Mammographic Breast Density Assessment: A Methods Study

Abstract

This pilot study was designed to find the best way to charac terize mammographic breast density for future epidemiologic studies. Fifty mammograms (24 high, 6 low, and 20 intermediate density mammograms, as defined by the original radiologist reading) were selected from women participating in a community medical surveillance program. Women were required to be Caucasian, 40-85 years old and have a Body Mass Index ≤30 at the time of the mammogram; and to have no history of breast cancer. Mammograms were selected by stratified random sampling on the year of mammogram and degree of density. The agreement between breast density category assignment using different methods was assessed using the kappa statistic. Substantial to perfect agreement was observed between density categories as determined from two radiologists’ readings (kappa statistic [K]=0.73). Perfect agreement was noted between two radiologists for high vs. intermediate (K=0.81) and high vs. low density combinations (K=0.86). When grouped by the year of the mammogram, substantial agreement was observed for all year intervals (K=0.77, 0.69, and 0.70). The agreement between density group assignment using information from digital images and using codes applied to the radiologist reading showed fair to moderate agreement (K 0.19-0.36). Using the radiologist reading for characterization of breast density from earlier conventional mammography films is better than using density quantification from digitized mammograms.