Interobserver agreement in the classification of rotator cuff tears using magnetic resonance imaging

Background: Although magnetic resonance imaging (MRI) is a standard method of assessing the extent and features of rotator cuff disease, the authors are not aware of any studies that have assessed the interobserver agreement among orthopaedic surgeons reviewing MRI scans for rotator cuff disease. Hypothesis: Fellowship-trained orthopaedic shoulder surgeons will have good interobserver agreement in predicting the more salient features of rotator cuff disease such as tear type (full thickness versus partial thickness), tear size, and number of tendons involved but only fair agreement with more complex features such as muscle volume, fat content, and the grade of partial-thickness cuff tears. Study Design: Cohort study (diagnosis); Level of evidence, 3. Methods: Ten fellowship-trained orthopaedic surgery shoulder specialists reviewed 27 MRI scans of 27 shoulders from patients with surgically confirmed rotator cuff disease. The ability to interpret full-thickness versus partial-thickness tears, acromion type, acromioclavicular joint spurs or signal changes, biceps lesions, size and grade of partial-thickness tears, acromiohumeral distance, number of tendons involved and amount of retraction for full-thickness tears, size of full-thickness tears, and individual muscle fatty infiltration and atrophy were assessed. Surgeons completed a standard evaluation form for each MRI scan. Interobserver agreement was determined and a kappa level was derived. Results: Interobserver agreement was highest (>80%) for predicting full- versus partial-thickness tears of the rotator cuff, and for quantity of the teres minor tendon. Agreement was slightly less (>70%) for detecting signal in the acromioclavicular joint, the side of the partial-thickness tear, the number of tendons involved in a full-thickness tear, and the quantity of the subscapularis and infraspinatus muscle bellies. Agreement was less yet (60%) for detecting the presence of spurs at the acromioclavicular joint, a tear of the long head of the biceps tendon, amount of retraction of a full-thickness tear, and the quantity of the supraspinatus. The best kappa statistics were found for detecting the difference between a full- and partial-thickness rotator cuff tear (0.77), and for the number of tendons involved for full-thickness tears (0.55). Kappa for predicting the involved side of a partial-thickness tear was 0.44; for predicting the grade of a partial-thickness tear, it was -0.11. Conclusions: Fellowship-trained, experienced orthopaedic surgeons had good agreement for predicting full-thickness rotator cuff tears and the number of tendons involved and moderate agreement in predicting the involved side of a partial-thickness rotator cuff tear, but poor agreement in predicting the grade of a partial-thickness tear.