries [37]. Our lab's own research [16] con rmed that the diagnostic value rank (from high to low) of commonly-used modalities was 3.0-T MRA, 1.5-T MRA, 3.0-T MRI, 7.5-MHz US, 1.5-T MRI, and < 7.5-MHz US in the diagnosis of full-thickness tears; however, in the diagnosis of partial-thickness tears, the diagnostic value rank (from high to low) was 3.0-T MRA, 3.0-T MRI, 1.5-T MRA, 7.5-MHz US, 1.5-T MRI, and < 7.5-MHz US. Moreover, other research performed in our lab's con rmed that MRI had a similar sensitivity and speci city to MRA in the detection of bursal-sided PTRCTs [19]. In addition, another meta-analysis revealed that MRI is by far the rst choice recommendation for imaging modality for the detection of acute labral lesions [32]. Recently, one study in 2020 using hierarchical summary receiver operating characteristic curves (HSROC) demonstrated that MRI was recommended to be a rst-choice imaging modality for the detection of rotator cuff tears [38]. Although MRA have a higher sensitivity and speci city, it cannot replace MRI after the comprehensive consideration of accuracy and practicality.
In this study, MRA has a sensitivity of 0.81 and speci city of 0.96 for detecting articular-sided PTRCTs; however, MRI shows a sensitivity of 0.78 and speci city of 0.97 in detection of the articular-sided PTRCTs. The differences in speci city and sensitivity between MRA and MRI are quite small, and that may be a good reason to avoid the potential risk/cost of MRA, this result is in accordance with a prospective study that the sensitivity and speci city of MRA were improved only by 3-4% when compared with MRI [36].
Several limitations in this research should merit consideration. At rst, the number of patients in the included studies is very small and most of the included studies are retrospective type. Evidence of heterogeneity in the pooled data existed cross the included studies. We used the meta-regression analysis to explore many factors that may bring the heterogeneity, including the patient sample size, publication year, magnetic eld strength, number of readers, blinding and QUADAS-2 score, design of study (prospective or retrospective) as well as muscle tendon were used as covariates. Even though we found the source of heterogeneity, the insu cient data cannot support us to conduct relevant subgroup analyses, which will reduce statistical e cacy power. Moreover, this metaanalysis included 11 studies, mainly involved in patients suspected of full-and partial-thickness RCTs and not speci cally articular-sided PTRCTs.
Furthermore, the eld density is very important for the diagnostic value of MRI or MRA in the detection of articular-sided PTRCTs, however, the insu cient data about the eld density of MRA/MRI cannot support us to analysis this important in uence. We assessed only the diagnostic value of the imaging modalities alone (MRA or MRI) without physical tests. Finally, the safety and cost-effectiveness of MRA/MRI in clinical practice should be assessed systematically.

Conclusion
This meta-analysis reveals that MRA has a better diagnostic value than that of MRI for the diagnosis of articular-sided partial-thickness rotator cuff tears, but only small improvement of sensitivity. Considering the price and invasion of MRA, MRI is recommended as an initial examination to detect patients suspected with articular-side partial-thickness rotator cuff tears.  MRA magnetic resonance angiography, MRI magnetic resonance imaging, T tesla, PLR positive likelihood ratio, NLR negative likelihood ratio, DOR diagnostic odd ratio, AUC area under curve, CI con dence interval. Figure 1 Selection process of the included studies.

Figure 2
Forest plots of the pooled sensitivity and speci city of MRA to diagnose articular-sided partial-thickness rotator cuff tears with the corresponding 95% con dence region.

Figure 3
Forest plots of the pooled diagnostic odds ratio (DOR) of MRA to diagnose articular-sided partial-thickness rotator cuff tears with the corresponding 95% con dence region. Summarized receiver operating characteristic curve (SROC) of MRA (A) and MRI (B) to diagnose articular-sided partial-thickness rotator cuff tears with the corresponding 95% con dence region Forest plots of the pooled sensitivity and speci city of MRI to diagnose articular-sided partial-thickness rotator cuff tears with the corresponding 95% con dence region.