College of American Pathologists Conference XXXV: Solid Tumor Prognostic Factors—Which, How and So What?: Summary Document and Recommendations for Implementation

In 1994, the College of American Pathologists (CAP) and its Cancer Committee sponsored a conference to address which prognostic factors were clinically useful in the examination of breast, prostate, and colon cancers. The conference was convened to examine levels of evidence supporting various prognostic factors, so that recommendations could be made to practicing pathologists about which prognostic factors were useful clinically. One hundred invited participants met for 2½ days to address these subjects. Following brief presentations of marker categories, the criteria needed to establish clinical utility were established (Table 1). Conferees developed consensus about existing prognostic factors, based on expert opinion, empirical evidence, and information about utility, accuracy, and reliability of current testing methods. At the conclusion of the conference, specific recommendations were made concerning which factors had proven clinical utility, and participants ranked other factors on the basis of levels of evidence and availability of standardized testing methods.1 

Since that time, literally thousands of other prognostic factors have been reported. However, since most studies reporting these factors have involved small heterogeneous groups of patients, such reports have increased the confusion and chaos in this area of pathology practice. As the Cancer Committee began deliberating about the importance of a second conference to address this topic, we realized that factors of proven utility for predicting outcome or response to therapy had not significantly changed in the intervening 5 years, despite the availability of many assays and numerous published studies on many factors. No new prognostic factors had achieved proven clinical utility in that extended period of time. In reflecting on this observation, the Cancer Committee determined that stagnation of the field of prognostic factors had occurred because the amount and quality of data concerning prognostic factors was so poor that no consensus of value could be achieved. Prognostic factors were performed by a variety of assays, using a variety of reagents, interpreted according to variable criteria, and reported in such a variable manner that data could not be collected systematically for evaluation. The purpose of the conference, as designed by the Cancer Committee, was to create a forum to address issues of variation in methods, interpretation, and reporting of prognostic factors, so that greater clarity of the value of these markers could be achieved. A multidisciplinary planning committee was convened to address this goal, as well as to reiterate which prognostic factors were of proven or promising value, based on the criteria elucidated in the first prognostic factor conference held in 1994.

Clinical and pathologic prognostic factors have been used to predict patient outcomes for many years. Such information serves many purposes; it is used to elucidate the natural history of cancer, to predict the results of therapeutic interventions, to identify homogeneous patient populations, to compare results of treatment, to identify subsets of patients with unfavorable outcome, and to plan follow-up strategies.

The most widely used prognostic factors are the stage groupings used in the TNM staging system, in which information about the local extent of disease (T category) is combined with information about the lymph node status (N category) and evidence of metastatic disease (M category). Based on the correlation with survival, the stage grouping provides a single statement of outcome (a single prognostic message), since each patient is assigned a stage based on the combination of the TNM variables.2,3 This system has achieved worldwide acceptance and is generally used to plan patient treatment and predict outcome. Other important prognostic factors are well known, including histologic grade, vascular or lymphatic invasion, mitotic index, performance status, symptoms, and, most recently, molecular and biochemical markers of disease status; however, a method to incorporate these other factors with the TNM staging system has been elusive. Current statistical methods are poorly equipped to handle the vast amount of data necessary for comprehensive evaluation and integration of factors.4,5 Newer statistical tools, such as neural networks, have been proposed as solutions to this problem, but these devices require the availability of large data sets.6 Because of the variability of prognostic factor information, such large data sets are not yet available.

Because of the plethora of candidate prognostic factors, criteria to assess their value are of paramount importance. Such criteria for identifying important prognostic factors were established by the American Joint Committee on Cancer (AJCC) and the International Union Against Cancer (UICC). The criteria include clinical importance (the factor is a powerful predictor that can be used in patient management), independence (the prognostic factor retains its prognostic value when other prognostic factors are combined with it), and significance (the predictive accuracy provided by the factor rarely occurs by chance).6 These criteria were used in both the previous and current CAP conferences to determine which factors should be considered in the discussion.

