Skip to Main Content

Last week STAT senior writer Matthew Herper wrote a story about a collaboration between Flatiron Health and the Food and Drug Administration, and a vexing topic called real-world evidence, which basically means the integration of types of data beyond clinical trials into the drug approval process. (You can read that story here.)

On Tuesday, Herper held a chat with STAT Plus subscribers. Here is an edited partial transcript (emphasis on partial — subscribers got to go much deeper) on the topic, particularly the area of “synthetic controls,” which Herper called “the biggest idea, the most appealing, and the scariest.” It’s the idea that computer-curated sets of health records could sometimes replace placebo groups. Read on. And if you want to participate in future chats, subscribe to STAT Plus.


Lexi G: How can real world data advance personalized healthcare/precision medicine? Any concrete examples?

We’ve seen three cases where real-world evidence was used in a drug approval — once for a first-time drug approval, and twice for an additional indication, according to Nancy Dreyer, the chief scientific officer at IQVIA, the clinical research organization.

The first-time approval is for Bavencio, the PD-L1 inhibitor developed by Pfizer and Merck KGaA to treat metastatic Merkel cell carcinoma. Data were compared to a historical control of matched patients. Blincyto, from Amgen, received an additional approval for Philadelphia chromosome-negative relapsed and refractory B-cell precursor acute lymphoblastic leukemia based on a single-intervention group trial. The results were compared with historical data from 694 comparable patients extracted from 2,000 patient records in the U.S. and E.U.


Perhaps the most interesting example is Invega Sustenna, the Johnson & Johnson long-acting antipsychotic. This was a pragmatic real-world trial, a randomized trial that used measures that are collected in clinical practice. It showed a six-month delay in relapse for those that got this long-acting injection compared to a basket of seven antipsychotic pills, and the FDA did approve adding to the product’s label.

But we’re still very much in the stage of defining what these tools are and what they mean.

Andrew L: One of the most-discussed use cases of real-world evidence (both in the Flatiron piece and elsewhere) is around use as a “synthetic” control arm to reduce or eliminate the need for placebo or comparator arms in clinical trials. What do you see as the biggest roadblocks to adoption of this use case?

This is the most discussed use case because it’s the biggest idea, the most appealing, and the scariest. There’s a lot of loaded emotion here on both sides, from patient advocates who don’t think cancer patients should ever have to receive placebo to clinical trialists who know from hard experience that only the answers from randomized clinical trials can really be trusted.

Right now we’re seeing synthetic control arms most in cases where there is no control available, and the treatment difference is large. I think we’re going to need a lot more validation before we start augmenting (what Flatiron calls a hybrid control) or replacing the control groups in randomized controlled clinical trials.

One of the biggest questions I have is how, if we move to a “synthetic” control world, we’ll know if the system breaks.

It’s also important to emphasize: Researchers and the FDA are making a distinction between purely synthetic controls and those constructed from contemporaneous data. This is confusing, because most critics of synthetic controls don’t make this distinction, and are usually referring to controls constructed from contemporaneous data.

Jason C: How do companies like FlatIron obtain electronic medical records? And what ability to patients have to either consent or deny their records be used as part of this ‘real world evidence’? Is there a bias introduced in who may or may not consent?

Flatiron owns an EHR company, which is widely used by community oncologists. Patients have the ability to opt-out of having their data or records used. It’s a very fair question to ask whether patients know this or what the right informed consent process would be.

Most patients consent. Also, there may be biases that are lost by patients not enrolling in a study and getting placebo. Is there a benefit to being in a clinical trial? If so, will a control arm constructed from EHR data be worse than what happens in a placebo group?  We need answers to ALL these questions, and we need to make sure we’ve fully considered the privacy issues as this becomes a bigger deal.

Thomas J: Is the FDA really going to use this to approve drugs? Front office all talking about it but when you ask any staffers they think it’s statistical bs.

This is the big question, isn’t it? I certainly think we’re going to see real-world evidence used for new indications before it’s used for new drug applications, and as supporting evidence before it’s accepted as a pivotal trial.

One really obvious place to use RWE in oncology is in approvals that use single-arm trials and response rates, where right now there is no comparator. Would knowing how survival, progression-free survival, overall response rate and complete response rate compared to a matched control from the same time period help? Maybe. Even there, there’s going to have to be a lot more validation, because reviewers and outside experts are going to wonder what the heck they’re looking at.

And that’s going to be the big issue, especially in cases (like oncology) where there may only be one source of data (say, Flatiron). Really, the use of real-world data is going to depend on a whole ecosystem of technologies developing that complement and support each other.

Comments are closed.