Login with your Reed credentials to view all events.

Bayesian Safety Surveillance with Adaptive Bias Correction -
In this presentation, we will discuss a collaborative project with the FDA CBER BEST Initiative to improve on post-market vaccine safety surveillance procedures through Bayesian sequential analysis. Post-market surveillance on approved vaccine products is essential for addressing safety concerns. The goal is to detect rare or high-risk adverse events that often go undetected in clinical trials due to limited sample sizes. At the FDA, safety surveillance is performed by sequential analysis of real-world observational healthcare data that accrue over time. The major challenge is that we need to control the testing error induced by sequential multiplicity, while being able to detect safety signals rapidly. Meanwhile, observational data are often systematically biased, which can substantially inflate decision error. The standard statistical approach for surveillance is Maximum Sequential Probability Ratio Test (MaxSPRT). It is designed to handle sequential multiplicity, but it requires a pre-fixed surveillance schedule and does not provide a coherent framework for adjusting for systematic bias. Collaborating with FDA CBER, we have developed a Bayesian alternative surveillance procedure that tackles these challenges in sequential analysis of observational data. Through comprehensive empirical evaluations on large-scale observational healthcare databases, we show that, compared to MaxSPRT, our Bayesian method offers more flexibility on the surveillance schedule, more transparency and interpretability in decision-making, and better error control through statistical correction of bias in observational data.

Event Details

0 people are interested in this event