Cookies are used for the functionality of our web site. For more information
please see our Privacy policy or Ok thanks

Pre-conference: Demystifying causal inference and mediation methods for public health researchers


Wednesday 20 November, 09:00 - 17:00, Room Riou

Organised by:

EUPHA Research Pillar and EUPHA Section on Public health epidemiology

Background

Mediation analysis is a common tool in public health researchers’ toolboxes, helping us to understand processes or casual mechanisms through which one factor affects another and to identify pathways where intervention might be most fruitful (particularly useful when the exposure of interest is less amenable to change). The so-called ‘Baron and Kenny’ approach to mediation is very widely used in public health research but makes a number of assumptions that can rarely be met. This can produce biased findings, potentially leading to flawed conclusions. Causal mediation methods can change or avoid the need for some of these assumptions, although they remain relatively inaccessible to those most likely to want to apply them.

Aims

In this workshop we will provide public health practitioners and researchers with an overview of the limitations of the ‘Baron and Kenny’ approach to mediation, including when and why this method fails to work. We will present alternative approaches in a way that is more accessible than in the methodological literature, drawing on applied research questions to show when different methods might be most appropriately used. During the session participants will learn the differences between direct and indirect effects, and between controlled and natural direct effects. They will work in small groups to draw Directed Acyclic Graphs (DAGs) to consolidate the learning. There will be a series of practical examples of the application of mediation analysis in public health using a variety of methods. Workshop participants should come with research questions in their mind relating to issues of mediation in their own research areas. Throughout the day there will be opportunities to ask questions, and the final session is designed to resolve any remaining issues that have arisen during the day.

Track record

Presenters John Lynch (Australia) and Anna Pearce (United Kingdom), both public health researchers, have a track record in applying these methods in their own work and teaching them to others. Chairs, Alastair Leyland and Linsay Gray, are statisticians working in public health with extensive experience in applying and teaching advanced statistical methods to mixed levels and disciplines. The course material has been successfully presented to a range of audiences, in one-off workshops to public health researchers and social scientists at the University of Glasgow and to epidemiologists and public health doctors at the UCL Institute of Child Health. It has also been run numerous times as part of a successful causal inference course in Australia.

Programme

Morning chair: Alastair Leyland, EUPHA Research Pillar, United Kingdom

9.00 – 9.05

Introduction to the workshop
Alastair Leyland, United Kingdom

 

9.05 – 9.45

Introduction to mediation analysis and DAGs
Anna Pearce, United Kingdom

 

9.45 – 10.30

Small group exercise: drawing a DAG 
Anna Pearce, United Kingdom

 

10.30 - 10.45

Feedback on small group exercise
Anna Pearce, United Kingdom

 

10.45 - 11:15

Coffee Break

 

11.15 - 12.00

Limitations of traditional mediation analysis: a potential outcomes view
John Lynch, Australia

 

12.00 - 12.45

Potential outcomes approaches to mediation analysis
John Lynch, Australia

 

12.45 – 13.45

Networking lunch 

 

Afternoon chair: Linsay Gray, EUPHA Section on Public Health Epidemiology, United Kingdom 

13.45 – 14.00

Recap of morning session, including questions
Anna Pearce, Uinted Kingdom

 

14.00 – 15.00

Applications of mediation analysis (3 x 20 min time slots)

1. Are certain racial groups really at greater risk of type 2 diabetes? A mediation analysis of income inequality and diabetes incidence in the United States
Katherine Ford, Luxembourg

2. An application of mediation analysis to examine the impacts of Indigenous culture on reduced drug use problems in Canada
Cheryl Currie, Canada

3. Mediation Analyses of Educational Differences in Walking Speed Decline in Older Adults over an 11-year Follow-up
Saila Kyrönlahti, Finland 

 

15.00 – 15.30

Coffee Break

 

15.30 – 16.30

Applications of mediation analysis (2 x 20 min time slots)

1. Mechanisms of life-course socioeconomic inequalities in adult systemic inflammation: findings from two cohort studies
Cristian Carmeli, Switzerland

2. Using structural nested mean models to estimate if social inequalities in smoking initiation could be reduced by intervening on mental health problems among children
Joost Oude Groeniger, The Netherlands

3. Education and type 2 diabetes: quantifying the impact of differential exposure and susceptibility to overweight and obesity
Jimmi Mathisen, Denmark

