Dear all-stat,
This is just a reminder to all those who are interested about the forthcoming RSS meeting in London on 26 May.
Please note the amended contact details; see here
http://membership.rss.org.uk/main.asp?group=&page=1321&event=1295&month=&year=&date=
for all costs and registration details.
Best wishes,
Paul Clarke.
---
Meeting of the General Applications and Medical Sections
Recent Developments in Observational Epidemiology using Mendelian
Randomisation
Mendelian randomisation involves using genetic variants that are known to
be reliably associated with modifiable risk factors to estimate the causal
effects of these risk factors. The last ten years has seen an explosion in
the number of published Mendelian randomisation studies, most of which rely
on instrumental variable techniques from econometrics. The aim of this
meeting is to highlight recent developments from biostatistics,
epidemiology and econometrics, and the impact that these will have on
future Mendelian randomisation stuides. An extremely full day will
comprise six presentations from prominent workers in the field. The meeting
will conclude with a panel discussion in which matters raised during the
day will be discussed and questions taken from the floor.
The meeting is sponsored by the ESRC-funded project "Impact of Family
Socio-economic Status on Outcomes in Childhood & Adolescence" (large grant
RES-060-23-0011); the MRC-funded project "Inferring Epidemiological
Causality using Mendelian Randomization" (collaborative project grant
G0601625); and the MRC Centre for Causal Analysis in Translational
Epidemiology (G0600705) at the University of Bristol.
Speakers:
The speakers will be Jack Bowden and Stephen Burgess (both MRC
Biostatistics Unit, Cambridge), George Davey Smith, Vanessa Didelez, Debbie
Lawlor, and Tom Palmer (all University of Bristol); John Thompson
(University of Leicester) will take part in the panel discussion.
Date and time:
Thursday 26 May 2011, 10:30-5pm (with morning coffee, afternoon tea and
lunch)
Place:
The meeting will take place at the Royal Statistical Society headquarters,
12 Errol Street, London, EC1Y 8LX (see
http://www.rss.org.uk/site/cms/contentCategoryView.asp?category=142 for a
map and directions).
Admission:
It is necessary to register for this event and there is a charge (??35 for
non-RSS members; ??25 for RSS members; ??22 for RSS fellows; and ??20 for
RSS reduced-fellows and CStats). Please email [log in to unmask] to register.
Abstracts and presentations:
Mendelian Randomisation: Where did it come from? What is it? Why is it
useful?
Debbie A Lawlor (MRC Centre for Causal Analyses in Translational
Epidemiology, University of Bristol)
Mendelian randomisation is a method for causal inference in observational
studies in which genetic variants reliably associated with the modifiable
risk factor of interest are used to assess the causal effect of the risk
factor. One of the earliest proponents of this method was M Katan, who in a
1986 letter to the Lancet suggested that if cholesterol levels were truly
causally related to increased cancer risk then one would expect
associations of genetic variants known to affect cholesterol to be related
to cancer risk. With the advent of genome-wide sequencing, the use of
Mendelian randomisation studies for causal inference has increased
considerably in the last 10 years. The statistical approach for Mendelian
randomization studies is an instrumental variables analysis, where the
genetic variant is the instrumental variable. In this introductory talk, I
will define Mendelian randomisation, illustrate how it relates to
instrumental variables analysis and discuss why genetic variants are less
likely than other (non-genetic) variables to violate the assumptions of
instrumental variables analyses; some examples of how this approach has
been used in epidemiology will be presented.
A comparison of different IV methods for binary outcomes and what we have
learned about them
Vanessa Didelez (Department of Mathematics, University of Bristol)
Mendelian randomisation is an instrumental variable (IV) technique that has
been used in epidemiology to investigate the causal effect of a modifiable
risk factor on (typically) a disease outcome by exploiting a genetic
predisposition to the risk factor. Much of the literature on IV methods has
focussed on linear models, where an IV analysis is particularly elegant and
simple. However, in epidemiology outcomes are often binary and hence linear
models are only of limited use. Unfortunately the binary/non-linear case
turns out to be less elegant and simple. A variety of IV methods for binary
outcomes have been proposed, and they differ in two important points: (1)
the exact interpretation of the target causal parameter; (2) the type and
restrictiveness of the distributional assumptions made. This presentation
will give an overview over the various approaches, and discuss two of them
in more detail: structural mean models and the "Wald-type" estimator.
