U.S. Army Medical Research and Materiel Command

June 9, 2005

NEW STRATEGIES AIM TO STRENGTHEN MAMMOGRAPHY ACCURACY

Points of Contact:
Lynn Rudinsky phone 212/886-2200, email lrudinsky@cooneywaters.com
In Philadelphia 6/8 - 6/11
Pennsylvania Convention Center, 215-418-2155
Gail Whitehead, Public Affairs Coordinator, phone (301) 619-7783, email gail.whitehead@amedd.army.mil

Research Results from the "Era of Hope "Department of Defense Breast Cancer Research Program Meeting

PHILADELPHIA, June 9, 2005 - While about eight of 10 breast biopsies turn out to be benign lesions, two studies presented at the "Era of Hope" Department of Defense Breast Cancer Research Program meeting describe unique approaches to strengthening the accuracy of mammography so fewer women receive unnecessary biopsies. In one study, researchers identified physician characteristics associated with more accurate mammogram interpretation, and another proposes the use of a computer algorithm to distinguish if a lesion previously identified by a radiologist for a biopsy is benign or malignant.

Physician Characteristics Can Predict Mammography Screening Accuracy
Physicians who have at least 25 years of experience, interpret 2,500 to 4,000 mammograms annually and have a practice focused on screening were found to more accurately interpret mammograms in research conducted at the University of California San Francisco.

Currently, there is an incredibly wide variation in how accurately mammograms are interpreted by physicians. In this study, physicians detected 77% of cancers on average, but for individual physicians the detection rate ranged from 29% to 97%. The average false positive rate was 10%, while the rate for individual physicians ranged from 1% to 29%.

While patient characteristics, such as age, family history and breast density, can determine how well mammograms work, the research team believed the profound variation in mammography accuracy also could be attributable to factors beyond the patient. They chose to focus their study on physician characteristics that could be associated with accuracy.

The researchers studied doctors' actual performance in interpreting more than 1.2 million mammograms, paired with data from statewide cancer registries in four states. Their study found that physicians who interpreted 2,500 to 4,000 mammograms annually were the most accurate and made the fewest false-positive identifications. Meanwhile, physicians who annually read fewer than 2,500 mammograms or more than 4,000 mammograms may tend to develop a lower threshold for identifying cancer, resulting in more false-positive identifications.

In addition to volume, two other physician characteristics were found to determine accuracy. Radiologists who focused their practice on screening, as opposed to diagnostic mammography, had a lower false-positive rate. And, the false-positive rate declined as physician experience - both age and years of practice - increased.

"Fewer false positives mean less anxiety for women and less expense through fewer unnecessary tests," said Philip Chu, a specialist in the department of radiology at the University of California San Francisco Medical Center.

Dr. Chu said the study findings suggest woman should make sure that before they make an appointment for a mammography they find out more about the facility where it will be performed.

"Find a big, busy facility," he advised. "That will mean a higher volume and a lot of experience reading mammograms."

Computer Algorithm May Reduce Unnecessary Biopsies
Researchers at Duke University have created a novel computer algorithm to distinguish whether a lesion previously identified by a radiologist for a biopsy is benign or malignant.

"The ultimate goal is to alleviate the pain and anxiety women suffer through the invasive biopsy procedure and to reduce the cost of unnecessary biopsy for the entire health care system," said Anna Bilska-Wolak, Ph.D., a research associate at the Duke Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center. In addition, she noted that surgical biopsies can cause distortion on future mammograms, which may make later breast cancer diagnosis more difficult.

The tool the researchers created is a likelihood ratio classifier (LRb), which is a computer algorithm. Using a 100% sensitivity cutoff threshold, it is able to predict if the mass is benign or malignant, based on characteristics of the mammogram - without missing any cancers. The computer tool's specificity is 26% - 44%.

"Originally this set of lesions had a zero percent specificity because all the cases were sent to biopsy as part of standard clinical care," Dr. Bilska-Wolak explained. "But by using the classifier we would have been able to spare between 26% and 44% of the lesions from biopsy."

Currently before a biopsy, the radiologist uses a standard lexicon of descriptors to describe the characteristics of the lesion as pictured on the mammogram. Taking it one step further, the radiologist could then feed those characteristics into the research team's computer tool to determine if the lesion is benign or malignant.

To test the tool, the research team initially ran 670 actual cases - 36% malignant and 64% benign - through the computer model to see if it could correctly determine the status of each lesion. On average, the tool identified all of the malignant lesions and also 32% of the benign lesions that could safely avoid a biopsy.

A preliminary evaluation then tested 150 more cases, with no further changes to the algorithm, and still received good results. It was able to identify 100% of the malignancies and 26% of the benign lesions that could safely avoid a recommended biopsy. Next, the research team plans to test cases from multiple medical centers to ensure results remain consistent.

"Going forward, a real challenge will be ensuring radiologists feel confident that the tool works," acknowledged Dr. Bilska-Wolak.

"Era of Hope" is a forum for the presentation of research supported by the U.S. Department of Defense's Breast Cancer Research Program (BCRP), an unprecedented partnership between the military, scientists, clinicians, and breast cancer survivors. Since 1992, the BCRP has been working to prevent and cure breast cancer by fostering new directions in research, addressing underserved populations and issues, encouraging the work of new and young scientists and inviting the voice of breast cancer survivors to be heard in all aspects of the program. One of many congressional research programs managed by the U.S. Army Medical Research and Materiel Command, the BCRP has received more than $1.8 billion to date from Congress for innovative breast cancer research.

# # #

"Physician Predictors of Mammographic Accuracy"
P Chu, R Smith-Bindman, D Miglioretti, C Quale, R Rosenberg, G Cutter, B Geller, P Bacchetti, E Sickles, K Kerlikowske
Concurrent Symposia Session V, Symposia 32: Saturday, June 11, 6:45 p.m. - 8:45 p.m., Room 203A/B

"A Likelihood Ratio Classifier for Computer-Aided Diagnosis in Mammography"
A Bilska-Wolak, CE Floyd
Poster Session: Saturday, June 9, 6:30 p.m. - 8:30 p.m., Exhibit Hall A