 

16.30 – 17.00

Closing Q&A and wrapping up
John Lynch, Australia

 

Rationale

Mediation analysis is a common tool in public health researchers’ toolboxes, helping us to understand processes or casual mechanisms through which one factor affects another and to identify pathways where intervention might be most fruitful (particularly useful when the exposure of interest is less amenable to change). The so-called ‘Baron and Kenny’ approach to mediation is very widely used in public health research but makes a number of assumptions that might not always be met. This can produce biased findings, potentially leading to flawed conclusions. Causal mediation methods can overcome or reduce some of these assumptions, although they remain relatively inaccessible to those most likely to apply them.

Learning objectives

In this workshop we will provide public health practitioners and researchers with an overview of the limitations of traditional regression approaches to mediation, including when and why this method may produce biased estimates. We will present alternative approaches in a way that is more accessible than in the methodological literature, drawing on applied research questions to show when different methods might be most appropriately used. During the session participants will learn the differences between direct and indirect effects, and between controlled and natural direct effects. They will work in small groups to draw Directed Acyclic Graphs to consolidate the learning. There will be a series of practical examples of the application of mediation analysis in public health using a variety of methods. Workshop participants should come with research questions in their mind relating to issues of mediation in their own research areas. Throughout the day there will be opportunities to ask questions, and the final session is designed to resolve any final issues that arise during the day.

Track record

Presenters Lynch and Pearce, both public health researchers, have a track record in applying these methods in their own work and teaching them to others. Chairs, Leyland and Gray, are statisticians with extensive experience in teaching advanced statistical methods to mixed levels and disciplines. This material has been successfully presented to a range of audiences, in one-off workshops to public health researchers and social scientists at the University of Glasgow and to epidemiologists and public health doctors at the UCL Institute of Child Health. It has also been run numerous times as part of a causal inference course in Australia.

Biographies

John Lynch, Professor of Epidemiology and Public Health, University of Adelaide

John Lynch is Professor of Epidemiology and Public Health at the University of Adelaide, South Australia. He is a Visiting Professor of Epidemiology in the Bristol Medical School:  Population Health Sciences at the University of Bristol in the UK. He spent 20 years working in North America and before returning to Australia in 2009 he held professorial positions at the University of Michigan in the USA, and at McGill University in Canada.

He is an internationally recognized scholar in epidemiology and public health. In 2005 he was awarded a Canada Research Chair in Population Health. In 2007 his work in social epidemiology and public health was recognized with an Honorary Doctoral Degree in Medical Science from the University of Copenhagen. In 2009 he was awarded a prestigious NHMRC Australia Fellowship. In 2015 he was elected as a Fellow of the Australian Academy of Health and Medical Sciences. 

He has over 350 academic publications with a Google H index of 97. Every year since 2014 he received Thomson Reuters’/Clarivate “Highly Cited Researcher” status that places him in the top 1% of cited scientists internationally in his field. In the last 5 years, he was a Chief Investigator on successful competitive research grants worth over $22 million. He currently serves on several international, national and local scientific advisory groups. He was an editor of the highly ranked International Journal of Epidemiology from 2005-16.

Anna Pearce, Research Fellow, University of Glasgow

Dr Anna Pearce is a Wellcome Trust Research Fellow at the MRC/CSO Social and Public Health Sciences Unit, University of Glasgow. She has fifteen years of experience researching socio-economic inequalities in child health, with a focus on analysing secondary data to understand the pathways through which inequalities in children’s health emerge and options for their reduction. Once a user of ‘Baron and Kenny’, Anna has spent the past five years developing an appreciation of advanced mediation methods and applying them in her own research. As a social epidemiologist with limited formal statistical training, she enjoys the experience of sharing with other researchers the exciting opportunities that advanced mediation methods offer in accessible way.

She is an Honorary Lecturer at the UCL Institute of Child Health, an Associate Fellow of the Higher Education Academy, and Honorary Treasurer of the Society of Social Medicine and Population Health. She was previously Associate Statistical Editor for the Journal of Child Health Care. She has received >£1,000,000 funding, including in personal awards from the Wellcome Trust and UK Medical Research Council, and for the application of causal mediation methods for understanding health inequalities.

Registration fee

The registration fee is € 140 which includes networking lunch and refreshments.