The CAP authorized the Cancer Committee to convene a planning committee to design this conference. The conference was limited to consideration of prognostic factors important in the clinical management of patients with colon, prostate, and breast cancer. It was designed to be an open conference to promote broad participation of pathologists, clinicians, industry representatives, and organizations concerned with cancer management. A panel of pathologists, surgeons, medical oncologists, radiation oncologists, and statisticians was selected for each site to organize working group sessions. Each group was asked to limit their considerations to tissue and serum factors detectable at the initial diagnosis of sporadic cancers. Factors useful in determining patient outcome (prognostic factors) and factors useful in determining responsiveness to various interventions (predictive factors) were both included. The expert panels met several times in telephone conferences to plan the working group sessions. Panel members were asked to evaluate and rank various prognostic factors using the AJCC criteria and the categories describing the clinical utility based on levels of evidence used in the previous CAP conference (Table 1). Before the conference, each panel created a draft document that described the nature of problems with methods, the interpretation or the reporting of each factor that limited its current utility, and recommendations for solutions. These documents formed the framework for the discussion in working group sessions. Finally, each expert panel was asked to consider strategies that would permit the implementation of the recommendations proposed. Individuals were selected from each panel to present these topics at the conference for discussion in the working group sessions. Participants of the conference selected working groups in which to participate and contribute to the discussion and consensus development.

The conference was organized to promote discussion, broad input, and future planning. An initial plenary session provided an overview of the status of tumor marker research, the impact of variation in medicine and pathology, and statistical issues related to prognostic factor research. Working group sessions for each cancer type interactively evaluated and refined the documents created by the expert panels. A second plenary session dealt with issues common to all 3 groups, including the problem of micrometastasis in lymph nodes in these sites; statistical issues that arose during the working group discussions; and issues of variation in methods, interpretation, and reporting of immunohistochemical assays. A faculty session brainstormed strategies that could be used to implement the changes recommended. This session included invited representatives of the Food and Drug Administration, Health Care Financing Administration, Centers for Disease Control and Prevention, National Cancer Institute, AJCC, and UICC. Cancer site and general recommendations were presented and discussed during a final session to achieve consensus of the conference participants and to address feasibility of implementation of these recommendations. A final discussion focused on future initiatives that might lead to implementation of the changes proposed at the conference by the various organizations represented. This report summarizes the general conference recommendations, cancer working group recommendations, and plans for implementation of the recommendations. Detailed reports of the working group recommendations are found in accompanying reports.7–9 

Table 2 summarizes the categorization of the factors by the Breast Cancer Working Group as ranked at the 1994 conference compared to the current ranking (1999). The categories are essentially identical, although the designations have changed semantically. Mitotic counting achieved consensus recommendation by the current working group as the method of choice for determining proliferation. Because the size of breast cancers has decreased dramatically during the 5 intervening years, estrogen receptor/progesterone receptor determination by immunohistochemistry has become more acceptable as an alternative to the dextran-coated charcoal method. However, only a few clinical trials or large outcome studies have tested whether these assays are truly comparable. c-erbB2 (Her2/neu) testing has emerged as an important prognostic factor, but may be more useful as a predictor of response to therapy than as a prognostic factor. There is considerable controversy about the best method to determine protein overexpression; evaluation of gene copy number by fluorescence in situ hybridization is another method whose clinical role is yet to be determined. Ki-67, a proliferation antibody useful on frozen tissue, has been replaced by MIB-1, an antibody against the same epitopes that can be used on paraffinized tissue. Angiogenesis evaluation was removed from category II because the methodologic considerations have limited the usefulness of such testing. The working group focused much of its work on specific recommendations to standardize methods of cancer evaluation, interpretation, and reporting of the ranked prognostic factors. Table 3 summarizes these recommendations, which are discussed in detail elsewhere. Other recommendations common to all working groups will be summarized in this article.

Recommended categorization of prognostic factors by the Colorectal Cancer Working Group are shown in Table 4. The major recommendations for colorectal cancer are designed to standardize the examination, interpretation, and reporting of tumor staging information. Table 5 lists the specific recommendations. A unique recommendation relates to the discovery of adequate numbers of lymph nodes in radical resection specimens. Studies cited in the working group session indicated that the finding of at least 12 negative nodes predicts regional node negativity. Another significant recommendation was the proposal of a 2-grade system for histologic grading, based on the amount of gland formation in the tumor. This recommendation is made to improve uniformity of grading designations. A comparison of the ranking of factors from 1994 with those of 1999 illustrates the influence of molecular testing on prognostic factor research. In 1994, no molecularly derived factors were proposed as promising factors to be used clinically. In 1999, microsatellite instability and the histologic manifestations of this genetic alteration, as well as loss of heterozygosity at 18q (DCC [deleted in colon cancer] gene loss), are both considered very promising prognostic factors.

