UU直播

Marginal Meta-Analysis for Combining Multiple Randomized Clinical Trials with Rare Events 鈥� Lessons Learned from Avandia Story听

Yi Huang, PhD, Associate Professor 鈥� Dept. of Mathematics and Statistics, University of Maryland, Baltimore

ALL ARE WELCOME

Tuesday, 16 May 2017; 12:15-1:15pm - Purvis Hall, 1020 Pine Ave. West, Room 25

Abstract: Meta analysis (MA) is commonly used in the post-marketing safety studies for FDA regulated medical products, including drugs, medical device, and etc.听 Avandia Studies (Nissen et al, 2007, 2010) is a powerful example to show how important MA is in real life for quantifying the safety concerns with policy impacts.听 However, the fact that the re-analysis of same Avandia data could reach different conclusions showed clearly the statistical challenges and difficulties associated with standard fixed effect and random effect MA methods. Specifically, the inclusion and exclusion of zero trials, changing the effect estimand to risk difference, and/or using other fixed effect MA methods rather than Peto, would all lead to different results.听 Lesson learned from Avandia studies inspired our discovery of the problems associated with 鈥渉omogeneous effects鈥� or 鈥渆ffect at random鈥� assumption 鈥� the validity assumption underlying standard MA approaches, and led to a set of more relaxed Study at Random assumptions. Additionally, two more concerns motivated our research on this marginal meta analysis: (1), rare events in safety studies often lead to low power in homogeneity test associated with standard MA approaches. Even though they may bias the results, various types of add-hoc continuation corrections were proposed and widely used to improve the performance of standard MA estimators. (2) Non-collapsibility issues associated with odds ratio limit the interpretability of many popular MA estimators too.听 As a result, based on the new flexible study homogeneity assumption, we proposed a marginal meta analysis approach with natural weights which provided a consistent treatment effect estimate for marginal causal effects combining randomized clinical trials in safety studies.听 This estimator is particularly useful when the outcome is rare, and double zero trials are naturally accounted in the estimation without any add-doc continuity correction.听 Systematic simulation studies show that the proposed estimator performs reasonably well under different rationales.听 This method is re-applied in Avandia safety evaluation as a real case application.听 This is a joint work with my students, Elande Baro, Yun-Yu Cheng, and colleague from FDA, Guoxing Soon.听

Yi Huang1,听 Elande Baro2, Yun-Yu Cheng1,听 Guoxing Soon2

1: Dept. of Mathematics and Statistics, University of Maryland, Baltimore County

2: Office of Biostatistics, OTS, CDER, US FDA

Keywords:听 Meta Analysis, Rare Events, Homogeneity Assumptions, Effect at Random, Avandia, Zero trials.

听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听听

Bio: Dr. Huang is Associate Professor in the Department of Mathematics and Statistics at University of Maryland, Baltimore County, and an affiliated faculty in the joint Doctoral Program of Gerontology, School of Medicine, University of Maryland.听 She completed the Biostatistics Ph.D. training at Johns Hopkins Bloomberg School of Public Health in 2007. As a biostatistician, her research focus on propensity score related causal inference methodology, comparative effectiveness research methods (e.g. meta-analysis), and public health oriented collaborations. Current projects include post-marketing safety studies, gerontology projects, health policy study for Maryland state, and enrichment design for efficacy trials.

听