JNCI: Mammo CAD falls short
CAD software is used currently for analyzing three out of four mammograms in the U.S. and carries annual direct Medicare costs of more than $30 million.
“Despite broad acceptance and use, it is unclear if the benefits of CAD during screening mammography outweigh its potential risks and costs,” wrote Joshua J. Fenton, MD, at the University of California, Davis, and colleagues, who explained that the ideal system would detect high-risk cancers earlier and reduce the incidence of advanced cancer.
One concern about screening CAD, noted Fenton et al, is that it may shift focus to more indolent cancers, such as ductal carcinoma in situ (DCIS).
To learn whether CAD leads to more accurate reading of mammograms, Fenton and colleagues analyzed data from more than 1.6 million film screening mammograms carried out at 90 facilities that participated in the Breast Cancer Surveillance Consortium from 1998 to 2006.
Of 90 total facilities, 25 adopted CAD and used it for an average of 27.5 months during the study period. The researchers collected information on women who had mammograms with and without CAD, including whether they were diagnosed with breast cancer within a year of screening.
Among facilities that implemented CAD, the researchers found specificity decreased statistically significantly from 91.9 percent to 91.4 percent after CAD implementation, while the recall rate increased from 8.4 percent to 8.9 percent.
In contrast, facilities that never implemented CAD realized a statistically significant increase in specificity from 91.1 percent between 1998 and 2002 to 91.3 percent between 2003 and 2006, with the recall rate dropping from 9.3 percent in the first time period to 9.1 percent in the latter.
Meanwhile positive predictive value held steady at non-CAD facilities but statistically significantly decreased from 4.3 percent to 3.6 percent at CAD sites.
Sites that used CAD experienced a statistically significant decline in overall cancer detection from before to after CAD implementation, which was primarily attributed to a decline in the detection rate of invasive breast cancer. The DCIS detection rate remained stable.
The cancers detected using CAD were no more likely to be smaller or at a lower stage or to have less lymph node involvement than those detected without CAD, the authors reported. The results were the same after adjusting for patient age, breast density, use of hormone replacement therapy and other factors that might influence mammography findings.
“[The] results suggest a limited impact of CAD on breast cancer detection, particularly with respect to invasive breast cancer detection,” wrote Fenton and colleagues. … These findings raise concern that CAD, as currently implemented in clinical practice, may have little or no impact on breast cancer mortality, which may depend on earlier detection of breast cancer.”
The researchers noted, “Although our analyses may lack sufficient power to exclude a small benefit of CAD in terms of invasive breast cancer detection, the principal contribution of CAD may be increased detection of DCIS--a precancerous lesion with an ill-defined long-term prognosis.”
Data from the study may be used to develop future statistical models that could quantify various parameters such as: the potential for CAD to improve breast cancer mortality, possible overdiagnosis of DCIS, patient preference for earlier treatment of DCIS versus later treatment of invasive cancer, the harms of additional false positive mammograms and societal costs, Fenton and colleagues added.
Fenton et al acknowledged some limitations of the study including a lack of digital mammography data, the assumption that CAD was applied to all mammograms in study sites and a lack of information on specific CAD software used at each site.
Ultimately, the researchers concluded, “As currently implemented in U.S. practice, CAD appears to increase a woman’s risk of being recalled for further testing after screening mammography while yielding equivocal health benefits.”
In an accompanying editorial, Donald A. Berry, PhD, of the department of biostatistics at MD Anderson Cancer Center in Houston, explained that it is nearly impossible to detect an incremental mortality benefit of CAD even in a very large randomized study. That’s because the additional cancers found are unlikely to metastasize in the window of time between detection with the aid of CAD and non-CAD-aided detection.
Plus, he noted, CAD may highlight less aggressive tumors or interval tumors. Detecting these tumors may not impact mortality.
The editorialist concluded that researchers should work to make CAD software more useful, but that "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."