PURPOSE: To evaluate (a) the relationship between mammogram interpreta
tion and diagnosis of new breast cancer and (b) interprovider variatio
n in mammogram interpretation. MATERIALS AND METHODS: Interpretations
of screening mammograms (133,668 mammograms in 114,899 women) acquired
during 21 months in a large health maintenance organization were cate
gorized (categories 1-5) with use of a standard format. During 1 year
after mammography, new breast cancer was identified with use of claims
data. Interprovider variation in the categories read was evaluated, a
nd percentages of these categories were correlated with breast cancer
detection. RESULTS: Over the 21 months, 1,018 mammograms were followed
by a diagnosis of new breast cancer. The category of mammogram interp
retation was strongly associated with the diagnosis of new breast canc
er; in 47.5% cases of category 5 mammograms, breast cancer was diagnos
ed. There was substantial interprovider variation in the percentages o
f category 3, 4, or 5 mammograms read. The percentage of category 4 an
d 5 mammograms read correlated inversely with the likelihood of cancer
detection (Pearson correlation coefficient [r] = -.4778 after log-log
transformation, P < .001). CONCLUSION: A strong correlation existed b
etween a mammographic abnormality suggestive of cancer and its detecti
on; however, substantial interprovider variation in the reading of cat
egory 3, 4, and 5 mammograms and their positive predictive values exis
ted. Reduction of interprovider variation should improve quality of ca
re because the number of false-negative and false-positive mammograms
should decrease.