In a new UC Davis study looking at 1.6 million mammograms taken across 90 facilities, Joshua Fenton and his co-authors found that computer-aided detection (CAD) in mammography actually increases false-positive readings, without significantly increasing detection rates of invasive breast cancers.
“The reason there is interest in having a second reader is because mammography is clearly an imperfect test. Twenty percent of the time when a woman has breast cancer, the mammogram is read as negative, as normal,” said Fenton, lead researcher and assistant professor at the UC Davis department of family and community medicine.
The detection software’s job is to analyze the mammograms and locate suspicious spots for radiologists to study more closely. However, it was found that radiologists were able to detect invasive breast cancers most of the time without the assistance of CAD.
The study, published in the August issue of the Journal of the National Cancer Institute, utilized test results that came from over 684,000 women from seven states, from 1998 to 2006. Of the 90 facilities that were examined, 25 had adopted the use of this new technology. Fenton found that after the detection software was installed, the rate of false positive readings increased from 8.1 percent before CAD to 8.6 percent after CAD.
Fenton noted that in the United Kingdom and other European countries, mammograms are typically read by two radiologists so that the second physician may catch suspicious signs that the first physician might have missed.
“The goal of having CAD would be that we might somehow automate a second read, and make mammography better than when a single reader does it,” Fenton said.
The technology was originally approved by the Food and Drug Administration (FDA) when some smaller studies showed that CAD utilization in addition to the radiologist’s reading could potentially locate more cancers without unacceptably raising false-positive readings. But Fenton’s study revealed that CAD was not associated with higher rate of breast cancer detection, or with earlier stage, size or lymph node status of the invasive cancer.
The current findings follow up on Fenton’s previous CAD-related study published in the New England Journal of Medicine in 2007. Critics of the previous study had argued that the findings were based on the use of the older CAD technology and didn’t accurately reflect the effectiveness of the newer CAD. This spurred Fenton to study the new technology from a greater sample, over a longer period of time.
“Conceivably, you could make a CAD program that works really well. But the actual technology that is widely disseminated in the United States, those that are currently FDA approved, don’t appear to work as well over practice as they seemed to have to work in the pre-marked studies for the FDA,” Fenton said.
In an accompanying editorial, Donald A. Berry, a biostatistician from University of Texas MD Anderson Cancer Center, agrees with Fenton’s conclusions.
“Researchers and device companies should work to make the software ever better,” Berry wrote. “But this should happen in an experimental setting and not while exposing millions of women to a technology that may be more harmful than it is beneficial.”
EVA TAN can be reached at science@theaggie.org.
