Improving Mammography Performance in Practice
Patricia Carney, PhD
Professor of Family Medicine
Associate Director for Population Studies
Oregon Health & Science University Cancer Institute
What's the problem?
Mammography is not a perfect test, partly because of the complex architecture of the
breast tissue being imaged and partly because the technology is imperfect. Moreover,
abnormalities are a rare event, with about 4 to 6 occurring in every 1,000 mammograms.
Accurately reading and interpreting screening mammograms is therefore an important
challenge for radiologists.
Shortcomings in human decision making, coupled with complicated practice settings, also
influence physicians' interpretive performance. Despite the proven value of mammography,
its efficacy depends in large part on radiologists' interpretative skills, and
radiologists differ substantially in their interpretations.
Although previous studies have shown marked differences in radiologists' interpretive
behavior, we still have much to learn about how to use that knowledge to maximize breast
cancer screening performance in community settings. Key unanswered questions include:
What factors influence radiologists' behavior as they read and interpret mammograms? What
is the best way to systematically provide feedback to radiologists on their interpretive
performance? What kinds of interventions could best help radiologists change their
behavior to improve their interpretive performance, and ultimately, to improve screening
mammography in community settings?
[Return to top]
How will this research address the problem?
In 1994, the National Cancer Institute established the Breast Cancer Surveillance
Consortium (BCSC), a research resource for studies designed to assess the delivery and
quality of breast cancer screening and related patient outcomes in the United States. The
BCSC is a collaborative network of five mammography registries and two affiliated sites
with linkages to pathology and/or tumor registries. The network is supported by a central
Statistical Coordinating Center. As of May 2008, the Consortium's database contained
information on more than 2 million women, 7.5 million screening mammographic examinations
interpreted by more than 1,000 community radiologists, and 86,700 breast cancer cases.
Dr. Patty Carney, recently the principal investigator of the New Hampshire Mammography
Network (one of the BCSC sites) and currently a professor at Oregon Health & Science
University Cancer Institute, has long had an interest in physician behavior. In
particular, she has been interested in understanding the factors that influence physician
behavior and in using that knowledge to develop interventions aimed at improving
performance. She has worked both with physicians in training and physicians in
The BCSC has provided a rich harvest of data that Dr. Carney and investigators at the
other BCSC sites have used to understand the factors that influence radiologist behavior.
Studies using these data, for example, have shown extensive variability among radiologists
in their interpretive performance and accuracy. This work also has noted great
variability among existing audits, the reports that mammography facilities use to provide
feedback to radiologists.
Since September 2000, Dr. Carney and investigators at three other BCSC sites have been
collaborating with a team at Harborview Medical Center, in Seattle, led by Dr. Joann
Elmore on a project called Factors Affecting Variability of Radiologists (FAVOR). FAVOR
has used data from the BCSC registries in New Hampshire, Colorado, North Carolina, and
Washington State and a survey of radiologists to estimate the accuracy of mammography at
an individual level and to better understand the reasons for variability in
Using the BCSC registry data and the results from the radiologist survey, Drs. Carney
and Elmore and their research teams have developed and are now testing an interactive
web-based educational intervention. This intervention is designed to help radiologists
examine and understand their interpretive performance over time and determine whether and
how they can improve their own accuracy. When a radiologist logs onto the system, the
intervention automatically populates all fields with his or her individual data.
The FAVOR web-based intervention has three modules. Module 1, Outcome Audits, explains
the basic components of an radiologist audit report (i.e., screening results, recall
rates, biopsy yields, sensitivity, and specificity), and allows users to compare their own
audit results with the average results of other US radiologists and with national
benchmarks established by the Agency for Healthcare Research and Quality. Module 2, The
Influence of Breast Cancer Risk Perceptions on Mammography Interpretation, helps
radiologists learn the breast cancer risk factors of pre- and post-menopausal women whose
mammograms they have interpreted, assess how their interpretive performance may be
influenced by perceptions of risk, and consider whether to change their clinical practice
once breast cancer risk is clarified. Module 3, Malpractice and Mammography, helps
radiologists learn the prevalence of medical malpractice suits among radiologists who
interpret mammograms, understand how perception of this risk is higher than actual risk,
and assess whether malpractice concerns influence their practice.
The study is a randomized controlled trial with an early group of radiologists
undertaking the intervention first, and a late group serving as a control group.
Currently, the study is in the 9-month window between implementation of the early group
and the late group. In January 2009, the late group will receive the intervention and
data analysis will begin shortly thereafter.
[Return to top]
Significance of the study and results
Breast cancer is a significant cause of morbidity and mortality in the United States.
Until it can be prevented, the best approach includes mammography screening for early
detection. A better understanding of ways to enhance radiologists' mammography
interpretive performance is needed to optimize the impact of mammography screening.
One of the distinguishing features of this work is that it takes advantage of a large,
multi-site surveillance research resource, the BCSC, to develop a web-based intervention
that can directly influence the practice patterns of radiologists in community practice
through its predisposing, enabling, and reinforcing elements. This program predisposes
radiologists to change (by helping them understand current levels of performance and their
goals for achievement), enables change (by facilitating a revision of how they use audit
reports to change practice), and reinforces change (by providing support to ensure the
enabling components occur). Whether it actually results in change is important, because
many new imaging technologies are on the horizon. To the extent that they rely on
interpretive behavior of humans, we need to show that behavior can optimized if we hope to
achieve the potential impact of the old as well as new technologies.
[Return to top]
Recent publications of interest
Taplin S, Abraham L, Barlow WE, Fenton JJ, Berns EA, Carney PA, Cutter GR, Sickles EA, Carl D, Elmore JG.
Mammography facility characteristics associated with interpretive accuracy of screening mammography.
J Natl Cancer Inst 2008 Jun 18;100(12):876-87.
Baernstein A, Liss HK, Carney PA, Elmore JG.
Trends in study methods used in undergraduate medical education research, 1969-2007.
JAMA 2007 Sep 5;298(9):1038-45. Review.
Carney PA, Yi JP, Abraham LA, Miglioretti DL, Aiello EJ, Gerrity MS, Reisch L, Berns EA, Sickles EA, Elmore JG.
Reactions to uncertainty and the accuracy of diagnostic mammography.
J Gen Intern Med 2007 Feb;22(2):234-41.
Miglioretti DL, Smith-Bindman R, Abraham L, Brenner RJ, Carney PA, Bowles EJ, Buist DS, Elmore JG.
Radiologist characteristics associated with interpretive performance of diagnostic mammography.
J Natl Cancer Inst 2007 Dec 19;99(24):1854-63.
[Return to top]