Reproducibility of Wolfe's classification of mammographic parenchymal patterns

Inadequate reproducibility of Wolfe's classification of mammographic parenchymal patterns may explain its limited use in clinical and screening practice. If the misclassification of mammographic parenchymal pattern categories is substantial, inconsistencies in study results will occur, which could at least partially explain the frequent inability to replicate Wolfe's findings in studies of breast cancer. In this article, results of a study to determine whether consensus to resolve inconsistencies between raters would improve the level of concordance in mammographic parenchymal pattern categorization and thus help reduce misclassification are presented. METHODS. One hundred consecutive mammograms from a large screening clinic in New York City were classified independently by two expert mammographers on two separate occasions, 8 days apart. Coding was repeated in two consensus conferences several weeks later. Reliability was estimated by computing intraclass correlation coefficients.

Initially, the average intraobserver reliability for the four patterns was 0.68 and the interobserver reliability was 0.65. After consensus, reliability improved markedly to 0.88.

It is concluded that the added complication and cost of consensus ratings will be more than offset by a substantial increase in precision.