Presentation: Multiplicative Effect Modeling for Meta-analysis
Candidate: Jiaqi Yin, Graduate Student, UW Biostatistics
Committee Members: Thomas Richardson (co-chair), Linbo Wang (co-chair), Jon Wakefield, Patrick Heagerty, Yanqin Fan (GSR)
Abstract: A case-control study is a retrospective observational study to help determine if an exposure is associated with an outcome/disease. This study is often conducted for rare diseases. A classic cohort study on a rare disease would require a long time to follow outcomes and secondly, to observe enough cases to be statistically significant, one would need to gather a large population, which is particularly challenging for rare diseases. However, case-control studies cannot determine a relative risk, because the prevalence of the disease is set by the study design. Even though an odds ratio which is identified from case-control data approximates relative risk in rare diseases, it is unclear when the approximation is appropriate. It is thus natural to ask: can we combine cohort and case-control studies to have a better estimation of the relative risk?
In this work, we focus on estimating relative risks in a meta-analysis including cohort and case-control studies. This project is an extension of our previous work, which has developed general methods for modeling the multiplicative effect of a categorical or continuous treatment on a binary outcome. Currently, to our knowledge, there are no methods for modeling relative risk directly from a combination of case-control studies and cohort studies. Studies typically approximate the relative risk by adjusting the odds ratio under an assumption regardng the prevalence of the disease or summarize the relative risk and the odds ratio separately from cohort and case-control studies respectively.
Monte Carlo simulations demonstrate the advantages of our proposed approach. A thought experiment also exemplifies our methods on real data.