K. Kerlikowske et al., Variability and accuracy in mammographic interpretation using the AmericanCollege of Radiology Breast Imaging Reporting and Data System, J NAT CANC, 90(23), 1998, pp. 1801-1809
Background: Several studies, which were limited by their small sample size
and selection of difficult cases for review, have reported substantial vari
ability among radiologists in interpretation of mammographic examinations,
We have determined, in the largest study to date, intraobserver and interob
server agreement in interpreting screening mammography and accuracy of mamm
ography by use of the American College of Radiology Breast Imaging Reportin
g and Data System (BI-RADS), Methods: The mammographic examinations were ra
ndomly selected on the basis of original mammographic interpretation and ca
ncer outcome from 71713 screening examinations performed by the Mobile Mamm
ography Screening Program of the University of California, San Francisco, d
uring the period from April 1985 through February 1995, The final sample in
cluded 786 abnormal examinations with no cancer detected, 267 abnormal exam
inations with cancer detected, and 1563 normal examinations, Films were rea
d separately by two radiologists according to BI-RADS, Cancer status was de
termined by contacting women's physicians and by linkage to the regional Su
rveillance, Epidemiology, and End Results Program, Results: There was moder
ate agreement between radiologists in reporting the presence of a finding w
hen cancer was present (kappa = 0.54) and substantial agreement when cancer
was not present (kappa = 0.62), Agreement was moderate in assigning one of
the five assessment categories but was statistically significantly lower w
hen cancer was present relative to when cancer was not present (kappa = 0.4
6 versus 0.56; two-sided P = .02), Agreement for reporting the presence of
a finding and mammographic assessment was twofold more likely for examinati
ons with less dense breasts, Agreement was higher on repeat readings by the
same radiologists than between radiologists. The sensitivity of mammograph
y was lower with BI-RADS than with the original system for mammographic int
erpretation, but the positive predictive value of mammography was higher, C
onclusion: Considerable variability in interpretation of mammographic exami
nations exists; this variability and the accuracy of mammography are neithe
r improved nor diminished with use of BI-RADS.