The interobserver reliability of the Neer and AO/ASIF classification systems were assessed on the basis of plain radiographs. At least three of four radiograph projections were made; these include an anteroposterior projection, an axillary projection, a scapula lateral projection and a bicepital groove projection of the shoulder. Using the Neer classification system, each fracture was classified into 16types. Using the AO/ASIF classification system, each fracture was classified into 3 types,9 groups and 27 subgroups. Forty-four fractures of the proximal end of the humerus were classified by three shoulder specialists and one orthopaedic resident. Then, kappa coefficients between every possible pair were calculated. The mean kappa coefficient for interobserver reliability of the Neer and AO/ASIF classification systems were 0.547 and 0.378, respectively. Simplification of the Neer and AO/ASIF classification systems, from 16 categories to 6 categories and from 27 categories to 9 categories, respectively, did not significantly improve the interobserver reliability. The mean kappa coefficient of the Neer system was higher than the AO/ASIF system. But both classification systems were difficult to apply in a reliable manner. The differences in classification between each observer may reach a different method of treatment and provide a surgeon with a different estimation of the outcome of that treatment.
