Orac wrote a brilliant take down of Stanislaw Burzynski's "personalized gene-targeted cancer therapy" claims yesterday and in it, raised an important reservation about the term "personalized medicine" that I happen to share (emphasis mine):
At the risk of annoying some colleagues I know, I'm going to point out that I never really liked the term "personalized cancer therapy" or its many variants, for the simple reason that it always struck me as more of a marketing term than a scientifically meaningful description of what targeting therapy to the genetic makeup of a patient's tumor will eventually entail. In fact, I think I now prefer another term, which has been used by Cancer Research U.K., namely "stratified medicine." The reason is that what we as clinicians are doing when we "personalize" or "individualize" therapies isn't really "personalizing" the therapy so much as using various measurements and biomarkers to place patients into groups of patients who respond to specific therapies. What the modern version of "personalized" therapy is really doing is producing more and more groups of patients, each of which, is smaller than the last, to be matched to more and more therapies. Whether the groups will eventually reach an N of 1, I don't know, but that is the goal. Only then will we truly have "personalized medicine."
Promises that "personalized medicine"—a therapeutic course tailored exactly to your specific presentation of a disease—is just around the corner seem to follow every announcement of a newly discovered mutation or newly developed technology. We are already largely there when it comes to "personalized genomics". Direct to consumer companies like 23andMe genotype enough markers that each customer's profile is pretty much guaranteed to be unique, and reasonably low-cost whole genome sequencing should be available Any Day Now(TM). Couple these with other technologies (e.g., high-throughput proteomics assays), and I'd venture a fairly confident guess that we'll be able to place each patient into a group of one within a decade.
What I'm much less sure about is how useful being placed in an "N of 1" group will actually be. Suppose the day of the sub-$1000 genome has arrived, and we can easily obtain the full somatic mutation spectrum of a given patient's tumor and even manage to separate driver and passenger mutations. How exactly would the uniqueness of this spectrum help? Any mutation truly private to this patient (e.g., in a gene never before observed to be mutated in cancer) would, with very high probability, be an impossible therapeutic target. The utility of knowing a mutation spectrum comes from our ability to use it to place in the patient in multiple groups, each with N much, much greater than one. In other words, the reason why we know targeted therapies work is because we've seen mutations in the target gene before.
We should also remember that medicine, and biology more generally, are largely a statistics game. Knowing that a breast cancer patient has a mutated HER2 gene is not a 100%-guarantee that Herceptin treatment will be effective—it just makes it far more likely. Similarly, even if we knew all of the risk-conferring or causative variants for a complex trait (height, type-II diabetes, etc.), gene–gene and gene–environment interactions would basically destroy our ability to make definitive statements about an individual's phenotype, regardless of how unique that person's genotype spectrum might be. Like I said before, these sorts of statistics are applicable not because each person is in a group of one but because he/she belongs to a unique combination of groups of many.
All of this then is my long-winded way of saying let's stop (over)using "personalized medicine" and switch to using "stratified medicine" instead. Not only is "stratified medicine" a clearer, much better approximation of our true meaning, it also stops us from making promises that either can't be fulfilled or are much less useful than most people would imagine.
No comments:
Post a Comment