6:34 PM
 | 
Mar 22, 2019
 |  BioCentury  |  Tools & Techniques

A frequentist and a Bayesian walk into a trial

How Bayesian methods are gaining ground in frequentist-dominated clinical trials

Editor's Note: This article was updated on Mar 25, 2019 at 10:49 AM PDT

In the stand-off between Bayesians and frequentists, the latter own the show for pivotal trials. But with non-profits leading the drive, Bayesians are poised to enter late-stage development via platform trials in cancer. The next areas to benefit might be pediatric and rare diseases, where the approach can maximize learnings from limited data.

Despite widespread panning of the limitations of p-values, and broad agreement that Bayesian analyses can reduce costs, shorten timelines and reduce patient exposure to treatments that don’t work, Bayesian designs have yet to take hold beyond dose-finding studies.

Some pharmas are using Bayesian trials to quickly prioritize which Phase II compounds to take to Phase III, or to mine clinical data post-hoc for new targets and biomarkers (see Sidebar: “Learn Beyond Confirm”).

But few are willing to take the leap and use the approach in registrational trials.

Part of this is because FDA leadership has given full-throttled support for Bayesian methods in designing adaptive trials, but no official endorsement for Bayesian use of prior data or predictive analyses in approval decisions (see Sidebar: “Different Likelihoods”).

Janet Woodcock, director of FDA’s Center for Drug Evaluation and Research (CDER), has repeatedly touted Bayesian adaptive designs for their potential to reduce trial duration and size.

Though FDA’s Center for Devices and Radiological Health (CDRH) has published a guidance document on using Bayesian statistics in medical device trials, including pivotal trials, the agency has not produced similar guidance for therapeutics. A CBER and CDER September draft guidance on adaptive clinical trial designs included considerations for Bayesian simulation methods, but did not directly address use of the approach for registrational trials.

“How we declare success is typically a frequentist approach, and a lot of that is driven by perceived regulatory requirements.”

Karen Price, Eli Lilly

“While we may use Bayesian methods in the design of the trial, how we indicate we’re going to declare success is typically a frequentist approach. A lot of that is driven by at least perceived regulatory requirement,” said Karen Price, a senior research advisor and Statistical Innovation Center lead at Eli Lilly & Co.

Laura Esserman, director of the Carol Franc Buck Breast Cancer Center at the University of California San Francisco, thinks it will be up to sponsors to create the change.

Esserman leads the I-SPY series of breast cancer platform trials, which pioneered adaptive Bayesian Phase II trial designs (see “I-SPY Adapts”).

She said while I-SPY is collaborating with FDA on the study design, it’s not waiting for the agency to define what it wants to see. “Instead of asking for guidance ahead of time, you want to generate the evidence that will help us to improve the way we write regulatory policy.”

I-SPY, the Global Coalition for Adaptive Research (GCAR) and the Pancreatic Cancer Action Network (PanCAN) are all lining up Phase II/III seamless trials in glioblastoma and pancreatic cancer with Bayesian predictive probabilities. GCAR’s GBM-AGILE is the first to announce a timeline, with dosing planned to begin in April.

The methods could also gain ground in pediatric and rare disease trials, where the scarcity of patients creates more pressure to maximize use of relevant data. In particular, these trials stand to benefit from the ability to include information from prior studies.


Figure: Bit by bit

Bayesian analyses calculate the probability of different hypotheses given observed data. This calculation is often represented as a probability distribution function, where...

Read the full 2871 word article

User Sign in

Trial Subscription

Get a 4-week free trial subscription to BioCentury

Article Purchase

$150 USD
More Info >