Level 2: Mid-level evidence
The use of computer-aided detection in screening mammograms was evaluated in a study of 222,135 women who had 429,345 screening mammograms at 43 separate facilities (N Engl J Med. 2007;356:1399-1409). The reference standard for breast cancer diagnosis was based on clinical diagnosis made within one year subsequent to screening mammography.
During the study, seven facilities (16%) introduced computer-aided detection. Comparing before vs. after computer-aided detection in these facilities, 84,900 vs. 31,186 screening mammograms were performed and 438 (0.516%) vs. 156 (0.5%) were associated with a breast cancer diagnosis within the subsequent year. Screening mammograms were associated with 80.4% vs. 84% sensitivity (not statistically significant), 90.2% vs. 87.2% specificity (P <.001), and 4.1% vs. 3.2% positive predictive value (P = .01).
Computer-aided detection was associated with a higher rate of breast biopsies performed per mammogram (1.47% vs. 1.76%, P <.001), but there was no significant difference in the rate of cancer detected per mammogram (0.415% vs. 0.42%).
With a 0.5% rate of breast cancer, a 3.6% increase in sensitivity and 3% decrease in specificity means that for every 100,000 screening mammograms, 18 more cancers will be detected, 2,985 more false-positive mammograms will occur, and 290 more breast biopsies will be performed. The original article reported only five more cancers (instead of 18) detected per 100,000 women, but this analysis does not adjust for the slight difference in breast cancer rates (0.516% vs. 0.5%) in the observational data.