Dear all,
Please find some information regarding the Causal Inference In
Epidemiology course hold by Andrea Rotnizky and Stjin Vansteelandt at
the Summer School on Modern Methods in Biostatistics and Epidemiology,
June 9-14, 2014, at CastelBrando, Cison di ValMarino, in the Province
of Treviso, Italy.
There are still some seats available. Please visit our course homepage
www.biostatepi.org for more detailed information.
Abstract:
Health policy and clinical decisions rely on the findings of clinical
and epidemiological studies of the causal effects of interventions,
treatments or exposures. In the course you will learn to critically and
systematically evaluate the pitfalls of non or imperfect experimental
studies, in particular all possible biases that can arise. You will
familiarize with state of the art causal analysis methods that help
squeeze as much evidence as these imperfect studies carry about the
causal effects of interest. These modern methods start from a clear
formulation of the effect measure that is of interest to answer the
causal question at stake and a model that encodes the temporal ordering
of variables and possible a-priori known causal relationships. The
subsequent analysis is then geared towards the specific effect measure.
The course will be centered around two complementary topics in
causal analysis, namely the potential outcome model and the causal
diagram. Using these tools we will give precise definitions of
traditional epidemiological concepts such as direct causal effect,
indirect causal effect, overadjustment, confounding, selection bias and
intermediate variables. You will understand that biases can be induced
in an analysis not only by omitting important confounders but also by
overadjusting for inappropriate covariates. You will gain insight to
when standard regression techniques apply or fail to infer cause-effect
relationships. You will learn graphical and statistical techniques to
investigate causal paths, the causal effect of time dependent exposures
and the use of instrumental variables to supply causal information when
not all important confounders are available. In the applied lab you will
apply these concepts to analyze real data sets as well as resolve
problem sets that will help you solidify these ideas.
The topics covered in the course will be:
- causal diagrams: the d-separation rule, the backdoor criterion
for the identification of causal effect, M-bias: reading from diagrams
the possibility of inducing bias by inappropriate adjustment of covariates.
- confounding adjustment of the effect of a point exposure on an
outcome: standard regression adjustment, standardisation and
G-computation, propensity score based methods that use
subclassification, matching, regression or inverse weighting,
double-robust estimators.
- disentangling direct from indirect exposure effects: controlled
direct effects, natural direct and indirect effects, the mediation formula.
- time-varying confounding adjustment of the effect of a
time-varying exposure on an outcome: marginal structural models and
inverse probability of treatment weighting
- instrumental variable methods.
The different concepts and techniques will be illustrated using
real data sets from epidemiology.
cordially,
rino
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|