SC.05

SC.05 Causal questions and principled answers: a guide through the landscape for practicing stat.

Sunday, 5 July 2020  |  9:00 - 17:00 | Location: TBD

Instructors:
  • Saskia le Cessie
  • Els Goetghebeur


Summary:
Taught by: The Causal Inference Topic Group of the STRATOS (STRengthening Analytical Thinking for Observational Studies) Initiative, including Els Goetghebeur (Ghent University, Belgium), Saskia le Cessie (Leiden University, Netherlands)

This course aims to support the practicing statistician in making it work: causal inference form observational data in a potential outcomes framework. How to choose among the many versions of exposure, the target population and indeed the estimand in a given setting? Different estimands serve different purposes while lending themselves (in)directly to natural constraints imposed on given datasets. We spend ample time exploring this in various contexts before explaining key features of estimation methods relying on either the `no unmeasured confounders’ assumption or the availability of `instrumental variables’. We review, apply and compare dedicated methods centered around outcome regression and stratification, inverse probability weighting and their double robust version, when relying on the no unmeasured confounders assumption. We discuss the value of two-stage-least-squares estimation when exploiting an instrumental variable and elaborate on pitfalls. We illustrate key results and estimation properties with published case studies, using either the original data or simulations. We elucidate links and differences between the methods. Hands on sessions will guide participants in using R, SAS or Stata with the data.


Prerequisites:

This is course is aiming at applied statisticians and epidemiologists, who have no sound understanding yet of the concepts and application of modern causal inference methods. The course will assume familiarity with regression methods.


Outline:

The course will consist of five sessions:

Morning - Session 1: Causal estimands

This session discusses the formulation of a causal question with a well-defined population of interest, an outcome that corresponds to the scientific question under study and a treatment/exposure with relevant levels. We discuss the concept of potential outcomes and show how different estimands can be defined for different purposes. Several case studies are provided.  We end the session by a group exercise around recent publications.

Session 2: Estimation and inference under no unmeasured confounding: outcome regression, matching and stratification.

Here we discuss how the estimand of interest can be estimated by the data at hand. We discuss general assumptions (positivity, no interference, no unmeasured confounding) needed to estimate the causal contrast of interest.  Methods discussed in this session are outcome regression, matching and stratification and the use of propensity methods. These methods are applied to data in the computer exercises with SAS, R or Stata.

Afternoon - Session 3: Inverse probability weighting and doubly robust estimation

In this session we discuss the principles of inverse probability weighting and double robust estimation. We will discuss the theory. Again, methods are applied to data in the computer exercises with SAS, R or Stata.

Session 4: Instrumental variables

So far, we discussed methods which assumed that all confounders were known. Instrumental variable analysis methods can deal with unmeasured confounding but require a different set of assumptions. The basics of instrumental variable methods are discussed in this session and are practiced in computer exercises.

Session 5: Summary

Here we compare the different methods, discuss the merits and pitfalls, and compare the results of the different methods in simulated data augmented with a series of potential outcomes.


Learning Outcomes:

After this course, students:

  • Are able to formulate a causal estimand which corresponds to a research question of interest
  • Know the general concepts of causal inference
  • Are familiar with the most commonly used methods to estimate causal contrasts for point exposures
  • Are able to apply these methods to real data using R, Stata or SAS
  • Are able to decide which method is most appropriate in certain situations

About the Instructors:

We have given the same course before (in English), at the ISCB 2017 in Vigo and at the Annual meeting of the Statistical Society of Canada in Montreal.

The course instructors are part of the STRengthening Analytical Thinking for Observational Studies (STRATOS) initiative, which is a large collaboration of experts in many different areas of bio statistical research. The Causal Inference Topic Group of STRATOS is composed of:

Professor Els Goetghebeur (chair), Department of Applied Mathematics, Computer Science and Statistics, Ghent University, Belgium. Els Goetghebeur held previous appointments at the London School of Hygiene and Tropical Medicine (UK),  Maastricht University (NL) and the Harvard School of Public Health (US) following a PhD obtained from the Limburgs Universitair Centrum (B). Her research focus on problems in survival analysis shifted to causal inference when confronted with noncompliance in clinical trials. More recently she had the pleasure of working with Belgian and Swedish registers on quality of care. She is generally interested in working with electronic health records. She is director of the Institute of Permanent Education at Ghent University and is Editor-in-Chief of Statistics in Medicine.

Associate Professor, Ingeborg Waernbaum (co-chair), Department of Statistics, Umeå University, Sweden. Ingeborg Waernbaum obtained her PhD in Statistics in 2008 from Umeå University. Since 2011 she is also Associate Professor at the Institute for the Evaluation of Labour Market and Education Policy, IFAU, Uppsala, Sweden, where she works with the development of statistical methods for analysing registry data. Her main research interests are in causal inference with register data with a broad focus on topics related to confounder control such as covariate selection, model specification and robustness.

Professor Bianca de Stavola, GOS Institute of Child Health, University College London, UK. Bianca De Stavola recently joined UCL GOS ICH after 23 years at the London School of Hygiene and Tropical Medicine where she was Professor of Biostatistics in the Department of Medical Statistics and co-Director of the Centre for Statistical Methodology. Bianca received her PhD from Imperial College London and MSc from the London School of Economics and Political Sciences, after graduating in Statistical and Economic Sciences at Padua University (Italy). Bianca's main research activities involve the understanding, development and implementation of statistical methods for long-term longitudinal studies, with specific applications to life-course epidemiology. As these often involve causal enquiries, in particular related to understanding pathways towards disease development, mediation analysis is her main interest.

Associate Professor Erica Moodie, Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Canada. Erica Moodie obtained her PhD in Biostatistics in 2006 from the University of Washington, before joining the Department of Epidemiology, Biostatistics, and Occupational Health at McGill University where she is now a William Dawson Scholar and an Associate Professor of Biostatistics. Her main research interests are in causal inference and longitudinal data with a focus on adaptive treatment strategies. She is an Elected Member of the International Statistical Institute, an Associate Editor of Biometrics and the Journal of the American Statistical Association. She holds a Chercheur-Boursier senior career award from the Fonds de recherche du Quebec-Sante.

Professor Saskia le Cessie, Department of Biomedical Data Sciences, Leiden University Medical Center, the Netherlands. Saskia le Cessie is a statistician with a joint appointment at the Department of Clinical Epidemiology and the Department of Medical Statistics. She obtained a master in mathematics (with minor in informatics) at the University of Utrecht and obtained her PhD in Medical Statistics at the University of Leiden. She is a broadly oriented statistician and involved in several large epidemiological studies of the Department of Clinical Epidemiology. Her research interests are in epidemiological and statistical methods for observational studies, in particular instrumental variable analysis, mediation analysis, competing risks and causal modelling.  She is heavily involved in graduate-level teaching for statistics, medical, and epidemiology students, is a member of the steering committee of the Statistical Science master program of Leiden University and is involved in the international programs in Sorbonne University Paris and the European Society of Clinical Microbiology and Infectious Diseases.


Recommendations for the Course:

Course notes, practicals and code for analysis will appear on the website ofcause.org

The course requires all participants bring their own laptops. If they wish to work with R, participants will need to install some packages beforehand. The instructors will inform the participants in a timely manner.