Nexavar misses. But did it really?

News today that Nexavar flunked its' Phase III trial in lung cancer. (NSCLC)

I'll be curious to see the data present at ASCO, because I'm not sure that they really failed. While Nexavar missed the primary endpoint (overall survival (OS)), there was an improvement in the secondary endpoint (progression-free survival(PFS.))

(I've yet to see the relevant figures. All I can go on are news reports.)

Also: every patient in the Nexavar trials had failed at least two previous (standard) therapies, with some having failed three. For NSCLC, you're talking about surgery, radiation, and chemo (2 or 3 of the 3 therapies) before being treated with the experimental targeted drug.

NSCLC is a huge market (~250,000 annual cases in the US alone) but has not yet been broadly cracked by a targeted therapeutic. (Iressa had approval, and Xalkori has approval for use in a very small %age of patients. Tarceva and Avastin both have approval, but clinicians are skeptical and the economic argument is suspect.)

But we have every reason to believe that NSCLC can be impacted by targeting EGFR and/or VEGF - Sutent (Pfizer's competitor to Nexavar) is also in late stage NSCLC trials, and a variety of not-yet-approved targeted therapies are also likely to try to impact NSCLC. Because of the large number of annual cases, and the need to audition as a thrid-line therapy, perhaps we're just setting the the bar too high or setting these trials up for failure.



(please pardon the upcoming rant)

Independent of the judgement on Nexavar in NSCLC, I am very disappointed at the FDA mandate of a placebo arm for oncology trials using late-stage, terminal patients such as the Nexavar trial discussed above. I fully appreciate that to run the most scientifically bullet-proof trial requires a control arm, but what is the humanity in dosing late-stage cancer patients with a placebo? Not only are you potentially generating false hope for the patient and their family, it seems utterly wasteful. With nearly a million cases of NSCLC every year, and decades of data tracking patient outcomes, don't we know by now what the OS and PFS are for NSCLC patients at a given stage?

I'd like to see an FDA/NCI-led project to calculate baseline survival expectation across the most lethal cancers, with these standards to serve as stand-ins for placebo arms. I'm not suggesting relaxing approval standards (I will in a minute), but instead normalizing for humanitarian reasons and economic reasons. Even better, such an FDA/NCI study would be a springboard to genomic-level understanding of outcomes.

Eliminating the placebo arm of trials of late-stage cancer patients would immediately double the base of trials-eligible patients - something very important as on of the hardest things of trial design is recruitment. (The Nexavar NSCLC trial was based on ~700 patients recruited at ~150 locations. Imagine what a pain it is to manage numbers like this, and how much $$$ could be saved if ~350 patients were not needed.)

The two reasons that the use of placebo arms continue (other than seeking perfection in experiments) are that

1) for many drug/disease combination, the survival benefit may be tiny (4.5 months versus 3), to the point of being very difficult to detect.

2) regulators have a negative biased (against approval), mostly for CYA reasons. No one ever got fired at the FDA for being too picky with approvals, while a sure way to get fired at the FDA is to be too permissive.

Combining these two arguments raises the bar for drug approvals. To obviate this, I would make the survival standards generated by the FDA/NIH (discussed above) a simple hurdle - if a drug clears the baseline hurdle survival rate or just ties it, approval is automatic, in contrast to the current approach of requiring placebo arms in order to demonstrate the statistical significance of the trial outcomes.

(Yes, I did just argue against a fundamental aspect of data analysis, but we're talking about cancer here. We need more shots on goal, and more importantly, more learning. We don't really learn anything from a placebo arm patient when we already have millions of data points in our databases.)

(end rant)