Computed tomography versus plain radiography in the Mitsuzawa classification for severe proximal humeral fractures:...

In a prior investigation, we demonstrated that the Mitsuzawa classification for dislocated and severely displaced proximal humeral fractures (PHFs) provided substantially higher reliability than the Neer or AO/OTA systems when applied to plain radiographs. The present study builds on that work by determining whether computed tomography (CT) alters classification outcomes or improves observer agreement.

Four independent reviewers-two shoulder specialists and two orthopedic residents-classified 70 PHFs according to the Neer, AO/OTA, and Mitsuzawa systems. Each fracture fulfilled criteria for dislocation or severe displacement requiring arthroplasty. Evaluations were performed at two separate timepoints: first with radiographs and then with CT scans, including three-dimensional reconstructions. Changes between modalities were recorded. Intraobserver consistency was assessed with the Cohen κ statistic, while interobserver reliability was evaluated with the Fleiss κ statistic.

The incidence of axillary artery injury was 7.1 %. After CT review, classification changes occurred in 31 % of Neer, 29 % of AO/OTA, and 11 % of Mitsuzawa assessments. With radiographs alone, intraobserver reliability was moderate for Neer (κ = 0.52), substantial for AO/OTA (κ = 0.63), and substantial for Mitsuzawa (κ = 0.73). With CT imaging, the corresponding values were 0.55, 0.59, and 0.73. Interobserver agreement ranged from 0.43 to 0.45 (Neer), 0.49-0.52 (AO/OTA), and 0.68-0.68 (Mitsuzawa) with radiographs, and from 0.47 to 0.50, 0.57-0.57, and 0.67-0.69, respectively, after CT. Significant pairwise differences (p < 0.05) in interobserver agreement were observed among all combinations of the three systems for both modalities.

The Mitsuzawa classification was minimally influenced by imaging modality and consistently provided substantial agreement. For dislocated and displaced PHFs, CT appears to have only a minor influence on Mitsuzawa-based categorization, indicating that well-performed radiographs remain sufficient for reproducible classification. These findings have direct implications for clinical decision-making, imaging selection, and anticipating risks such as neurovascular compromise.