Thursday, August 9, 2018

The JAMA/Flatiron Study of Cancer Gene Panels: Why It Couldn't Have Worked

On August 7, 2018, JAMA published a large retrospective observational study of the relative impact of large gene panel testing in stage III/IV lung cancer in a community population (Presley et al.) along with a detailed Op Ed by Bunn et al.   See also coverage and interviews on August 8 at Genomeweb (here).

From a Flatiron Heath database of some 5700 patients in 191 practices, the authors identified about 875 received panel testing (>30 genes) and 4800 received "routine" testing such as EGFR gene and ALK gene alone.  In propensity-matched analysis, there was no increase in 12-month survival in the gene panel group: survival was about 43% at one year in either group.

The study authors, and the Op Ed authors, note various caveats and limitations.  Stil, the study authors conclude with concerns there is limited validation here for the wide use of costly gene panel testing in lung cancer patients.    The Op Ed authors note that the number of testable genes and paired drugs is rapidly increasing, even since an area that spanned 2011-2016. so that results of panel testing in one several-year-period may not apply to the next period.

What's Wrong

The biggest issue for me is that an RCT which is pivoted on the diagnostic test can be a hopeless trial design in some situations, and this is probably one of them.

Surely if you take 200 advanced lung cancer patients, and pivot the design to test only for ALK in one half of them, it's unlikely to show a population wide survival benefit, since the ALK fusion will appear in only a few percent of patients in the second arm, and even then, a good drug may work only half or 2/3 the time. 

Case Study with a Diagnostic for Herceptin

It's like playing a movie backwards: this is exactly why the classical pivotal trial for Herceptin was run in Her2-positive women and not in all women

Let's pretend for the sake of real simple math that 30% of women are Her-2 positive, and that in advanced breast cancer it doubles your survival from one year to two years half the time, with half the patients being responders. 
So in a control arm of 100 patients (no Her2 testing), all will get chemo and all survive one year.    
In the Her2 test arm, 100 patients get Her 2 testing.   30 are positive, and 15 respond to drug and live 2 years; the other 85 in Arm Two live one year.    
Result:  In the Her2 test arm, the average survival of the 100 patients getting Her2 testing is 1.15 years, very unlikely to be different than 1 year, when you add in all the diverse noise of surival data in cancer patients.   Yet, you also already KNOW that if you treat the Her2 positive women with Herceptin, the responders will have double survival (2 years) and half of them are responders.
OK.  In the JAMA report this week, we've just run the vision of precision medicine backwards by pivoting the study on the diagnostic test.   You know some patients will do better in the gene panel test population, but they will be washed out in most situations in a randomized control trial pivoted on the precision medicine diagnostic test in the whole population.   This is the whole rationale for setting up precision medicine trials on the positive and negative populations, not the whole population, in the first place.

N.B.:  Authors Found Amazing Lack of Prescription Follow-Through

Another key result of the real world database study was that only about half of patients with a pivotal mutation (like EGFR or ALK) got the corresponding targeted therapy, in data running almost up to the present, to 2016. 

That's an issue that should be getting red-light attention and plans for an intervention.


Interesting citation.   From just a few weeks ago, June 29, the Op Ed authors cite Sharma et al in J Precis Oncol (here).  This is: Eye-Tracking Study to Enhance Usability of Molecular Diagnostics Reports in Cancer Precision Medicine.  It's a cognitive psychology study that concludes:  "Focused usability studies can help drive our understanding of the clinical workflow for use of molecular diagnostic tests in cancer care. This in turn can have major effects on quality of care, outcomes, costs, and patient satisfaction. This study demonstrates the use of specific usability techniques (eye tracking and think-aloud protocols) to help clinical laboratories improve MDX report design in a precision oncology treatment setting."


See a March 2018 study by Flatiron on real-world use of immunotherapy in lung cancer (pembrolizumab); Khozin et al., here.


See an August 2018 study by Marquart, Chen, Prasad, estimating that "7% of patients can benefit from genome driven oncology."  While this is arguably a low number, it also makes the point if you randomize cancer patients into two arms, treating one arm with "genome driven oncology," you won't see a population-average benefit.  Jama Oncol 4:1093.

Note in the figure the tiny % at the top of each paired left bar is exploded into the paired right bar.