SEATTLE – A high-tech tool now used on more than 90 percent of U.S. mammograms doesn't improve breast-cancer detection and may lead to missed diagnoses — all while adding at least $400 million to the nation's annual health care tab, a study by investigators in Boston and Seattle has found. Computer-aided-detection (CAD) for mammography, which aims to double-check radiologists' screening results, didn't improve accuracy by any measure, according to the largest study to date of the controversial tool, published Monday in the journal JAMA Internal Medicine.

"This is a perfect example of something that has taken off without adequate analysis of the harms and benefits," said Diana S.M. Buist of Seattle's Group Health Research Institute, which helped conduct the National Cancer Institute-funded study.

An editorial accompanying the study suggests that Congress should stop allowing Medicare to reimburse health care centers for the procedure, and a University of Washington expert says women may want to demand CAD-free mammograms.

"They should not be required to pay for it," said Dr. Joann G. Elmore, a professor of medicine who has studied CAD and specializes in diagnostic accuracy. She was not involved in the study.

Group Health investigators worked with lead author Dr. Constance Lehman of Massachusetts General Hospital in Boston to analyze more than 625,000 digital mammograms read by 271 radiologists at 66 sites across the United States between 2003 and 2009.

The study, which reviewed mammograms from nearly 324,000 women, found that radiologists detected cancer in about four of every 1,000 women — and invasive cancer in about three of those four — whether they looked at nearly half a million digital mammograms with CAD or nearly 130,000 without it.

Additionally, among 107 radiologists who interpreted mammogram results with and without CAD, the study found they were more likely to miss cancers when they used the computer-aided review. Radiologists' sensitivity, or the proportion of times they correctly identified cancer, was 83 percent when they used CAD — and nearly 90 percent without it.

"What this shows is that when they used CAD, they did worse," Buist said, adding that some radiologists may use the device as a crutch.