Adaptive Design Guidance Explained

CDRH Division of Biostatistics deputy director Gerry Gray defined adaptive trial designs in a 9/22 webinar as “clinical studies that include some prospectively planned opportunity or set of opportunities to change the study design based on data accumulated during the course of the study without undermining the integrity and validity of the study.” Gray was discussing a final guidance for adaptive designs issued by CDRH and CBER at the end of July.

Researchers should consider using an adaptive design, he said, when assumptions of trial outcomes may not go according to plan. Moderate levels of uncertainty, he said, can be addressed through adaptive designs.

The primary adaptation that Gray and his colleagues have seen is in sample size adjustment. “There’s all kinds of opportunities for adaptation as long as they’re pre-specified,” he said. Adaptive designs are usually advantageous, he said, if there is some uncertainty in the parameters used to design a trial that can be addressed by the adaptation.

While adaptive designs can be used in studies at any phase, Gray said they are most often seen in pivotal trials. “We are concerned that the estimation of treatment effect is going to be appropriate, and the confidence intervals,” he said. “Sometimes you have to worry about statistical biases that can creep in because of the nature of the adaptive design. Operational bias is always an issue…. And everyone wants a trial that has a high probability of succeeding, so power is important. There also are practical considerations related to trial infrastructure, trial conduct, and availability of alternative treatments, etc., that come into play.”

Speaking for CBER, Office of Biostatistics and Epidemiology director Estelle Russek-Cohen said the Center’s devices are mostly diagnostic, with a few that are therapeutic. So the clinical trials they see tend to be similar to trials for biologics.

Both presenters discussed the need for sponsors to communicate with the agency on adaptive designs. “The more complex the design is,” Russek-Cohen said, “it may result in more than one round of discussion with the agency. And sometimes acceptance of a particular design is not just statistical but could also be clinical.”