This article presents a case study of 4 contemporary examples of personalized medicine biomarkers in colorectal cancer, describing the available clinical evidence for each example. All 4 of these biomarkers have been evaluated for inclusion in the wide-reaching NCCN Guidelines. Based on this structured review, the authors conclude that the level of published clinical evidence for these biomarkers is variable, and in some cases, discordant in content. This finding suggests that, by necessity, the domains of expert experience, clinical judgment, and consensus may play a greater role than published clinical trial data in guideline development for new personalized medicine biomarkers.
Reliance on expert opinion may be both a strength and a potential limitation when evaluating new technologies with rapidly evolving data, such as biomarkers. On the one hand, reliance on expert opinion enables timely review and incorporation of new information, resulting in the most up-to-date, accessible, and useful guidelines for stakeholders who must assimilate new technologies. The above-discussed qualification in B-raf
recommendations soon after presentation of new data reflects the dynamic and adaptive nature of the NCCN Guidelines. Other guidelines that rely on formal systematic review of the evidence are more laborious, require mature data sets and studies, and are therefore slow to respond to new data, rendering them less useful to practitioners.26
In a recent survey of 459 breast cancer surgeons, expert opinion followed by guidelines and consensus statements have been shown to have the strongest influence on decision-making in areas of scientific uncertainty. 22
Conversely, however, the strong reliance on expert opinion introduces potential for bias. The NCCN Guidelines development process uses rigorous safeguards, including strict conflict of interest disclosure requirements, an iterative process with review by and input from practitioners at member institutions, and inclusion of panel members representing diverse specialties and viewpoints.29,110
Among the 4 examples studied, the 3 tests that have been recommended by the NCCN Panel for Colon Cancer (K-ras
mutation analyses and MMR testing) are not proprietary to any single commercial entity, unlike the Oncotype
DX Colon Cancer Assay, which has not been adopted by the guidelines.
These measures, however, are not protective of subtle factors that might influence the uptake of new personalized medicine technologies into oncology guidelines. For example, in the case of the new B-raf mutation analysis recommendation, it is possible that the momentum and enthusiasm generated by the K-ras biomarker discovery influenced panel members’ impression of level of evidence and likelihood of improvement in patient outcomes from the related downstream biomarker, B-raf V600E mutation. The threshold for biomarker recommendation may also vary by the type of malignancy because of differences in research funding, patient advocacy, the risk inherent to the specific tumor type, treatment efficacy and toxicity, and subspecialty bias, factors that are not addressed by this case study.
This case study highlights several other intriguing aspects of biomarker evaluation and uptake by these influential guidelines. First, the examples of K-ras
mutation analysis, MMR testing, and the Oncotype
DX Colon Cancer Assay highlight the challenge of applying a uniform methodology to categorize the variable level of data generally associated with biomarker studies.1,111
Across the 4 tests, the clinical evidence consists largely of retrospective subset analyses of patient subsets derived from case series, cohorts, and prospective, randomized trials. The current NCCN Categories of Evidence and Consensus ratings do not clearly discriminate differences in level of available evidence for K-ras
, and MMR testing. All 3 of these recommendations are assessed as Category 2A level of evidence and consensus, whereas this structured review suggests a significantly lower level of evidence for B-raf
testing compared with the other 2 tests at the time of its inclusion in the guidelines.
These examples also show that it is very difficult to show improvements in patient outcome, one of the domains included in the NCCN Guideline evaluation methodology, for biomarkers. Biomarker studies may require a different set of standardized end points than studies of therapeutic modalities.1
Inherent to the challenge of selecting appropriate end points for biomarker studies is the importance of determining whether a biomarker’s association with a specific outcome is because it is a predictive marker for response to the treatment being studied, or because it is a strong prognostic marker independent of treatment. Many studies claiming predictive value do not uniformly include a control arm to exclude the contamination of a strong prognostic marker, as exemplified by studies of both B-raf
mutation analysis and MMR testing. In the case of the B-raf
V600E mutation, the lack of controlled studies impedes clear designation of this marker as prognostic, predictive, or both; because of this, the initial NCCN Guideline recommendation for B-raf
interpretation has been amended to reflect the inconsistency suggested by new, controlled data sets.24,76
Another observation from this case study is that NCCN Guideline inclusion often follows presentation of preliminary data at a national conference such as the ASCO Annual Meeting or a subspecialty symposium. This coincident timing likely follows naturally from the maturation of available evidence and consensus expert opinion from both organizations. However, given the previously discussed importance of guideline inclusion in the decisions of stakeholders, including policymakers, payors, and practitioners, this association in timing may merit further study.
Finally, although this case study focuses specifically on personalized medicine biomarker tests, the findings may apply to other types of nascent technologies and treatments in oncology that may be available to practitioners before the supporting evidence base is complete.