Course 7
Date
23 – 25 January 2025
Faculty
Prof. Kate Tilling
Bristol Medical School, University of Bristol, United Kingdom
Prof. James Carpenter
London School of Hygiene and Tropical Medicine, University of London, United Kingdom
Medical Research Council Clinical Trials Unit, 90 High Holborn, London, United Kingdom
Venue
CH – 3823 Wengen | Hotel Jungfraublick
Course description
This course is designed for epidemiologists and applied statisticians who have to handle missing data in the analysis of observational or randomised studies.
The aim of the course is to equip you to understand the issues raised by missing data, the likely impact of missing data on a complete records analayis, and when and how to use multiple imputation to reduce bias and recover information.
We will describe the use of directed acyclic graphs (DAGs) to capture the assumptions around missing data and their likely impact on the analysis. Following this, we introduce the concepts of data being missing at random and missing not at random, relating these to the concepts of exchangeability and no unobserved confounding, and to the DAGs.
Using examples, we will introduce multiple imputation and show how it can be used to handle missing values in both causal modelling of observational data (using multivariable regression and also using propensitiy score analyses) and the analysis of randomised controlled trials with missing outcomes.
As the assumptions underlying both complete records analysis and missing at random are untestable, we will describe how sensitivity anlyses can be formulated and implemented using multiple imputation.
The final session of the course will give the opportunity for participants to present their data, and challenges with missing data, for discussion by the tutors and class.
Practicals will use Stata; the majority will also be available in R.
Recommended reading:
- Multiple Imputation and its Application (2nd Editon, 2023) published by Wiley
- Framework for the treatment and reporting of mising data in observational studies: the Treatment And Reporting of Missing data in Observational Studies framework (Lee et al, Journal of Clinical Epidemiology, 2021, 134, 79-88, doi:10.1016/j.jclinepi.2021.01.008) https://pubmed.ncbi.nlm.nih.gov/33539930/
Course objectives
By the end of this course participants will:
- Learn how to use directed acyclic graphs to assess the likely consequences of missing data for a complete records analysis
- Understand the concepts of exchangeablility, missing at random and missing not at random as they relate to causal and trials analysis
- Be able to use multiple imputation for missing data in (i) observational analyses; (ii) propensity scores for causal modelling, and (ii) missing outcome data in trials
- Understand the role of sensitivity analysis and learn how to use multiple imputation to carry out sensitivity analyses
Course audience
This course is designed for epidemiologists and applied statisticians who have to handle missing data in the analysis of observational or randomised studies.
Course outline
Thursday, 23 January 8:00 am – 12:00 pm | 4:30 pm – 6:30 pm
Friday, 24 January 8:00 am – 12:00 pm | 4:30 pm – 6:30 pm
Saturday, 25 January 8:00 am – 12:00 pm | 1:00 pm – 3:00 pm
Credits
1.0 ECTS
Course materials
Students should bring their own portable computers. A course license for Stata® will be available to install before arrival. Practicals will use Stata; the majority will also be available in R (http://www.r-project.org/).
Onsite University of Bern IT staff provides support upon e-mail () request.
Course book
On the first day of the course we provide:
Multiple Imputation and its Application (2nd Editon, 2023) published by Wiley
Course fee
PhD Bern: CHF 650
PhD other: CHF 850
Academic: CHF 1050
Industry: CHF 2050
Registration
Accomodation
Book your accommodation separately. Please see recommendations for special prices.