NEW ORLEANS—Steps can be taken to reduce the number of false-positive results with screening mammography, W. Phil Evans, MD, said at the American Society of Breast Disease annual meeting. “But false positives may not be such a catastrophe. Reducing the false-positive rate should never overshadow the fundamental goal of screening mammography—the early detection of clinically occult breast cancer,” said Dr. Evans, medical director, Susan G. Komen Breast Center, Baylor University Medical Center, and clinical associate professor of radiology, University of Texas Southwestern Medical School, Dallas.
A study by Elmore et al (N Engl J Med 338:1089, 1998) called for new methods to reduce the false-positive rate while maintaining sensitivity. The study found that among 870 screened women, 24% had at least one false-positive mammogram; 13%, at least one false-positive clinical breast exam; and 32%, at least one false-positive result for either test. Over 10 years, almost one third of women screened for breast cancer will have a false-positive test result.
False-positives add $33 to every $100 spent on screening. “But the bigger cost comes when cancer is not found,” Dr. Evans said. “The average cost of treating early-stage breast cancer is about $11,000 and for late-stage lesions, $140,000, with a reduction in survival from 96% to 20%.”
To reduce the number of false-positive mammograms leading to biopsy, there must be a way to reduce the recall of patients in whom cancer is not likely to be found, he said. “To do so, one should obtain previous mammograms for comparison, confidently and correctly identify summation artifacts (tissue superimposed upon itself so it incorrectly appears to be a mass), and learn to ignore subtle mammographic findings of doubtful significance,” Dr. Evans said.
One article reported 543 lesions of “doubtful significance” from 5,514 mammograms, almost half of which were calcifications. The lesions were evaluated about 30 months later, with the following results: 74% did not change, 21% were less apparent or absent, 4% were more apparent but not suspicious, and 1% (6 cases) were more suspicious.
Of these 6 “more suspicious”cases, 3 proved to be benign without biopsy. Of the 3 that were biopsied, one was a low-grade DCIS (Wolverton DE, Sickles EA: Am J Roentgenology 167:1041, 1996). “The conclusion was that it is almost impossible to prospectively identify lesions that turn into cancer,” Dr. Evans said.
To reduce the rate of requests for biopsies after full breast imaging workup, Dr. Evans said that the radiologist must learn to confidently and correctly identify those summation artifacts that require more than standard screening views by properly tailoring the diagnostic mammography. Doing spot compression views or changing the angle can sometimes obviate the need for biopsy, he added.
Periodic mammographic surveillance can also be substituted for tissue diagnosis for “probably benign” findings. Probably benign describes lesions that have a 98% chance of being noncancerous.
Another method for reducing the false-positive rate is the use of consensus double reading. Two radiologists independently interpret a set of screening mammograms, then meet to discuss all abnormal findings. They collectively discuss each case and reach a consensus decision regarding which patients to recall for further evaluation. Use of this method in one facility has resulted in 35% fewer and more appropriate recalls, with a 50% increase in the number of cancers detected at biopsy, Dr. Evans reported.
This technique does add to the cost of screening. “It takes about 4.5 additional hours of the radiologist’s time to consensus double read 75 mammograms, adding about $500 a day, or $100,000 annually,” he said.
Computer-Aided Detection
Systems for computer-aided detection, such as the ImageChecker System (R2 Technology, Los Altos, California) may improve interpretation. With this technology, computers digitize the mammogram, and a computer algorithm searches for calcifications or crossing radiating lines. Areas of concern are marked and reviewed by the radiologist.
A multicenter study has shown that computer-aided detection finds many lesions overlooked by human interpretation, he said. There is no increase in the overall recall rate but an increase in recalls of true positives.
