Journal of the National Cancer Institute Advance Access originally published online on March 25, 2008
JNCI Journal of the National Cancer Institute 2008 100(7):456-461; doi:10.1093/jnci/djn095
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© Oxford University Press 2008.
NEWS |
Biomarker Developers Face Big Hurdles
Personalized medicine is the ultimate goal of modern cancer treatment. Its success depends on the availability of tumor markers that can be used to guide treatment. Yet few clinically useful biomarkers have been validated despite decades of intense effort.
To improve the success rate, individual scientists and formal committees are working to develop methods and guidelines for biomarker development. Their work aims to improve everything from clinical trial design and issues of statistical validation to specimen collection and the quality of manuscripts reporting study results. But even with these efforts, no one expects that the changes will be quick or easy.
The Difficult Task
"It is very difficult indeed to show that something you measure in the blood or in the tumor will predict reliably the benefit of a new treatment," said Marc Buyse, Ph.D., executive director and biostatistician at the International Drug Development Institute in Louvain-la-Neuve, Belgium. "It is a hard statistical problem, it is a hard clinical problem, and it is a hard biological problem."
Currently, nearly all published biomarker studies report a simple association between a clinical response and a biomarker or a change in a biomarker. "That is potentially useful," Buyse said. "But that is not the end of the story."
What's needed, he said, are randomized trials showing that a change in biomarker level induced by some treatment also translates into a change in a clinical endpoint produced by the same treatment. "If you haven't shown that, then in fact you haven't shown that changing the level of the biomarker will affect the clinical endpoint. You've only shown a correlation, and the world is full of correlations."
Part of the problem is that many studies are those of convenience, said Daniel Hayes, M.D., professor of internal medicine at the University of Michigan Medical School in Ann Arbor. Often one researcher has an assay, another has a set of samples in the freezer, and they decide to collaborate. "It is not good enough to just pull samples and run an assay and see if you get a P value of [less than] 0.05," Hayes said. In such a case, a low P value indicates only that the survival or progression-free survival curves (or another endpoint being measured) are likely to differ between the two patient groups. The value doesn't say whether the separation between the groups is useful for clinical decision making; they may be separate but still so close together as to not be clinically meaningful.
|
Both Buyse and Hayes say that even well-trained clinical and laboratory scientists seem to forget basic experimental principles when they collaborate on biomarker projects that inevitably span the two fields. And that tendency includes forgetting about good clinical trial methods, such as having a written protocol, adequate sample numbers, and good statistical analysis. "So it's not surprising, in fact, that the same errors occur with biomarkers today as was the case with drug trials 25 years ago," Buyse said.
Improving the Development
To obtain the quality of data required to validate a biomarker for clinical use, researchers need to test putative markers with the same rigorous approach that clinical scientists use to test new treatments, Hayes said. In fact, he and colleagues described the levels of evidence necessary to validate clinical biomarkers more than 10 years ago. Adapted from the levels of evidence used in other areas of medicine, their recommendations call for the use of comparative trials, preferably prospective randomized trials.
But that is a lot to demand in terms of both financial and patient resources, Hayes admits. "I am worried that I am creating a monster here," he said. "But in the absence of that, you end up having patients being treated based on marker results that you can't trust—and that's not good either."
Pharmaceutical companies will probably be willing to fund the studies necessary to identify biomarkers that predict response to targeted therapies, said William N. Hait, M.D., Ph.D., senior vice president of hematology and oncology for Johnson & Johnson in New Brunswick, N.J. In such situations, the drug developers have a sufficient incentive to pay for trials because positive results would make it more likely that insurers and others will continue paying for targeted therapies, which can cost tens of thousands of dollars per patient.
However, orphan markers—those that show whether a particular patient is likely to respond to therapy in general rather than to a specific therapy—will have a harder time finding support, Hayes said. So setting the standard as high as he suggests could actually mean that good markers aren't developed.
New Guidelines
Two groups are currently working on guidelines that could improve biomarker development. The AACR–FDA–NCI Cancer Biomarkers Collaborative is working on a report that describes what is necessary to incorporate a predictive biomarker into clinical trials and, eventually, into clinical practice. The group has focused on four aspects of biomarker development: assay validation, bioinformatics, specimen quality and collection, and data sharing.
Johnson & Johnson's Hait, who is involved in the collaboration as a participant and as president of the American Association for Cancer Research, expects the report to be released this spring. "It is going to be the definitive work on predictive biomarkers," he said. However, he noted, it will not address other types of biomarkers such as pharmacodynamic markers, which show the molecular effect of a drug, or prognostic markers, which provide information about the natural history of a patient's disease. "It is going to be fantastic, but it is only the smallest slice of the pie, and we could just about get our arms around that."
The biomarkers collaborative is part of the U.S. Food and Drug Administration's Critical Path Initiative. The agency expects to use the report to develop policies about what will be required for biomarker approval or use.
Meanwhile, researchers are also working to improve how biomarker studies are reported. Two years ago, Lisa McShane, Ph.D., of the biometric research branch of the NCI's Cancer Diagnosis Program and her colleagues published the REMARK (Reporting Recommendations for Tumor Marker Prognostic Studies) guidelines, which address the minimum standards for reporting biomarker studies. For example, the guidelines request that authors specifically state their hypothesis or the goal of the study and the basics of the study design, including sample selection, preparation, and statistical analysis. McShane and colleagues are now working on a follow-up manuscript that will explain and elaborate on the REMARK guidelines. Although it is too early to tell if the REMARK guidelines have improved reporting, McShane is hopeful. "There are some promising signs. The REMARK publications have now been cited over 100 times, so we know that people are becoming aware of them," she said. "I am hearing that some journals are now asking reviewers to check submitted papers for adherence to the REMARK guidelines."
Whether the collaborative report on biomarkers or publication guidelines will lead to more effective biomarker development is unknown. But Hayes, who regularly talks about the issue at meetings and other public venues, is adamant that the situation must improve.
"A bad tumor marker is every bit as harmful as a bad drug," he said. "For some reason, those of us who have been in this interface between the lab and the clinic, developing these markers, have forgotten this concept. For example, would you use a drug if you don't have good data about how the drug is useful, its efficacy, or its concentration? I suspect the answer would be no. And yet, every day we expect our pathologists to run tumor marker assays when we don't have a clue."
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||