THE USE OF DIGITISED RADIOGRAPHS IN DETERMINING THE CONSISTENCY OF THE AO AND NEER CLASSIFICATIONS OF FRACTURES OF THE PROXIMAL HUMERUS

Abstract

For any fracture classification, a high level of intraobserver reproducibility and interobserver reliability is desirable. We compare the consistency of the AO and Neer classifications for proximal humerus fractures with an assessment of the digitised radiographs of 100 fractures by 10 orthopaedic surgeons and 5 radiologists using the General Electric Picture Archiving and Communications System (PACS), allowing manipulation of the image. This process repeated 1 month later.

Reproducibility and reliability moderate for both the AO and Neer systems. Reproducibility using the AO/ ASIF system was slightly greater. The assessor’s level of experience and specialty did affect accuracy. The ability to electronically manipulate images does not improve reliability and their sole use in describing these injuries and comparing similarly classified fractures from different centres is not recommended.

Fractures of the proximal humerus are common. Most undisplaced or minimally displaced, and treated conservatively. Up to one fifth may benefit from surgery. As decisions regarding treatment are based on the fracture type, a radiological classification should be easy to use and have a high degree of reliability and reproducibility to serve as a useful discriminator, creating standards by which treatment can be recommended and outcomes compared.

Radiographs of 100 fractures of the proximal humerus selected. A true anteroposterior, scapular lateral, and axillary radiograph taken for each fracture. 10 orthopaedic surgeons and 5 radiologists recruited as assessors, including 5 specialist registrars. Each given a printed description of both Neer and AO classifications, a goniometer and ruler. The assessment preceeded by short lecture. Radiographs could be manipulated digitally for size, contrast, brightness, orientation and the negative image displayed. We did not require assessors to determine subgroups for reasons of simplicity. Reproducibility and reliability analysed using Kappa statistical methods. Coefficients for agreement compared using the Student t test incorporating the standard errors of kappa for these groups. A comparison made between radiologists and surgeons, and then consultant orthopaedic surgeons and trainees.

In each case the AO/ASIF system was statistically (p< 0.01) more accurate.

Agreement was greater for less complex (one and two part, and type A) fractures.

Level of experience produced a statistically (p< 0.01) significant difference in accuracy. Specialty did not.

Our analysis comparing the Neer and AO systems uses the largest group of assessors reviewing the largest number of radiographs reported in the literature.

We concur with others in concluding that using these systems in isolation in determining treatment and comparing results following treatment cannot be recommended