An article published in Nature last month describes a recent trend in which “oncologists are starting to prescribe expensive new drugs that target the genetic profiles of their patients’ tumors, even when those treatments have not been approved for the particular cancer involved.” Such “off-label” use of drugs is nothing new. Claims have been made that more than 60% of prescriptions for cancer drugs in the US are for off-label uses.

But the stakes are definitely higher when personalized medicines are used off-label as compared to other drugs. Personalized medicines are expensive, and, as with all drugs, everyone but the drug company loses when they confer no benefit – with personalized drugs those losses are much greater.

Personalized medicines are also complex, which potentially decreases the likelihood that off-target uses are going to be successful. Giving a pancreatic cancer patient an EGFR inhibitor may seem like a good idea if EGFR is over-expressed in the patient’s tumor, but which downstream pathway(s) is one ultimately trying to inhibit, and what are the implications for mutations in key signaling genes (e.g. KRAS) in those pathway(s)?

That is not to say that the off-label use of personalized medicines is an intrinsically bad thing, and indeed the practice is clearly understandable for late-stage cancer patients whose physicians are trying to buy them a little extra time.

In fact, off-label uses can actually advance the practice of medicine, but only in cases where the patient’s genetic/proteomic profile is known, and where the outcome of the off-label effort is captured so that it can inform future patient care. Otherwise the knowledge gained from these N-of-1″ clinical trials is unable to contribute to improving the use of personalized therapies for other patients.

There are efforts underway to build databases that can capture this information, but they may simply be contributing to the fundamental problem in personalized medicine of unconnected data silos that individually have limited ability to inform the optimization of personalized medicine, when what is really needed is one massive silo that is accessible to all.

Where there is unambiguously a problem is when patients are getting personalized medicines without the necessary companion diagnostic testing having been performed in the first place. In such cases there is no ability to learn from the success or failure of an off-label use, and a prescription is just a shot in the dark that can never inform future therapeutic decision-making.

Understanding how often physicians test for biomarkers that are linked to personalized medicines in comparison to the prescription rates for those drugs can help us understand the scale of any potential “over-prescription” problem. Layering in the incidences for diseases that are treated by those drugs can additionally help to sort out the extent to which personalized medicines are even part of the therapeutic discussion.

We recently received from CMS the total reimbursement volume and amounts for all IVD CPT codes over the last five years for all Western states. I have so far analyzed the reimbursement volume for BRAF and KRAS tests and compared that to disease incidence data, and the analysis seems to suggest that as few as 48% of newly diagnosed melanoma patients are being tested for BRAF, and as few as 36% of newly diagnosed colorectal cancer patients are being tested for KRAS.

How many of those untested patients are receiving the drugs that target BRAF and KRAS? This is definitely the next question that needs answering and I am actively looking for the necessary drug prescription data. Knowing the prescription volume will help us to understand whether low testing rates are due to 1) physicians prescribing personalized medicines blindly; or 2) physicians not being aware of, or motivated by, personalized therapeutic options.

Stay tuned as there is much to learn in this area. And BTW, if you know of a good database of drug sales/prescription volume by state and by drug please contact me asap!