About the course
The course content is structured in six sections, each consisting of
preparation through self-study of selected literature (3.5 hours preparation
per session) and 3-hour in-person sessions that include a theoretical component
(90 minutes per session) and a bioinformatics lab (90 minutes per session). The
final exam consists of a 2-page structured essay (mini-review) about a
pre-specified causal inference topic (workload 20 hours).
The specific contents of each course section is listed below.
Session 1:
Introduction to causal thinking
Theory
Ø
Objective: Introducing basic
concepts of causality.
Ø
Topics:
·
History and philosophy of causality
·
Introduction to causal reasoning and its importance in
research
→
Preparation: Reding excerpts from Modern Epidemiology
(Rothmann, et al.)
→
Lecture (45 minutes): Clemens Wittenbecher
→
Group work (45 minutes): Moderated discussion on the
causal mechanism behind published associations making explicit use of different
causal concepts and causal language.
Lab
Ø
Objective: Describing
univariable and multi-variable distributions.
Ø
Topics: Introducing
statistical measures and basic concepts of frequentist hypothesis testing.
·
Measures of Location and Distribution
·
Measures of Correlation and Association
·
Interpretation and misconceptions concerning P-values
→
Preparation: Install R and software packages
→
Lecture (30 minutes): Jakub Morze
→
Group work (60 minutes): Generate histograms, violine
plots, heat maps, and association tables.
Session 2: Causal
concepts and statistical models
Theory
Ø
Objective: Linking causal
concepts to data analysis.
Ø
Topics: Introduce counterfactual thinking and methodologies
to formalize causal assumptions.
·
Counterfactuals and Potential Outcomes
·
Directed Acyclic Graphs (DAG)
→
Preparation: Reding excerpts from The Book of Why
(Judea Pearl), Observation & Experiment (Paul Rosenbaum)
→
Lecture (45 minutes): Clemens Wittenbecher
→
Group work (45 minutes): Using DAGs to formalize and
explain specific causal mechanisms.
Lab
Ø
Objective: Understanding that a
statistical model can encode causal assumptions.
Ø
Topics:
·
Multi-variable linear regression
·
Derive sufficient adjustment sets from DAGs estimates.
→
Preparation: Download datasets and install R packages
→
Lecture (30 minutes): Jakub Morze
→
Group work (60 minutes): Analyze cross-sectional data
and compare effect estimates across models with different adjustment sets.
Session 3: Study
design I ecological and cross-sectional studies
Theory
Ø
Objective: Understanding
cross-sectional study designs and their strengths and limitations in
etiological research.
Ø
Topics: Introduce basic cross-sectional study design and
highlight important contributions.
·
Ecological studies
·
Cross-sectional studies
·
Reverse causation and other study design-specific
sources of bias
→
Preparation: Paper I (Cross-sectional studies)
→
Lecture (45 minutes): Clemens Wittenbecher
→
Group work (45 minutes): Extracting the causal
assumptions from Paper I and evaluating the validity of causal conclusions
Lab
Ø
Objective: Learning to analyze
cross-sectional data in linear and logistic regression models.
Ø
Topics:
·
Continuous outcomes and multi-variable linear
regression models
·
Categorical outcomes and logistic regression models
→
Preparation: Download datasets and install R packages
→
Lecture (30 minutes): Jakub Morze
→
Group work (60 minutes): Analyze cross-sectional data
and visualize effect estimates and confidence intervals.
Session 4: Study
design II longitudinal and prospective studies
Theory
Ø
Objective: Understanding the
design of prospective studies and the advantages of longitudinal data
collection.
Ø
Topics: Introduce basic longitudinal and prospective studies
and highlight important contributions.
·
Prospective cohort studies
·
Nested case-control and case-cohort studies
·
Examples of repeated exposure assessment
·
Remaining sources of bias and the role of measurement
error in prospective data analysis
→
Preparation: Paper II (Prospective studies) and Paper
III (Repeated measurements)
→
Lecture (45 minutes): Clemens Wittenbecher
→
Group work (45 minutes): Extracting the causal
assumptions and evaluating sources of bias from Paper II or Paper III and
evaluating the validity of causal conclusions
Lab
Ø
Objective: Learning to analyze
matched case control studies in conditional logistic regression models and
time-to-event data in Cox PH regression models.
Ø
Topics:
·
Conditional logistic regression models
·
Cox PH regression models
→
Preparation: Download datasets and install R packages
→
Lecture (30 minutes): Jakub Morze
→
Group work (60 minutes): Analyze prospective cohort
data in conditional logistic regression (nested matched case-control studies)
and time-to-event data (Cox PH regression) visualize effect estimates and
confidence intervals.
Session 5: Causal
mediation analysis
Theory
Ø
Objective: Understanding the
design, assumptions, and interpretation of causal mediation analysis.
Ø
Topics:
·
Decomposition of total effects into direct and
indirect effects
·
Counterfactuals and do calculus
→
Preparation: Reding excerpts from Causality (Judea
Pearl) and Paper III (applied mediation analysis)
→
Lecture (45 minutes): Clemens Wittenbecher
→
Group work (45 minutes): Evaluate the applied
mediation analysis in Paper III against the assumptions for causal
interpretation of the mediation analysis results
Lab
Ø
Objective: Conducting causal
mediation analysis.
Ø
Topics:
·
Statistical approaches and R-packages for causal
mediation analysis
·
Communicating results from causal mediation analysis
→
Preparation: Download datasets and install R packages
→
Lecture (30 minutes): Jakub Morze
→
Group work (60 minutes): Implement and report a
pre-specified causal mediation analysis.
Session 6: Effect
modification analysis
Theory
Ø
Objective: Understanding the
assumptions underlying causal inferences based on population averages and
recognizing potential sources of effect heterogeneity.
Ø
Topics:
·
The concept of effect modification.
·
Interaction in biomedical research: synergies and
antagonistic effects.
→
Preparation: Paper IV and paper V (additive
interaction analysis theory and example)
→
Lecture (45 minutes): Clemens Wittenbecher
→
Group work (45 minutes): Using DAG(s) to illustrate
the causal mechanism(s) underlying known interactions and effect modifications
in biomedical research
Lab
Ø
Objective: Learning basic
statistical approaches to detect and quantify effect heterogeneity.
Ø
Topics:
·
Stratification
·
Additive vs. multiplicative interaction tests
·
Reporting of model estimates in the presence of effect
modifiers
→
Preparation: Download datasets and install R packages
→
Lecture (30 minutes): Jakub Morze
→
Group work (60 minutes): Conduct interaction tests and
communicate study results when interactions were detected.
More information
Literature
Textbook excerpts: Modern Epidemiology (Rothman et al.), The Book of Why
(Judea Pearl), Causality (Judea Pearl), Observation & Experiment (Paul
Rosenbaum)
Session-specific selection of original research articles
