Expertise vs evidence in assessment of breast biopsies: an atypical science.

Abstract

Breast biopsies are performed in 1.6 million women in the United States each year1 and yield results ranging from benign to atypical hyperplasia to carcinoma in situ to invasive cancer, eachwith specific implications for subsequent management. The critical tissue diagnosis from the anatomic pathologist directly determines patient management. That diagnosis is based on morphology, the relationship between cellular and architectural features, devoid of any molecular evidence. The accuracy of the pathologist’s diagnoses is relativelyunderstudiedand represents an important knowledge gap at a timewhenmedicine is becoming evermore evidence-based. In this issue of JAMA, Elmore and colleagues2 report an analysis to quantify the diagnostic discordance in the interpretationof breast biopsies amongpathologists in 8US states. A2-stepdesign involvedtheassemblyofa test setof singleglass slides of 240 breast biopsies that purposely oversampled for cases with atypia and ductal carcinoma in situ (DCIS) as well as cases fromwomenaged40 to49years and thosewithmammographicallydensebreasts. The referencediagnosis for each casewasderived fromaconsensuspanel of 3 expert breast pathologists who unanimously agreed on the diagnosis for 75% of cases in independent review and 90%of cases after using a modified Delphi approach. Four test sets of 60 breast biopsies each (drawn from a master collection of 240 specimens including 23 invasive cancer, 73 DCIS, 72 atypical hyperplasia or atypia, and 72 benign without atypia cases) were then distributed to 115 randomly selected participating pathologists whowere representativeof thegeneral populationofpathologistswith regard to experience, practice locationand size, and demographics. The key finding was that the overall concordance rate of diagnostic interpretations with the reference diagnosis for participating pathologists was 75.3% (95% CI, 73.4%-77.0%), identical to the initial level of unanimous agreement between the 3 expert pathologists. This ranged from 96% (95% CI, 94%-97%) for invasive cancer to 84% (95% CI, 82%-86%) for DCIS to 48% (95% CI, 44%-52%) for atypia to 87% (95% CI, 85%-89%) for benign lesions without atypia. Statistically significant differences in disagreement with the reference diagnosis were seen for biopsies for women with higher breast density (but not younger than 50 years) and for pathologists who interpret fewer weekly case volumes or work in smaller practices or nonacademic settings, although the absolute differences were generally low. These findings are disconcerting but perhaps not all together surprising. Morphologic diagnosis rests on broad acceptance of key criteria, some ofwhich are easily and specifically defined but many of which are very subtle and difficult to describe in words. Diagnoses that rest on well-defined criteria (eg, diagnosis of invasive breast cancer) are more likely to be observer-independent whereas interpretation of subtle criteria suchas architectural irregularity andnuclearpleomorphism will vary between observers as a function of training, experience, andperhaps innate cognitive skills.With these caveats it is reassuring that these practicing pathologists agreed with the reference diagnosis for virtually all diagnoses of invasive cancer and 87% of benign lesions without atypia. The greatest challenge lies in the diagnosis of atypia, which constitutes about 10%of breast biopsies performedeachyear and forwhich this study suggests that overinterpretationmay occur in 17% of cases. The findings fromthiswell-designed, conducted, andanalyzedstudyare importantandsuggest that improvementsneed to be made. However, the study does have some significant limitations. First, by design, the case mix does not represent that of a typical pathologist, and thepurposeful increased representation of challenging cases is likely to magnify diagnostic discordance. Second, interpretationwas basedon abiopsy on a single glass slide without the opportunity to examine or recutmore tissue.This too coulddecrease concordance, as tissue recuts are frequently performed in difficult cases. Third, participating pathologists did not have the option to consult withothers, aswouldbecommonindailypractice.Fourth,participating pathologists had unlimited time to arrive at a diagnosis, a luxury that they are unlikely to have in routine practice. Fifth, and perhaps most critically, there is no outcome information to suggest that the reference diagnosis as identified through the modified Delphi process between 3 experienced breast pathologists was the “correct” diagnosis. Nonetheless the study by Elmore et al should be a call to action for pathologists and breast cancer scientists on several fronts. A deeply concerning aspect of this study is the 4% miss in the diagnosis of invasive cancer and the 16% discordance in the diagnosis of DCIS, as specific treatment algorithms have demonstrated value for these 2 diagnoses. Also this study demonstrates some variability in performance of individual pathologists based on practice location and experience and suggests that pathologists in low-volume practices should consult with more expert colleagues about challenging cases. Just as the Eskimos have many words for snow, pathologists have many terms for atypia. It is incumbent on Related article page 1122 Opinion