The use of computer-aided detection to help radiologists focus on suspicious mammographic findings actually seems to reduce the accuracy of interpretation, according to a New England Journal of Medicine study. The use of computer-aided detection to help radiologists focus on suspicious mammographic findings actually seems to reduce the accuracy of interpretation, according to a new report in The New England Journal of Medicine for April 5.

The US Food and Drug Administration approved the use of computer-aided detection in 1998, and since then, many radiology centers have adopted this technology. Whether it actually improves the accuracy of mammogram interpretation, however, was unclear.

To investigate, Dr. Joshua J. Fenton, from the UC Davis Health System in Sacramento, California, and colleagues analyzed data from 222,135 women who had received a screening mammogram between 1998 and 2002 at 43 facilities in the US. Of these women, 2351 were diagnosed with breast cancer within 1 year of screening.

Seven of the 43 facilities used computed-aided detection during the study period, the researchers note. At these centers, implementation of the technology was associated with a drop in diagnostic specificity from 90.2 per cent to 87.2 per cent and in the positive predictive value from 4.1 per cent to 3.2 per cent. Biopsy rates also rose by 19.7 per cent (p < 0.05 for all).

Diagnostic sensitivity rose from 80.4 per cent to 84.0 per cent with use computer-aided detection, but this change was not statistically significant. Similarly, the cancer-detection rate did not improve with implementation of the technology.

Analysis of data from all 43 facilities confirmed that use of computer-aided detection lowered diagnostic accuracy (area under the ROC is 0.871 vs. 0.919, p = 0.005).

In a related editorial, Dr. Ferris M. Hall, from Beth Israel Deaconess Medical Center in Boston, comments that while the current findings will not end the use of computer-aided detection for screening mammography, "they constitute a substantial hit to this technology."

He adds that "I recommend the conduct of larger, controlled studies of computer-aided detection that assess not only cancer diagnosis, but also the gold standard: mortality."

N Engl J Med 2007;356:1399-1409, 1464-1466.