The Prostate Cancer Working Group devoted significant effort to discussion of prognostic versus predictive factors in prostate cancer. Recommended ranking of categories by the Prostate Cancer Working Group are shown in Table 6. The categorization of prognostic factors does not differ from the previous ranking of 1994, except that surgical margin status and tumor volume considerations were specifically addressed in these recommendations, and fewer factors are included in category II because few recently published studies supported the value of these factors as clinically important. Recommendations of the working group focused on improving the examination of specimens and the interpretation and reporting of prognostic factor information. These recommendations are shown in Table 7.

Statisticians played an active role in each of the working groups. In addition to their role in refining site recommendations, Drs Thomas Pajak, Gary Clark, and Daniel Sargent gave 2 illuminating plenary session presentations, a summary report of which is included in this issue.12 Summary recommendations are listed in Table 8. The most important issues that were raised concerned the development of consensus about which factors are most important and how multiple factors can be best analyzed. Publication guidelines were advocated to assure that prognostic factor reports contain elements that allow for statistically valid conclusions. Continued partnership of pathologists, clinicians, and statisticians will be needed to assure future progress.

Considerable discussion during working group and plenary sessions concerned the appropriate handling of micrometastases. It was agreed that a micrometastasis is a histologically detected focus of cancer in a lymph node equal to or smaller than 2 mm in maximal dimension. To detect micrometastases, it is recommended that all grossly uninvolved lymph nodes be entirely submitted for histologic sections and that at least 1 hematoxylin-eosin–stained section of each node be examined. Tumor foci smaller than 2 mm in maximal dimension that are detected only by immunohistochemical or molecular methods should be reported separately from those that are detected histologically. Such patients should be regarded as pN0 until current trials evaluating the importance of such findings are published. Working groups for breast, colorectal, and prostate concurred with these definitions.

Immunohistochemical methods are used to detect numerous prognostic factors, including estrogen receptor protein, progesterone receptor protein, Her2/neu protein, p53 protein, apoptotic proteins like Bcl-2, proliferation markers (MIB-1), and endothelial epitopes, such as factor VIII-related antigen, CD31, and CD34, which are used to stain the microvasculature and assess angiogenic potential. Methods used for prognostic factor evaluation are currently performed, interpreted, and reported in highly variable ways. For prognostic factor information to become clinically useful, consensus will need to be developed about optimal methods, best criteria to assess assay positivity, and recommended reporting formats. Conference participants agreed that a first step in achieving this goal would be adoption of uniform guidelines for use of reagent and methodologic controls, documentation of assay methods, guidelines for interpretation, and reporting standards, adapted from those advocated by Dr Clive Taylor in a recent report and in his accompanying editorial.13,14 Summarized recommendations are listed in Table 9. We hope that consensus by participants at the conference and publication of the recommendations will lead to adoptions of these guidelines.

The conference's final session was a lively discussion of panel and general recommendations. In this session, a list of strategies to promote adoption of the conference recommendations was discussed. It was recognized that these recommendations will only produce improvement if they are accepted and used in clinical practice. The strategies advocated to promote adoption of these recommendations follow.

Conference proceedings and protocol modifications need to be communicated effectively to pathologists. The proceedings of the conference are published in this issue of Archives of Pathology & Laboratory Medicine to achieve their recommendation. Publication of the summarized recommendations are planned as editorials or short submissions to Cancer, Journal of Clinical Oncology, and The International Journal of Radiation Oncology Biology and Physics, to assure that clinicians are also informed of the key recommendations. Symposia about the conference have also been planned, including sessions at the fall 1999 and 2000 American Society of Clinical Pathologists/CAP meetings. Summary presentations will occur at various specialty conferences dealing with prostate, breast, and colon cancer, as well as at the fall 1999 AJCC conference on use of novel statistical methods for prognostic factors.

Educational initiatives are needed for breast, colorectal, and prostate cancer so that recommendations can be understood and implemented. The CAP cancer protocols for breast, colorectal, and prostate cancer have been updated to conform to the recommendations and are included in this issue. Seminars and other educational offerings will be prepared to educate pathologists about how to initiate the recommendations. For example, the Prostate Working Group plans to have an educational symposium at the urologic companion meeting at the United States and Canadian Academy of Pathology meeting in March 2000. The Colorectal Working Group will write an article about how to perform the modified grading that they have recommended. Members of the Breast Working Group participated in a National Cancer Institute–sponsored workshop dealing with methodological issues related to HER2/neu protein testing, building on the principles espoused at the working group meeting. The CAP Cancer Committee has initiated a new educational offering, known as the Cancer Curriculum, which will provide a series of educational exercises dealing with conference recommendations in specific cancer sites, such as breast. The Cancer Committee, working with the Quality Practices Committee of the CAP, also plans to measure how well the recommendations have been incorporated into clinical pathology practice and to provide this information back to pathologists.

