RESUMO
Using the American Society for Bone and Mineral Research Task Force case definition for atypical femoral fractures, sensitivity and specificity of radiographic fracture characteristics were calculated. Fracture pattern was the most sensitive and specific characteristic. This suggests that some characteristics should be weighted more heavily when identifying these fractures. INTRODUCTION: To estimate the sensitivity and specificity of each radiographic criterion in the 2013 ASBMR atypical femoral fracture (AFF) case definition for distinguishing AFF from other subtrochanteric/diaphyseal fractures (non-AFF) among women enrolled in a large integrated health care organization. METHODS: Radiographs from 55 physician-confirmed AFFs and a sample of 39 non-AFFs were reviewed by four independent expert reviewers representing four medical specialties. One image per fracture was selected for review. Using a standardized data collection tool, based on the 2013 AFF case definition, reviewers indicated the presence or absence of the following characteristics viewable on radiograph: fracture pattern, comminution, periosteal and/or endosteal thickening, and cortical thickening. Sensitivity and specificity for each characteristic was calculated for each reviewer and summarized across reviewers with the mean and range. Agreement across reviewers was quantified using Fleiss's kappa (FK) statistic. RESULTS: The most sensitive factors distinguishing AFF from non-AFF were lateral cortex transverse fracture pattern (mean 93.6 %, range 85.5-98.2 %), medial cortex transverse or oblique fracture pattern (mean 84.1 %, range 72.7-98.2 %), and minimal/non-comminution (mean 93.2 %, range 89.1-98.2 %). Specificity was the greatest for lateral cortex transverse fracture pattern (mean 95.5 %, range 92.3-97.4 %). Agreement across reviewers was the highest for lateral cortex transverse fracture pattern (FK 0.83) and incomplete fracture through the lateral cortex only (FK 0.80). CONCLUSION: Lateral cortex transverse fracture pattern was the most sensitive and specific characteristic and the most highly agreed upon across reviewers. Other characteristics were less readily agreed upon across reviewers. Measurement of discrete combinations of individual characteristics may enhance sensitivity and/or specificity.