When and to What End Do Pathologists Agree?

We are certainly pleased with the central message of the study by Wells et al. (1) that appears in this issue of the Journal which we take to be that pathologists are generally in agreement regarding diagnoses of benign and malignant breast disease It was also a pleasure to find that so many pathologists were willing to take part in a cooperative exercise regarding diagnostic agreement However, several points of clarification regarding the study and its conclusions are needed. These discussion points relate to special considerations of diagnostic implications for clinical management (2) and the performance of clinical trials. Indeed, a major use of this study may be to pinpoint those situations in which pathologists are likely to disagree so that we may target these elements for study. The practical question presented with every medical diagnosis is the diversity of the decision tree, and, fortunately, for most breast biopsies, that is straightforward, because most of the lesions are benign as was the case in this exercise conducted by Wells et al. (1). This dominance of benign diagnoses was particularly true in the premammographic era when large, well-developed lesions (e.g., fibroadenomas, cysts, cancers, etc.) made up the majority of the specimens submitted to a surgical pathologist (3). It is certainly refreshing to focus on agreement, not disagreement. The agreement presented in the study by Wells et al. (1) relates to diagnoses of “cancer-yes” versus “cancer-no,” a simple dichotomy. However, Wells et al. (1) document the difficulty in making an assignment of invasive or in situ status for some biopsy specimens, and the authors state that “there are situations when anything less than perfect agreement may be clinically unacceptable.” The category of atypia was accorded generic status by Wells et al., yet they did not reference the body of literature on the subject, which includes studies of histologic criteria linked to epidemiologically validated outcomes (4–7). Indeed, the paper by Schnitt et al. (5) is incorrectly cited as a study that documents diagnostic disagreement. In fact, that study was designed to target particularly difficult lesions of atypical hyperplasia, benign hyperplasias, and minimal carcinomas in situ, and the results indicated that there was greater than 90% agreement when the several pathologists involved used agreedupon criteria. Epidemiologists are now documenting different disease patterns for different patterns of atypical hyperplasia using these same criteria (7), which have also been adopted in large part by the breast cancer screening program in the U. K. Use of these criteria for the recognition and separation of atypical hyperplasia from minimal examples of ductal carcinoma in situ (DCIS) appears to foster interobserver agreement (5,6,8,9). Another way to look at disagreements has to do with biologic and clinical relevance. Since Wells et al. provide very little in the way of specifics regarding the type of specimen and the extent of the lesions for each of the different categories, we can only speculate about the levels of disagreement (and the reasons for them). Although at face value we might think that any disagreement between benign and malignant could not be tolerated in clinical practice, in actuality, atypical ductal hyperplasia and non comedo DCIS straddle the benign/malignant fence. For small examples of noncomedo DCIS, the biologic potential is probably more similar to atypical ductal hyperplasia than to lesions toward the other end of the spectrum of malignant potential (i.e., possibly comedo DCIS and certainly lesions in the malignant invasive category). We also might think that it should be very straightforward to separate noninvasive from invasive carcinoma, but consider the example of high-grade (comedo) DCIS with areas suggestive of microinvasion-such lesions are often greatly distorted by sclerosis, making the diagnosis of microinvasion quite difficult. In the vast majority of cases, microinvasive carcinoma behaves no differently from DCIS (10,11), suggesting that the clinical behavior may dictate the seriousness of a disagreement and reduce its importance in this special case. This consensus on benign versus malignant diagnoses was obtained without recourse to discussions of criteria or special circumstances, such as the small core biopsies in which less than a certain diagnosis might be rendered, and appropriately so for purposes of clinical management in this setting. Guidelines have been suggested that foster the acceptance of the need for wider excision if patterns of a typical ductal patterns of hyperplasia are present (12,13). Wells et al. document that agreement is more difficult in these smaller core needle samples, and we agree. Wells et al. make a major point that most diagnostic agreement exercises in histopathology have been conducted with the involvement of academic pathologists. However, the most demanding exercise of reliable histologic grading of invasive carcinomas has been demonstrated with the involvement of pathologists from general practices (14) as well as from the ranks of academia (15). We disagree with the conclusion that central review is not necessary for clinical trials, because this assumes that clinical trials will remain as they have been in the past, a measure of therapeutic effectiveness in a setting structured to avoid possible subtype confounders at the onset, with only the effect of the therapy being measured. Rather than accepting breast cancer as one disease, and being content with a 3%-4% error rate in diagnosis in a trial because the statistical validity of determining the therapeutic advantage is not altered (16), we really want to know what elements in the individual tumor will interact with the therapy to determine the outcome. This latter approach is a shift of paradigms from the oneness of breast cancer to an acceptance of its heterogeneity. The reference to National Surgical Adjuvant Breast and Bowel Project, using central review that is now questioned as being unnecessary by Wells et al. (1), has been criticized as being incomplete even with a central review, at least in the important category of DCIS (17). Indeed, a current intergroup trial led by the Eastern Cooperative Oncology Group (trial E5194) is using central review because of the importance of reliable determination of grade, size, and margins in this disease. The area of DCIS appropriately emphasized as of increasing importance by Wells et al. is rapidly entering a phase in which validation of subcategories to guide various therapeutic options (9,18,19) is going beyond the consideration of invasive versus in situ disease presented by Wells et al. (1). There is one resounding truth in all of this: complete reliance is given to pathologic diagnoses-they are the basis of the major surveys of cancer incidence (20) and basic therapeutic decisions. Pathologists are properly sensitive and responsive to the needs of clinicians, and the discipline of surgical pathology is in constant and dynamic change as patterns of disease and therapeutic options change. Mammography has brought challenges to pathologists and all others involved in managing breast disease because small, unusual lesions have replaced the more obvious benign or malignant lesions that comprised the bulk of diagnoses in an earlier era.