Doubts on Mammogram Software: Lower Accuracy May Mean Unneeded Tests, Report Says.

Apr 05, 03:53 PM

Current Headlines: By Carrie Peyton Dahlberg, The Sacramento Bee, Calif.

Apr. 5--An increasingly popular computer-aided way to read mammograms is less accurate than human analysis alone, leading more women to go through repeat tests and unnecessary biopsies, according to a study in today's New England Journal of Medicine.

For women, the study leaves unanswered a life-or-death question. Among the repeated tests, would some handful of lives be saved because a computer alerted radiologists to tumors they would have missed?

"The benefits of this technology at this point are uncertain," said UC Davis Dr. Joshua Fenton, first among 11 co-authors who studied mammogram results of more than 220,000 women.

About three more women of every 1,000 screened underwent a biopsy after their radiologists used the computer software.

That's nearly a 20 percent increase in biopsy rates.

By comparison, radiologists in the study found 4.15 cancers per 1,000 mammograms before adopting computer-aided detection, and 4.20 per 1,000 afterward -- too small a difference to be statistically significant.

The study will "surprise and disappoint" most mammographers, the journal said in an editorial.

The study is the latest in a flurry of recent medical news about breast exams and cancer, suggesting different exam schedules or different technologies depending on each woman's risk.

The bottom line, Fenton stressed, is that "women need to understand that the benefits of mammograms are there, and are important to take advantage of."

Among doctors, debate about cancer screening techniques is complicated by issues unrelated to the disease.

Companies that make screening equipment have lobbied Congress heavily for Medicare reimbursement, even when supporting evidence has been slight, Dr. Ferris Hall, a Harvard radiology professor, wrote in the medical journal editorial.

Radiologists are so wary of being sued over a missed tumor that some contend it's hard to attract young doctors to the specialty.

In that atmosphere, computer-aided detection spread rapidly since FDA approval in 1998.

The National Cancer Institute estimates it helps interpret 30 percent of U.S. mammograms.

In the Sacramento region, Sutter Health uses it for all mammograms and Kaiser uses it for about 15 percent.

It is not used by UC Davis Medical Center.

Computer programs are cropping up not just for breast cancer detection, but for finding colon polyps and lung nodules, said Dr. John Rego, the lead radiologist for Kaiser Permanente throughout Northern California.

The technology is intended to highlight potential areas that an expert should look at more closely -- not to replace human judgment, Rego said.

That sentiment was echoed by Dr. Michael Norton, director of breast imaging at Radiological Associates of Sacramento, which has used computer-aided detection, shorthanded as "CAD," since 2002.

"Reading mammograms is a lot like finding Waldo," Norton said. "All the CAD system does is highlight potentially worrisome areas. Then the radiologist makes the decision."

Radiological Associates, which reads all mammograms locally for the Sutter Health system, used to have each mammogram evaluated by two experts but now uses the computer and a single human reviewer.

Of the 400,000-plus mammograms researchers reviewed in the new study, Norton noted, only 31,000 were computer-aided, and many of those were done by less-experienced radiologists.

In the hands of experienced radiologists, Norton said, computer-aided detection did not increase calls for more testing in his group.

Kaiser, which has been gradually introducing the technology at new facilities and some existing ones, noticed an upswing in callbacks at first, but that subsided as radiologists got used to the program, Rego said.

In the Sacramento area, computer-aided detection is used by Kaiser at its Point West facility and in Lincoln.

UC Davis' medical system has opted against the computer-aided approach, said Dr. Karen Lindfors, a radiology professor and chief of breast imaging.

"It was a group decision among our faculty," said Lindfors.

Doctors worried that the software would turn up too many false positives, she said.

And they felt that given the level of expertise at the academic medical center, they didn't need extra backup from computers.

Lindfors said she wants to see more research into any possible benefit of having computer programs help read mammograms.

"Most women are very afraid of breast cancer," Lindfors said.

"If there is a chance that computer-assisted detection might see something the radiologist wouldn't see ... I think a lot of women would say, 'Oh I'll take my chance with that.' "

It's likely that plenty of women will get to take that chance.

In the journal editorial, Harvard radiologist Hall wrote that while the findings are a "substantial hit" to computer-aided detection, they're unlikely to stop a practice that showed early promise and nets an extra $20 per mammogram from Medicare.

-----

Copyright (c) 2007, The Sacramento Bee, Calif.

Distributed by McClatchy-Tribune Business News.

For reprints, email tmsreprints@permissionsgroup.com, call 800-374-7985 or 847-635-6550, send a fax to 847-635-6968, or write to The Permissions Group Inc., 1247 Milwaukee Ave., Suite 303, Glenview, IL 60025, USA.