Validation of a novel 3-dimensional classification for degenerative arthritis of the shoulder

Abstract

Introduction A novel three-dimensional classification to comprehensively describe degenerative arthritis of the shoulder (DAS) was recently published by our group. The purpose of the present work was to investigate intra- and interobserver agreement as well as validity for the three-dimensional classification. Materials and methods Preoperative computed tomography (CT) scans of 100 patients who had undergone shoulder arthroplasty for DAS were randomly selected. Four observers independently classified the CT scans twice, with an interval of 4 weeks, after prior three-dimensional reconstruction of the scapula plane using a clinical image viewing software. Shoulders were classified according to biplanar humeroscapular alignment as posterior, centered or anterior (> 20% posterior, centered, > 5% anterior subluxation of humeral head radius) and superior, centered or inferior (> 5% inferior, centered, > 20% superior subluxation of humeral head radius). Glenoid erosion was graded 1–3. Gold-standard values based on precise measurements from the primary study were used for validity calculations. Observers timed themselves during classification. Cohen’s weighted κ was employed for agreement analysis. Results Intraobserver agreement was substantial (κ = 0.71). Interobserver agreement was moderate with a mean κ of 0.46. When the additional descriptors extra-posterior and extra-superior were included, agreement did not change substantially (κ = 0.44). When agreement for biplanar alignment alone was analyzed, κ was 0.55. The validity analysis reached moderate agreement (κ = 0.48). Observers took on average 2 min and 47 s (range 45 s to 4 min and 1 s) per CT for classification. Conclusions The three-dimensional classification for DAS is valid. Despite being more comprehensive, the classification shows intra- and interobserver agreement comparable to previously established classifications for DAS. Being quantifiable, this has potential for improvement with automated algorithm-based software analysis in the future. The classification can be applied in under 5 min and thus can be used in clinical practice.