PURPOSE: To describe measures of mammography performance in a geographicall
y defined population and evaluate the interpreter's use of the Breast Imagi
ng Reporting and Data System (BI-RADS).
MATERIALS AND METHODS: Mammographic data from 47,651 screening and 6,152 di
agnostic examinations from November 1, 1996, to October 31, 1997, were link
ed to 1,572 pathologic results. Mammographic outcomes were based on BI-RADS
assessments and recommendations reported by the interpreting radiologist.
The consistency of BI-RADS recommendations was evaluated.
RESULTS: Screening mammography had a sensitivity of 72.4% (95% Cl: 66.4%, 7
8.4%), specificity of 97.3% (95% Cl: 97.25%, 97.4%), and positive predictiv
e value of 10.6% (95% Cl: 9.1%, 12.2%). Diagnostic mammography had higher s
ensitivity, 78.1% (95% Cl: 71.9%, 84.3%); lower specificity, 89.3% (950/0 C
l: 88.5%, 90.1%); and better positive predictive value, 17.1% (95% Cl: 14.5
%, 19.8%). The cancer detection rate with screening mammography was 3.3 per
1.000 women, with a biopsy yield of 22.4%, whereas the interval cancer rat
e was 1.2 per 1,000. Nearly 80% of screening-detected invasive malignancies
were node negative. The recall rate for screening mammography was 8.3%. Ul
trasonography was used in 3.5% of screening and 17.5% of diagnostic examina
tions. BI-RADS recommendations were generally consistent, except for probab
ly benign assessments.
CONCLUSION: The sensitivity of screening mammography in this population-bas
ed sample is lower than expected, although other performance indicators are
commendable. BI-RADS "probably benign" assessments are commonly misused.