Q: Your April 5, 2017 blog post that proposed a new “Conditional” category for FDA drug approval elicited a number of positive and negative responses. Please explain the proposal in more detail to enable better reader understanding.
A: ￼In Response to “Conditional Approval: Right Solution for the Wrong Problem” by
We appreciate Ms. Brownlee’s comments on our recent blog post, but think she missed our key points—perhaps we weren’t clear enough:
- Conditional approval isn’t intended for cancers that have other options—like ￼some types of breast cancer—but rather for the invariably fatal cancers like ￼glioblastoma multiforme (GBM), for which there aren’t any good FDA-approved options.
- Conditional approval requires not just evidence of safety, but evidence that the drug has promise, at least in some patients.
- Using the registry is just a different way of collecting the data that traditionally is collected in clinical trials. Instead of a traditional trial in which only highly selected patients participate and relatively few doctors collect the data, we will be collecting data on the entire range of patients from many doctors.
Having many doctors and patients participate removes a lot of the bias inherent in clinical trials. For example, consider a doctor using an experimental treatment on “thousands” of brain tumor patients in 100 clinical trials in which only a handful of doctors are collecting all of the data. All of the doctors involved have a major financial interest in the outcome. We still have no idea if this treatment helps or hurts—and it has been going on for 40 years. With our system, we would know in a year if it helps or hurts.
￼Moreover, Ms. Brownlee’s arguments extolling the virtues of randomized clinical trials ￼(RCTs) may reflect an unfamiliarity with the cost-effectiveness of “registry trials,” especially when coupled with modern Bayesian designs.
With the traditional approach, we have been lucky to get 3 or 4 treatments approved in the last 40 years. With our approach, we can expect at least 3–6 new treatments conditionally approved each year, as the cost to get approval would be on the order of 1–2% of the cost using the current traditional approach.
- As Ms. Brownlee states, “It took more than a decade to accumulate enough patients” in an RCT to demonstrate the ineffectiveness of autologous bone marrow transplant (ABMT) in breast cancer. Further, “A few patients may actually have been helped by the Avastin, but there is no way to predict ahead of time whether patients would be helped or harmed.” These points demonstrate the limitations of classical RCTs versus Bayesian point-of-care trials that can be run on top of a registry.
- The big challenge and opportunity regarding investigational treatments that show ￼promise in some patients is to identify and continuously refine the cohort for whom an intervention is effective, as efficiently as possible. This requires a Bayesian approach, which can rapidly replicate successes and discard failures, not accruing a large trial to ￼test a specific hypothesis, which is likely to be wrong.
- A registry that captures biomarkers, treatments, and outcomes from patients ￼undergoing a variety of interventions can provide rigorous cross-controls, which are ￼every bit as valid as those provided by randomization.
Using Ms. Brownlee’s example of ABMT, in our system, the registry would have quickly picked up that it was not as good as standard chemotherapy. The current system allowed 41,000 women to use a treatment whose effectiveness was unknown because nobody was tracking it. An RCT took 10 years to find an answer that our registry may have had in as little as 1 year, saving many lives.
Ms. Brownlee states that the registry can’t show efficacy. We disagree. If you have the majority of patients being tracked in the registry, you can use all of the patients who are NOT taking the treatment as the control group. Comparing to the old historic control is useless for brain cancer—almost everyone died and we did not collect the necessary biomic data to correctly match them with current patients.
The comparison that needs to be made is which of many possible new treatments and combinations works the best. Some patients would probably stick with the old standard of care—we would be encouraging them to participate as well—so we can see if the current standard is better than any of the new treatments.
We are talking about using conditional approval only in cases where there really is no acceptable standard of care. And we ARE looking for new drugs and combinations that have extraordinary power to improve outcomes, not looking for something that extends life by weeks. We agree it wouldn’t make sense in a disease where the standard treatments offer hope.
In summary, our proposal is not to minimize the research, it is to maximize the amount of research done to a drug, just in a different time period—after phase 1 instead of before the end of phase 3.
As to picking up idiosyncratic reactions, by definition, these are rare and do require a large group of patients being tracked to identify how frequently they occur. With our registry, it is simple to analyze the data and get an early warning. Having the genomic data would allow us to try to figure out which patients are most likely to have such a problem. With traditional trials, which usually only allow a select population to be tested, a reaction that only occurs in the elderly or in minorities—which are underrepresented— may never be found. Once a drug is approved and underrepresented groups use it, side effects are not tracked as closely as they would be in our registry.
For more reference, please read “Rapid Learning for Precision Oncology” published in Nature on Jan 21, 2014. We include a free PDF download. Please see below for a 3-minute video showing how the Bayesian approach can be applied to medical research in a new method called Global Cumulative Treatment Analysis (GCTA).