AR: More mammo experience = more accurate reads
With breast cancer as the second most common form of cancer in the world and mammography as the standard screening technique, the authors sought to investigate radiologist characteristics associated with better or worse performance in identifying breast lesions.
Warren M. Reed of Medical Radiation Sciences at the University of Sydney in Australia, wrote that while “much work has been done to explore the impact of technical features, such as acquisition devices, displays, and environmental conditions, relatively less work has focused on the relationship between diagnostic performance and expert reader practice and experience.”
The researchers measured sensitivity, specificity and Az values (area under receiver operating characteristic (ROC) curve) to assess the performance of 69 radiologists, each reading 50 cases. All radiologists read the same 50 cases and the same four images of each patient (200 mammography images total), using identical software, room and screen settings, etc. and blinded to all other readers’ findings.
All cases had been read before the study, with 30 percent of patients having been diagnosed with single malignant lesions, all of which were positively confirmed by biopsy. Moreover, all 15 of the positive findings had been missed by one of the original radiologists reading the cases, indicating that the positive cases in this study were of at least moderate difficulty to detect.
The researchers found that radiologists who read 2,000 or more mammography studies per year had significantly higher Az scores than radiologists who read 1,000 or fewer exams annually, demonstrating significant positive relationships between annual caseload and accuracy of reading. The study also discovered that more years of certification as a radiologist and more hours spent reading mammography each week were both associated with significantly higher scores.
The authors emphasized the implications of the study for the broader public. “Expectations of 100 percent cancer detection across a number of experts is unrealistic and, for a variety of reasons, a more realistic acceptable minimum value should inform public debate and policy documents.”
Reed and colleagues interpreted their findings as meaningful for public debate about establishing reading experience benchmarks for radiologists. They noted that, while in Canada, the U.K. and Australia, nationally recommended mammography experience criteria satisfy their findings, in the U.S., the recommended weekly caseload is only 480, well below the 2,000 threshold the researchers found to be associated with more accurate readings. The article cautioned that the findings and comments could inform, but should not set, public or political debate.
“The overall median ROC score at close to .85 demonstrates a good overall performance, but shows that across expert readers, 100 percent accuracy is not achieved and should not be expected by policy makers, judicial systems, or the public,” the authors concluded.