Computer systems designed to make mammograms more accurate turn out to make them less reliable, according to the largest study to evaluate the increasingly popular high-tech versions of the common test for breast cancer.
The study of more than 429,000 mammograms found that “computer-aided detection” systems did not help radiologists find more tumors and significantly increased the number of false alarms.
The researchers and others said the findings, published in today’s issue of the New England Journal of Medicine, show that the expensive systems should undergo more scrutiny.
“The goal of these systems was to make mammography better,” said Joshua Fenton of the University of California at Davis, who led the study. “Our study suggests that, if anything, they appear to be doing more harm than good.”
The study highlighted how new, more expensive technologies can proliferate before enough research has been done to evaluate them, Fenton and others said.
The findings are the latest in a flurry of recent developments that have renewed widespread debates about breast cancer screening. The American College of Physicians this week challenged the long-standing recommendation that all women in their 40s undergo mammograms every year or two, saying that the risks may outweigh the benefits for many. Last week, the American Cancer Society recommended for the first time that women at greatest risk of breast cancer have annual MRI scans, triggering similar arguments over that sensitive but expensive technology.
Together, the developments underscore that breast cancer screening remains imperfect. But several experts stressed that they hope the debate will not dissuade women from having the exams, noting that for the first time, fewer women sought mammograms last year than the year before.
“No screening test is perfect,” said Robert Smith, the society’s director of screening. “But we should never lose sight of the fact that mammography is an important tool in reducing the toll of the disease.”
Breast cancer strikes more than 212,000 U.S. women each year and kills more than 40,000, making it the second-leading type of cancer and No. 2 cancer killer among women.
Computer-aided detection systems use computer programs to analyze X-ray images, flagging suspicious areas for radiologists to scrutinize. The Food and Drug Administration approved the first system in 1998.
The new study marks the first large attempt to assess how well the computer-aided systems work. Researchers evaluated 429,345 mammograms of 222,135 women at 43 centers in Colorado, New Hampshire and Washington state between 1998 and 2002.
Overall, the systems correctly identified which women had tumors and which did not 87.2 percent of the time, compared with 90.2 percent for standard mammograms, the study found. The systems also led to 31 percent more women being called back for more tests and 20 percent more undergoing biopsies, but detected no more cancers.
The researchers estimate that for every additional woman who received a breast cancer diagnosis, 156 had unnecessary additional tests and 14 had unnecessary biopsies.