Bayesian modeling in COVID-19-disease modeling, therapeutics, and vaccines

ABSTRACT

COVID-19 efforts have dominated the headlines in 2020. These efforts have involved efforts across the medical and statistical spectrum, from modeling of the pandemic to the development of therapeutics to the testing of possible vaccines. Novel methodologies have been utilized, such as platform trials, Bayesian modeling of pandemic uncertainty, and Bayesian adaptive trials to facilitate timely vaccine delivery. In this session, we will present four real examples of Bayesian methods across this range of activities. These include the official modeling of the epidemic within Los Angeles County by the leader of the team, both design and execution of platforms trials within the COVID-19 pandemic, and the Bayesian Pfizer vaccine trial. All speakers confirmed. Roger Lewis is the leader of the COVID-19 epidemic modeling team for Los Angeles County, California, advising government officials on the progress of the epidemic and projecting future developments. He will discuss the Bayesian SEIR modeling performed for Los Angeles, including capturing uncertainty in the predictions and real-world issues in data collection and adjusting modeling in the presence of evolving medical care and government policies. Ben Saville will discuss therapeutic adaptive platform trials like PRINCIPLE and REMAP-CAP (focus on PRINCIPLE). Both trials are ongoing adaptive platform trials investigating multiple therapies for COVID- 19. PRINCIPLE is a UK national priority trial and is focused on ambulatory participants with suspected COVID-19 and a higher risk of morbidity (e.g. .50 years age with comorbidities). The trial is open-label and has co-primary endpoints of subject-reported time to recovery and hospitalization. REMAP-CAP includes both open-label and blinded interventions focused on hospitalized patients in the intensive care unit across eight countries. The primary endpoint is the number of organ support-free days, and includes multiple interventions within multiple therapeutic domains, for example, antiviral agents, corticosteroids, or immunoglobulin. Both trials use innovative Bayesian modeling that account for potential drift over time, with frequent interim analyses allowing early decisions of futility or superiority. Response-adaptive randomization is used to increase allocation to interventions with better observed outcomes, which can increase statistical power for finding effective therapies and result in better participant outcomes. Mark Fitzgerald will present some of the challenges of execution for the statistical analysis committee for a trial that is rapidly adapting to an ongoing pandemic, with a focus on REMAP-CAP. REMAP-CAP is an adaptive platform trial that explores the efficacy of interventions across a range of treatment domains, including the combinations across domains, that utilizes a novel endpoint organ-support free days. The international effort combines data from five continents, evaluates thousands of treatment combinations, and rapidly evolves to accommodate information from external sources. The statistical analysis committee faces unique challenges in adjusting to rapid changes when combining data from disparate sources, updating models and reports to incorporate new design features, and producing results for public disclosure for closed domains or interventions, all while ensuring proper communication and maintaining trial integrity. Satrajit Roychoudhury will discuss the design of the Pfizer Bayesian adaptive vaccine trial. This trial incorporates multiple interim analyses, each based on achieving a sufficiently high Bayesian posterior probability of vaccine efficacy. The trial also incorporates early stopping for futility based on Bayesian predictive probabilities. In November 2020, the trial is currently ongoing. Additional information may be publicly available at the time of SCT 2021 that may be discussed, but this will depend on future events at time of submission.