Special IoMH meeting on Mediation analyses

Mediation analyses aim to identify intermediate mechanisms on the pathway from an exposure to an outcome. Crucially, such intermediate mechanisms can provide targets for compensatory interventions when early exposures are difficult to or cannot be change to psychology, psychiatry and many other disciplines. Statistical approaches to mediation analyses have considerably evolved in the past decades. In particular, recent research has relied on the counterfactual framework (i.e. a theoretical framework to formalize causal inference) to introduce causal mediation analyses. This approach provides a new set of concepts and analytical methods that deserve to be more widely understood and applied. Here, our three speakers will outline the novelty of these approaches compared to previous estimation methods, provide an overview of novel concepts and available analytical methods, and discuss their usefulness in identifying mediators of treatment effects in randomised controlled trials.

Speakers and talks:

Bianca L De Stavola

Professor of Medical Statistics , UCL Great Ormond Street Institute of Child Health

Title: On the two mediation analysis schools

The study of mediation has a long tradition in the social sciences and beyond anchored within path analysis and structural equation models (SEMs), and a more recent school mainly developed in epidemiology which stems from methods developed within the potential outcomes (counterfactual) approach to causal inference.

I will introduce the mediation effects defined in the SEM literature and then revisit them through the lens of counterfactual-based causal inference. This will allow an appreciation of the similarities of the two approaches, and hint at the advantages given by the greater formality and flexibility of the latter.

Biography: Bianca De Stavola is Professor of Medical Statistics at UCL Great Ormond Street Institute of Child Health, London, UK. 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 the University of Padova, Italy. Her main research activities involve the understanding, development, and implementation of statistical methods for long-term longitudinal studies, with specific applications to causal enquiries in life-course epidemiology. In 2020 Bianca received the Suffrage Science Award for Mathematics and Computing and in 2021 the Royal Statistical Society Bradford Hill Medal in Medical Statistics.

Rhian Daniel (Cardiff).

In a growing number of modern applications, it is contended that interest lies in decomposing the effect of an exposure on an outcome into its effects via a large number of mediators. In the biomedical sciences, these are often transcriptomic, proteomic or metabolomic measurements, or other such high-dimensional biomarkers. For example, when trying to understand the mechanism through which a particular genetic variant affects a cardiovascular outcome, one might attempt to decompose the total effect of the variant into individual path-specific effects through hundreds of blood metabolite measurements, such as different molecular sub-types of cholesterol.

It is often somewhat loosely argued that the aim of such an analysis is to gain insight into mechanisms linking an exposure to an outcome, since this understanding may lead to ways of disrupting or improving it.

In this talk, I will discuss how such aims relate to what is targeted by a conventional mediation analysis, and two ways in which an estimation strategy may be developed that does not rely too heavily on either structural or parametric assumptions about the high-dimensional mediators.

I will illustrate the ideas using an analysis of a dataset taken from the UK Biobank in which the multi-dimensional mediators of genetic associations with Covid-19 hospitalisation are investigated.

Biography: Rhian Daniel is a statistician working at the Division of Population Medicine, Cardiff University. Her work lies at the intersection of methodology and medical applications, with a particular focus on estimation and inference for causal estimands.

Professor Richard Emsley

Department of Biostatistics & Health Informatics, Institute of Psychiatry, Psychology & Neuroscience, King’s College London ; NIHR Maudsley Biomedical Research Centre

Title: Applications of causal mediation analysis in mental health trials

In this talk, we will consider four specific challenges that can arise when applying causal mediation analysis methods in randomised clinical trials in mental health:

1. Why intervention-specific measures like therapeutic alliance and fidelity are not mediating variables but should be considered as post-randomisation effect modifiers
2. How to account for outcomes collected during the intervention window as time-varying confounders of the mediator-outcome relationship using parametric g-formula
3. Combining adjustment for departures from random allocation and mediation analysis by extending the definition of the causal mediation estimands to include compliance and estimating these using instrumental variable structural equation models
4. Accounting for differential uptake of components of ‘treatment as usual’, for example, the use of medication in psychosocial intervention trials, by addressing the appropriate question in the context of the estimand framework

We illustrate these issues using applied examples from trials of psychological and digital interventions in people with psychosis.

Biography: Richard is an NIHR Research Professor and Professor of Medical Statistics and Trials Methodology at the Institute of Psychiatry, Psychology and Neuroscience. Before joining King's, he held a personal Chair as Professor of Medical Statistics at The University of Manchester. His research interests are in clinical trials methodology and developing statistical methods for efficacy and mechanisms evaluation using causal inference approaches. The applications of these methods focus on randomised trials of complex interventions in mental health. He is lead of the KCL Trials Methodology Research Group, and co-lead of the Statistical Analysis Working Group in the MRC-NIHR Trials Methodology Research Partnership.