In this episode of "In the Interim…", Dr. Scott Berry chats with Frank Harrell, a professor of Biostatistics at Vanderbilt University and W.J. Dixon Award winner. Harrell describes his transition from frequentist to Bayesian clinical trial design, prompted by a decisive meeting with Dr. Don Berry, informed by David Spiegelhalter’s published work. The dialogue addresses persistent academic opposition to Bayesian methods, operational constraints in trial implementation, regulatory work at FDA, and technical Bayesian modeling details.
Key Highlights
- Harrell credits Don Berry’s direct influence for converting him to Bayesian methods during his early career at Duke, despite entrenched academic resistance.
- Discusses early cardiovascular research at Duke, experiences with large multicenter trials, and later founding Vanderbilt’s Biostatistics department.
- Details the compromise of using Bayesian interim monitoring and frequentist primary analyses under NIH and regulatory mandates.
- Outlines design and publication of the ORBITA cardiovascular trial (Imperial College London), using all-Bayesian longitudinal ordinal methodology—Lancet reviewers required all analyses remain Bayesian, rejecting inclusion of a mixing frequentist and Bayesian analyses.
- Critiques simulation of Type 1 error within Bayesian trial designs.
- Addresses deficiencies in eliciting utilities for clinical endpoints, underscoring operational challenges in longitudinal ordinal modeling and ethical imperatives for efficient early stopping.
