What is the formula for a great event? What are the critical ingredients?
Well, first, you need some good speakers. They must have experience and knowledge that others can learn from and they must be willing to share. Second, you need to focus on a topic that is of professional interest to others in the field. It is one thing to talk about the latest and greatest, but it is more important to provide practical perspectives that are valued and can be used to improve solutions. Third, a good moderator is always a plus, someone who might not be in the thick of that specific research but clearly understands what the experts have to share, also knows the audience, and can work the technology to field questions that bubble up.
IF you have all of these things lined up, ... then what happens is exactly what happened at this latest NISS/Merck Meet-up! The hour and a half went by in a flash! And, there were over 500 attendees who logged in to listen to Frank Harrell, Professor of Biostatistics at Vanderbilt University's School of Medicine, Amy Xia, Vice President of the Biostatistics, Design & Innovation division at Amgen and Telba Irony, Deputy Directory of Biostatistics and Epidemiology at the FDA’s Center for Biologics Evaluation and Research. The focus of the session was on the use of Bayesian statistics in the drug development process. And, as evidenced by the number and variety of questions that were asked by the audience their remarks were thought provoking!
Frank Harrell provided an overview of the advantages of Bayesian approaches in the drug development process. He began by comparing both frequentist approaches and Bayesian and in doing so asked the following question as part of the understanding the challenges involved in drug development and whether results become actionable. “Would you rather know the chance of making an assertion of efficacy when the treatment has no effect, or the chance the treatment is effective?” Throughout his in-depth review of both approaches he demonstrated how Bayesian methods aligned with the research questions asked within the drug development process but also pointed out that, “You can't compute a current probability without having a starting probability.” In response, Harrell provided an example that demonstrated how possibilities exist for fully continuous trials with unlimited looks. In summary, while the computations might be a bit more complicated, the interpretation is much clearer.
“Approaches for pediatric drug development need to be efficient and flexible while maintaining valid and persuasive evidentiary standards.” This was one of the final comments of the second speaker of the session, Amy Xia. Her remarks involved a fascinating, in-depth example of her work extending information from studies in adult populations to pediatric populations for the purpose of minimizing exposure of children to clinical trials and increase the efficiency of pediatric drug development programs. Specifically, Xia’s presentation provided a detailed walk-through of a study of cinacalcet, a treatment of secondary hyperparathyroidism (HPT) in adult patients with chronic kidney disease. Her example highlighted the use of a 3-level Bayesian hierarchical model which allowed for understanding how data from one population could be used to inform implications for other population subgroups. In summary, Xia remarked that, “While the use of statistical extrapolation to support pediatric trials is an emerging tool, a Bayesian extrapolation approach helps with sample size limitations and missing control arms in pediatric settings.”
The final speaker of the session was Telba Irony. She provided an overview of Bayesian approaches that have found use within regulatory settings and the value they bring to the drug development process. She expertly reviewed the role of the prior distribution, Bayesian adaptive designs, simulations and predictive probabilities - important points to be considered in each. She then quickly moved to the lessons that were learned for what works in which clinical settings and where opportunities exist for further work using these Bayesian approaches. In conclusion Irony listed six areas that show great promise including implementing the likelihood principle in flexible clinical trial designs, using decision analysis to develop rational / transparent decision rules, and rationally determining the required strength of evidence by medical need, patient tolerance for risk and perspective on benefit, and/or severity and chronicity of the disease.
Throughout the session attendees posted questions for the speakers to address. Not only did the questions quickly get to the heart of the issues being discussed, the responses and back and forth between the speakers demonstrated the excitement and potential that Bayesian approaches bring to drug development, clinical trials and other epidemiological research areas. Review the session recording below to see what points the speakers were intent on making!
As with all of the NISS/Merck Meet-Up events, Dan Holder (Merck) served as general organizer and was moderator of the session.
Recording of the Session
Slides used by the Speakers
Frank Harrell, (Vanderbilt University)
Amy Xia, (Amgen)
Telba Irony, (FDA)