In Review - News from the 2004 Meeting of the RSNA

Even veterans benefit from double-read mammography

By: Deborah R. Dakins

Highly experienced radiologists can overlook cancers, a fact that bedevils breast imagers. Techniques for bolstering detection include both double reading and computer-aided tools, but neither strategy is considered the definitive standard of care.

Previous studies have demonstrated the impact of double reading on detection rates by inexperienced radiologists. It turns out that even old hands at mammography can benefit from the practice.

Dr. Daniel Kopans, director of breast imaging at Massachusetts General Hospital, reported a 2.5% increase in breast cancer detection among experienced breast imagers, with only a fractional increase in patient callbacks, after the hospital instituted a double reading system.

The retrospective study evaluated 183,585 screening mammograms from 59,047 women acquired between 1995 and 2004. A primary staff radiologist read the studies and marked abnormalities requiring additional evaluation. A second staff radiologist quickly reviewed the cases, looking for abnormalities missed by the first reader. Concern by either reader resulted in a patient callback.

The average rate of callbacks by first-reader radiologists was 6.5%. An additional 0.27% of patients were recalled based on second-reader evaluations. The average for callbacks for both sets of readers totaled 6.8%.

The cancer detection rate of the first reader was about 4.07%, or about four/1000. Eighteen cancers were found by the second reader that were missed by the first, resulting in an average increased cancer detection rate of 2.5%.

"The use of double reading did increase our cancer detection rate, but at the expense of increased callbacks," Kopans said. "It goes to show that even experienced radiologists overlook cancers. But there is no reimbursement for double reading in the U.S., and it is not the standard of care."

As a result, double-reading programs must be cost-effective and efficient, he said. On average, a first reader at MGH takes about 90 minutes to review 60 or so cases. A second reader can complete that many studies in about five minutes.

Radiologists can receive an incremental payment for use of computer-aided detection when interpreting screening mammograms, and many practices now employ the technique. In one of the largest studies of CAD to date, it increased cancer detection by 20% in 13,000 women over a one-year period. (Radiology 2001;220:781-786).

But CAD is no substitute for an experienced radiologist's clinical judgment, said Dr. Diane Georgian-Smith, a breast imaging specialist at MGH who presented a study comparing double reading and CAD in a screening population.

"A CAD system is only as good as the radiologist who uses it," she said. "CAD alone cannot compensate for differences in clinical judgment."

Georgian-Smith's study compared CAD and double reading in 6000 consecutive screening mammograms over a two-year period. Eight radiologists with an average of 14 years of breast imaging experience participated in the study, which involved three separate interpretations of each mammogram by two radiologists, including a baseline without CAD, a second read with CAD, and a third interpretation by a radiologist blinded to the CAD results.

Callbacks resulting from the baseline reader accounted for 7.5% of all cases; reading again with CAD increased that percentage to 8%. Using CAD did not increase the cancer detection rate of two/1000. The second reader called back an additional 34 patients, resulting in the same 8% callback rate as CAD but yielding an additional two malignancies. A retrospective review, however, found that both those cancers had been marked by CAD but ignored by the baseline reader.

Ultimately, there was no significant difference in performance between the CAD reader and the second reader, Georgian-Smith said, but she expressed a personal preference for double reading.