Despite their growing popularity in the research community, DNA microarray-based gene expression assays
have not yet been approved or licensed by the U.S. Food and Drug Administration for clinical or ...regulatory
decision-making. While recent publications demonstrated the technical reliability of microarray technology,
there are several challenges that still need to be addressed so that microarray-based gene expression data
can be accepted for routine use in clinical and regulatory environments. First, appropriate quality-control
metrics and thresholds are needed for objectively assessing the quality of microarray data from individual
laboratories. Secondly, consensus on the analysis of microarray data is needed. Thirdly, adequate evaluation
and validation of microarray results is needed so that non-reproducible chance correlations can be avoided
in handling high-dimensional microarray data.
Felix Frueh, PhD: I'll make just a couple of observations about the question itself, which I found quite interesting. First, you ask about healthcare product development, and I think it's important ...that we don't equate that only with drugs. There are many healthcare products, and therefore many venues that can actually look at tailoring something new to individual patients and identify subgroups. Second, you mentioned effectiveness rather than efficacy. I think that's another critically important distinction. When you're developing a package to submit to the FDA for regulatory approval, you're demonstrating clinical utility, but not necessarily clinical effectiveness. Often, information on effectiveness comes much later, when we actually realize how a new medical product performs in a real-world setting. Third, the question focuses on the development process. If you look at the pipeline, I believe that about two-thirds or so of all drugs that are currently being developed, be that small molecules as well as biologies, have a biomarker associated with them in one form or another. So in that sense, I think that the advent of personalization, or the more precise delivery of care, is currently part of most ongoing development processes. Frueh: I have a slightly different opinion. I think current da- tabases can be used for personalized healthcare in certain cir- cumstances. They're certainly not designed for this particular use, but the data that's contained might in fact be very useful to at least create hypotheses that you can then take forward and either confirm or disprove. For example, when the reports came out that Plavix® (Clopidogrel) was being metabolized predominately through an enzyme called cytochrome P450 2C19, we didn't really have a lot of genetic information to fig- ure out whether or not the presence of this enzyme translates into different patient outcomes. What we did have, however, was drug-drug interaction information, where we could look at drugs that interact with this pathway and mimic the genet- ics that we're interested in. So by pheno-copying genetics and looking into longitudinal pharmacy databases-and then outcomes databases to see whether or not patients who were on both drugs have worse outcomes than patients who are only on Plavix-we actually had a proxy for the impact genet- ics could have. If we consider that interaction as a hypothesis that we could then take forward and prove valid or not valid in clinical practice, I think it's that alternative use that we have to respect when looking at the current databases. That doesn't mean that we have the perfect system. I completely agree that there are new types of databases, more complex databases, and databases in particular that address molecular features of individual patients that we need to look at, but they're being created as we're moving along. I don't think that you're going to go out there and just design one database that becomes your personalized healthcare database for an individual pa- tient. There are going to be various different iterations. I personally have a problem with PCORI leaving econom- ics completely out of the equation if we're talking about comparative effectiveness. You can't just look at one bucket versus the other bucket, and not look at whether or not it's ec- onomically feasible to use any one of these buckets in the real world. I think there needs to be a reality check that is based on the economic environment we're working in. And I just find it peculiar that, while we're talking about healthcare re- form focused predominately on cost, we're creating a PCORI that leaves the cost part completely out of the equation. Third, I think PCORI should work together with payers to develop better processes for coverage with evidence development, pragmatic clinical trials, and other evidence development schema. This is going to be how we learn the best methods to make this happen from an evidence stand- point. And finally, PCORI needs to work within the context of demonstration projects to understand what personalized medicine means within the microcosms of healthcare. It's not just BRAF to BRAF-positive melanoma; it's really BRAF inhibitors in the context of melanoma that's got an appropri- ate BRAF mutation for a particular person where we've got the right line of therapy, where we understand other issues around this person's tumor, and we understand issues in terms of toxicity over time, as well as how far they live from a place that can give them that particular drug, etc. There are a lot of things we have to do in order to figure out how to take care of this particular person within this particular context, and that requires demonstration projects.
Current advances in genomics, proteomics, and metabonomics would result in a constellation of benefits in human health. Classification applying supervised learning methods to omics data as one of the ...molecular classification approaches has enjoyed its growing role in clinical application. However, the utility of a molecular classifier will not be fully appreciated unless its quality is carefully validated. A clinical omics data is usually noisy with the number of independent variables far more than the number of subjects and, possibly, with a skewed subject distribution. Given that, the consensus approach holds an advantage over a single classifier. Thus, the focus of this review is mainly placed on how validating a molecular classifier using Decision Forest (DF), a robust consensus approach. We recommended that a molecular classifier has to be assessed with respect to overall prediction accuracy, prediction confidence and chance correlation, which can be readily achieved in DF. The commonalities and differences between external validation and cross-validation are also discussed for perspective use of these methods to validate a DF classifier. In addition, the advantages of using consensus approaches for identification of potential biomarkers are also rationalized. Although specific DF examples are used in this review, the provided rationales and recommendations should be equally applicable to other consensus methods.
