Page 230 - Special Topic Session (STS) - Volume 4

STS582 Mariza de A.

Mendelian randomization, causal relationship,

and statistical approaches
Mariza de Andrade, PhD
Mayo Clinic

Abstract
In Mendelian Randomization, one needs the outcome (Y), the instrumental
variable (IV) and the mediator variable (M). In the field of biostatistics and
epidemiology the genetic markers are the IVs that can vary from one to
multiple genetic markers (G). Several approaches are used in MR: Structural
Mean Models (SMMs) are semi-parametric models that use IVs to identify
causal parameters that include multiple instrument variables for multiplicative
and logistic SMMs. In this case one can use the generalized method of
moments (GMM) estimator. Other approach when one have multiple
mediators are to apply the two-stage least squares (2SLS) that consists of two
regression stages: the first-stage regression of the exposure on the genetic
IVs, i.e., (G –M) regression, the exposure is regressed on the IVs to give fitted
values for the exposure (X |G) and the second-stage (X – Y) regression, the
outcome is regressed on the fitted values for the exposure from the first stage
regression, the outcome can be continuous and binary. One can also use
likelihood-based, Bayesian, and semi-parametric methods (that includes the
GMM and SMMs). We will use available data from the VTE meta-analysis as
well as from the lung cancer.

Keywords
Instrumental Variables; Mediators; Two-stage Methods: Wald method, Missing
data

1. Introduction
In Epidemiology studies, the researchers rely on observational data that
can lead to confounding and reverse causality. In this paper will introduce
concepts of making inferences in causal effects based on observational data
using genetic instrumental variables as known as Mendelian randomization
(MR) (1). Since children inherit their genomes from their parents at random,
which means that reverse causality can be ruled out and genetic variants are
not related to the environmental confounder (2). Multivariate MR (MVMR) is
an approach that can be used to estimate the effect of two oe more exposures
on an outcome (3). The concepts of Mendelian randomization will be
introduced that include outcomes, mediators, instrumental variables among
others. One of the early example of Mendelian randomization was the levels
of C-reactive protein (CRP) and the risk of coronary heart disease (CHD).

219 | I S I W S C 2 0 1 9