One important strategy to motivate individuals to use the recommendations, such as the statistical recommendations listed in Table 8, would be to incorporate them into the publication guidelines of pathology and cancer journals. For example, statistical recommendations could be prepared in the form of a checklist for editors and reviewers to make sure that elements are included in publications. Specific information that should be included concerning assay methods, interpretation of assay results, and format of assay reporting of prognostic factors could also be included in publication guidelines.

Conference recommendations involved standardizing how prognostic information would be reported for several key elements, such as the cancer volume in prostate, HER2/neu status in breast, and histologic grading in colon cancer. If these definitions are widely distributed, they can represent important improvements in the quality of data being gathered across diverse pathology practices. After such standardization occurs, partnering with tumor registries may help ensure that data are collected and maintained in a standard fashion. Finally, Standardized Nomenclature of Medicine (SNOMED) coding of prognostic factor information has been planned and is in the process of implementation by the CAP. A grant from the Centers for Disease Control and Prevention for the coding of cancer elements using SNOMED has been awarded to the CAP and will result in SNOMED coding of these elements. Funding for associated educational initiatives are part of this grant award.

The National Cancer Institute Cancer Diagnosis Program has begun a pathology informatics initiative to promote aggregation of cancer pathologic information and tissue acquisition. Such an effort will be enhanced by the use of standard data definitions, like the standard TNM staging system definitions that have led to international standardization of staging information. If fields for data can be clearly defined, data from diverse sources can be aggregated, allowing for analysis of larger population groups. This will speed the process of definitive evaluation of important prognostic factors.

Immunohistochemistry is conducted in a highly variable manner. Promoting standardization of methods, interpretation of assays, and reporting of information will be helpful, but reagent controls are also needed so that the quality of reagents can be determined and interlaboratory comparisons can be made. Such standardization will allow for aggregation of data by making larger patient populations available for inclusion. National Institute of Standards and Technology and the Food and Drug Administration are willing to work to develop reagent standards for prognostic factor testing, such as providing cell lines with standardized expression of estrogen and progesterone receptors. Vendor partnerships will be needed to test reagent controls and share the information with pathologists. Vendors have expressed interest in carrying out this type of reagent testing to promote the production of the highest quality of reagents. Finally, pathologists will need to be educated in the use of standardized reagents when they are available. For example, if cell lines are used as standards for HER2/neu testing, ways to promote production of cell line controls must be disseminated to practicing pathologists.

Each working group considered how multiple factors could be assessed and found that no uniform methods were available at this time. It will be important for statisticians to investigate new ways to evaluate data, and conferees recommended that the National Cancer Institute be encouraged to promote seminars or training grants to create statistical methods for such evaluations.

The participants recognized that making more rapid progress in prognostic factor research would require collaboration and participation by those organizations and individuals holding responsibility for various aspects of this field. Without a common agenda of what should be done and how to approach the problems, little concerted progress is likely. To this end, participants suggested that the CAP Cancer Committee petition the Board of Governors of the CAP to sponsor a day-long planning meeting to address how a common agenda and goals for prognostic factor implementation can be achieved. By including all organizations that have concern for the conduct of such studies, it is hoped that prioritization of the most important issues can occur. Without such planning, it is likely that the next 5 years will produce a similar dearth of progress in this field.

Conference XXXV on Solid Tumor Prognostic Factors generated specific recommendations as to how various prognostic factors should be evaluated, how they should be interpreted, and how they should be reported so that better data will be forthcoming in future years. For this information to have a meaningful impact on the clinical practice of pathology in our country, significant work will be required to implement the conference recommendations. If recommendations such as these are accepted and used, pathologists will be able to produce prognostic factor reports that will be clinically relevant. Since prognostic information will be analyzed and reported in a standardized manner, larger data sets should be available to promote prediction of therapeutic responses. Finally, the conference has produced a framework with which to evaluate new factors and to establish the relative value of those factors more quickly and effectively.

The authors thank Sheila Taube, PhD, for her helpful suggestions regarding the manuscript.