TITLE: Likelihood Ratio Methodologies for Safety and Risk-Based Monitoring
SPEAKER: Lan Huang, Zhihao (Howard) Yao and Ram Tiwari, CDRH, FDA


With greater technological advancements and medical innovations, post-market safety surveillance plays a critical role in ensuring public health. Companies manufacturing medical products evaluate the safety issues/signals of the medical products during the product development and, also monitor the safety issues/signals after the product is on the market after approval of FDA. Consistent with the mission to protect and advance the public health, FDA also monitors the benefit-risk balance of medical products over their life cycle. A safety signal identified in the development and post-market surveillance includes information suggesting a possible safety concern that need further assessment.

In this interactive course, we will review some common Frequentist and Bayesian data mining methods for evaluation of safety signals in large post-market data including individual case reports obtained from spontaneous reporting systems (SRS), such as the FDA adverse event reporting system

(FAERS). The main focus of this course is on the Likelihood Ratio Test (LRT) method and its extensions, which detect safety signals with reasonable power and sensitivity, and have control of overall type-I error rate and false discovery rate.

The topics covered include the basic LRT method based on Poisson model for safety signal detection in large data without exposure information; Longitudinal LRT methods for active safety surveillance with exposure information; LRT methods for data from multiple studies; LRT method modified for comparing safety issues in treatment vs. control groups using clinical trial data; LRT methods extended for signals in a drug class and a modified LRT method for signals including single drug or drug combinations; ZIP-LRT method for modeling data with extra zero counts; normal-LRT method for continuous outcomes; a spatial-cluster signal detection method using LRT; and the use of LRT in site selection.

We will give demos of the LRT software/tool in OpenFDA for drug and AE signals, and BIMO LRT Inspection Statistical Software (BLISS) for site selection in medical device, both programmed using R and JavaScript, and displayed using Shiny and Shiny dashboard by RStudio (a R package that enables results to be presented in an interactive web application format). The audience will learn the basic concept of LRT methods for safety signal detection in data- mining and its extensions for different applications in drugs and devices, and the use of some R codes on examples for different scenarios.

Instructor’s Bio: 

Ram C. Tiwari, Ph.D. is the Director for Division of Biostatistics, CDRH, effective June 27, 2016. He joined FDA in April 2008 as Associate Director for Statistical Science and Policy in the Immediate Office, Office of Biostatistics, Office of Translational Sciences, CDER. Prior to joining FDA, he served as Program Director and Mathematical Statistician in the Division of Cancer Control and Population Sciences at National Cancer Institute, NIH; and as Professor and Chair, Department of Mathematics, University of North Carolina at Charlotte.

Dr. Tiwari received his MS and PhD degrees from Florida State University in Mathematical Statistics. He is a Fellow of the American Statistical Association and a past President of the International Indian Statistical Association. He has published 200+ research papers on a wide range of statistical topics. His current research interests include developing frequentist and Bayesian methods in clinical trials and pre-and-post market drug/device safety evaluation.

Dr. Lan Huang received her Ph.D. in Statistics from University of Connecticut in 2004. From 2004 to 2009, Dr. Huang worked on cancer surveillance at national cancer institute (NCI). Dr. Huang joined FDA/CDER in 2009 as a statistical reviewer and moved to FDA/CDRH in 2016. She has reviewed submissions for both therapeutic and diagnostic products/devices and has participated in regulatory research for methodologies to improve the quality of review in statistical analysis in clinical trials and safety surveillance in CDER and CDRH.




Zhihao (Howard) Yao, received his master’s degree from University of Rochester in 2009. He has over 10-year experience of data analysis and software development. Before Mr. Yao joined FDA as a mathematical statistician in 2016, he worked as a database programmer at department of Biostatistics University of Mississippi Medical Center and a data analyst at NIH. His works are focused on software development, data analysis and data visualization

This entry was posted in . Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *