Predicting Radiology Resident Errors in Diagnosis of Cervical Spine Fractures 1

Abstract

Our objective was to identify factors associated with resident errors of cervical spine fractures to enable targeted education. We performed a retrospective cohort study of consecutive cases of after-hours resident interpreted cervical spine fractures over 27 months at a single level 1 academic trauma center. The outcome measure was appropriate identification of all fractures by the resident. Potential predictors of resident error or discrepancy were identified from chart review and included: age, gender; fracture location/pattern (upper/lower cervical spine, occipital condyle, C1 ring, dens, C2 pars, vertebral body, posterior column, lateral mass, transverse process); consecutive and nonconsecutive additional fractures; radiologist distracting factors (number of noncervical spine injuries); number of noncervical spine studies performed. Risk ratios with confidence intervals were calculated for categorical variables using epidemiological 2 x 2 tables, and for continuous variables using difference of means. There were 59 errors among 492 cervical spine fractures in a total of 327 patients. Fifty-seven of the errors were on computed tomography and 2 errors were on radiographs. Upper cervical fractures were significantly more likely to have been errors than lower cervical fractures: risk ratio (RR) of 2.2 (confidence intervals (CI) 1.3, 3.5; P = .001). Occipital condyle fractures were more likely to have been discrepant: RR = 2.2 (CI 1.3, 3.9; P = .006). Dens fractures were also significantly more likely to have been discrepant: RR = 2.0 (CI 1.0, 3.8; P = .05). Other potential predictors were not associated with significantly increased risk. Upper cervical spine fractures, in particular occipital condyle and dens fractures were significantly associated with an increased relative risk of resident missing or misinterpreting the fracture. These findings suggest that resident education should focus in particular on upper cervical spine injuries, occipital condyle, and dens fractures. The methods used in this study could also be applied to other imaging modalities and anatomic regions in the future to target resident education to more challenging areas.