The Effect of Two Factors on Interobserver Reliability for Proximal Humeral Fractures

The purpose of this study was to assess whether training observers and simplifying proximal humeral fracture classifications improve interobserver reliability among a large number of orthopaedic surgeons.

One hundred eighty-five observers were randomized to receive training or no training in a simple classification for proximal humeral fractures before evaluating preoperative radiographs of a consecutive series of 30 patients who were treated with open reduction and internal fixation.

The overall interobserver reliability of the simple proximal humeral fracture classification system was low and not significantly different between the training and the no training group (κ = 0.20 and κ = 0.18, respectively; P = 0.10). Subgroup analyses showed that training improved the agreement among surgeons who have been in independent practice ≤5 years (κ = 0.23 versus κ = 0.14; P < 0.001), surgeons from the United States (κ = 0.23 versus κ = 0.16; P = 0.002), and general orthopaedic surgeons (κ = 0.42 versus κ = 0.15; P = 0.021).

Simplifying classifications and training observers did not improve the interobserver reliability for the diagnosis of proximal humeral fractures. However, training observers improved interobserver reliability of a simple proximal humeral fracture classification system among surgeons from the United States and, in particular, younger and less specialized surgeons. This finding may suggest that our interpretations of radiographic information might become more fixed and immutable with experience.