Measuring the accuracy of diagnostic imaging in symptomatic breast patients: team and individual performance

Abstract

The combination of mammography and/or ultrasound remains the mainstay in current breast cancer diagnosis. The aims of this study were to evaluate the reliability of standard breast imaging and individual radiologist performance and to explore ways that this can be improved. A total of 16,603 separate assessment episodes were undertaken on 13,958 patients referred to a specialist symptomatic breast clinic over a 6 year period. Each mammogram and ultrasound was reported prospectively using a five-point reporting scale and compared with final outcome. Mammographic sensitivity, specificity and receiver operating curve (ROC) area were 66.6%, 99.7% and 0.83, respectively. The sensitivity of mammography improved dramatically from 47.6 to 86.7% with increasing age. Overall ultrasound sensitivity, specificity and ROC area was 82.0%, 99.3% and 0.91, respectively. The sensitivity of ultrasound also improved dramatically with increasing age from 66.7 to 97.1%. Breast density also had a profound effect on imaging performance, with mammographic sensitivity falling from 90.1 to 45.9% and ultrasound sensitivity reducing from 95.2 to 72.0% with increasing breast density. The sensitivity ranges widely between radiologists (53.1-74.1% for mammography and 67.1-87.0% for ultrasound). Reporting sensitivity was strongly correlated with radiologist experience. Those radiologists with less experience (and lower sensitivity) were relatively more likely to report a cancer as indeterminate/uncertain. To improve radiology reporting performance, the sensitivity of cancer reporting should be closely monitored; there should be regular feedback from needle biopsy results and discussion of reporting classification with colleagues.