Mendelian randomization analysis of case-control data using structural mean
models
Jack Bowden (MRC Biostatistics Unit, Cambridge)
'Instrumental Variable' (IV) methods provide a basis for estimating an
exposure's causal effect on the risk of disease. In Mendelian randomization
studies genetic information plays the role of the IV and analyses are
routinely performed using case-control data, rather than prospectively
collected observational data. Although it is well-appreciated that
ascertainment bias may invalidate such analyses, ad hoc assumptions and
approximations are made to justify their use. This presentation attempts to
explain and clarify why the standard approach may fail.
Using the Structural Mean Model (SMM) framework, consistent estimators of
the causal relative risk and odds ratio are proposed if a priori knowledge
is available regarding either: (a) the population disease prevalence; or
(b) the population distribution of the IV (e.g. population allele
frequencies). Procedures for efficiently incorporating multiple IV's into a
single SMM are also described. These methods are demonstrated on
case-control data from the recently completed EPIC study, in an attempt to
assess the evidence for a causal relationship between C-reactive protein
levels and the risk of Coronary Artery Disease.
Bayesian methods for meta-analysis of Mendelian Randomisation studies
Stephen Burgess (MRC Biostatistics Unit, Cambridge)
Precise estimation of causal association using Mendelian randomization
requires large amounts of data, typically necessitating synthesis of
evidence from heterogeneous sources and meta-analysis. We introduce a
Bayesian framework for instrumental variable analysis which combines a
two-stage analysis of a genotype-risk factor-outcome model and a
hierarchical meta-analysis. This framework is used to analyse data on the
causal association of C-reactive protein on coronary heart disease,
comprising individual participant data on over 150 000 participants from
over 40 different studies measuring with 20 different genetic variants. We
address issues of use of multiple genetic variants, variation of genetic
variants measured between the studies, non-measurement of the risk factor
in certain studies, and between-study heterogeneity of genetic effects and
of the causal effect.
Estimation using structural mean models with multiple instruments
Tom Palmer (MRC Centre for Causal Analyses in Translational Epidemiology,
University of Bristol)
In this talk we describe how to estimate structural mean models (SMMs), as
proposed by Robins, using multiple instrumental variables (IVs) in the
generalized method of moments (GMM) framework common in econometrics. The
GMM approach is flexible as it allows estimation of over-identified models
in which there are more instruments than endogenous variables. It also
allows assessment of the joint validity of the multiple instruments through
Hansen's over-identification test. In the case of the logistic SMM the GMM
approach can incorporate different first stage association models. We
discuss the relationship between the additive SMM and the linear IV moment
condition. We discuss the relationship between the multiplicative SMM and
the multiplicative GMM estimator. For the multiplicative SMM we show,
analogously to Imbens and Angrist (1994) for the linear case, that the
estimate is a weighted average of local estimates using the instruments
separately.
We describe implementation of the additive, multiplicative, and logistic
SMMs in Stata and R. To demonstrate the models we use a Mendelian
randomization example, in which genotypes found to be robustly associated
with risk factors from genomewide-association studies are used as
instrumental variables, investigating the effect of being overweight on the
risk of hypertension in the Copenhagen General Population Study.
Mendelian Randomisation: The Future
George Davey Smith (MRC Centre for Causal Analyses in Translational
Epidemiology, University of Bristol)
The availability of multiple genetic variants related to a modifiable risk
factor allows the use of multiple independent combinations of variants to
generate instrumental variable (IV) estimates, and through examination of
heterogeneity between these to assess the potential role of reintroduced
confounding (primarily through pleiotropy). This can considerably
strengthen the causal inferences that can be drawn from Mendelian
randomization studies. Complex networks of associated biological measures
have been identified in aetiological studies of common complex diseases,
and node-by-node investigation of these through Mendelian randomization
methods can help separate causal from epiphenomenal associations. Such
networks include gene expression and epigenetic (in particular methylation)
data, and analogous approached to Mendelian randomization, which have been
referred to as "genetical genomics" and "genetical epigenomics", can be
applied to these. The presentation will cover the above potential
developments of Mendelian randomization studies.
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|