
Could you introduce yourselves?
Pierre Masselot: I am Assistant Professor in Statistics and Environmental Epidemiology in the Environment and Health Modelling (EHM) Lab at the app of Hygiene & Tropical Medicine (LSHTM). My background is initially in applied statistics, and then I started working in environmental epidemiology during my PhD in Quebec.
Max Eyre: I am an Assistant Professor in the Environmental Health Group (EHG) at LSHTM. I originally studied engineering before moving into the world of statistics and epidemiology during my PhD.
What does your current research focus on?
Pierre: My current research is looking at modelling the association between mortality and environmental stressors, mainly temperature and air pollution. I am particularly interested in understanding the homogeneity of vulnerability to these stressors between different populations and what drives this heterogeneity. Alongside this, I am developing new statistical methods to model the mortality risks in a wide range of locations and subpopulations, which can allow us to predict these risks in populations for which we have less information.
Max: My research focuses on understanding how environmental factors, climate, and water, sanitation, and hygiene (WASH) provisions influence the transmission of infectious diseases, particularly climate-sensitive pathogens in urban settings with several collaborations in Brazil. I develop and apply a range of non-spatial and spatial statistical models to:
- fine-scale community-based studies to understand the mechanisms that drive the transmission of these pathogens
- national-level datasets to make predictions about disease burden to inform public health policy.
You recently ran an event series on causal inference in in environmental epidemiology. Could you tell us a bit about this topic?
Pierre: Causal inference is a framework which aim is to derive conditions and methods for estimating causality in statistics. It is only beginning to be applied in environmental epidemiology, especially for understanding the causality from air pollution to human health. This would help in understanding the effect of various air quality policies, to give one example. The seminars covered a wide range of subjects within this topic including a general introduction to the causal inference framework, spatial confounding, exposure misclassification due to people travelling, and statistical properties of causal inference estimators.
Max: To add to Pierre’s response: One of the key challenges in environmental epidemiology is that the exposures we are interested in (e.g. temperature, rainfall, air pollution) usually cannot be randomly assigned to subpopulations. This means that gold standard approaches to assessing causality via experimental study designs, like randomised-control trials, are largely unfeasible. Instead, we have to use observational data to learn more about their impact on human health, which can make inferring causality challenging.
Causal inference aims to present a solution to this – it consists of a framework and set of methods and assumptions that help us to define, identify, and estimate causal relationships. These methods aim to produce more reliable estimates of the health effects of environmental exposures, to better inform our understanding public health.
Why did you choose this particular topic for the event series?
Pierre: I personally thought that a seminar series was a great way for us to learn more about the causal inference framework, and its potential for environmental epidemiology. The framework is still not used much in our field and we wanted to understand what opportunities causal inference can bring, alongside its limitations and challenges. On the other side, environmental epidemiology is a good topic to expand the horizons of causal inference, accounting for our field specificities. Ultimately, we really wanted to kickstart a discussion within DASH about both of these topics.
Max: Exactly. There’s so much exciting work happening in causal methods, but it often doesn’t make its way into environmental epidemiology, where we’re dealing with complex exposures and systems. Environmental exposures are often unavoidable, affect large populations, and are deeply entangled with structural and social inequalities. Getting the causal story right really matters, both for science and for policy.
For me, this series was about creating a space for learning and exchange of ideas across disciplines (between statisticians, epidemiologists, environmental scientists, and social scientists), as well as building bridges between the methodological advances in causal inference and the real-world complexities of environmental health data. We wanted to use this series as a space to reflect critically and ask: when are these approaches useful in environmental epidemiology, when might they fall short, and how do we adapt them thoughtfully to our context?
Tell us something particularly interesting that you learnt from the seminars.
Pierre: I learned a lot from all the seminars. But the one fact that has really stuck with me was from Heejun’s talk on the impact mobility potentially has in our estimates of mortality related to air-pollution. This is something that might be true more generally for other environmental stressors.
Max: I think something that came across really clearly from Brian Reich’s talk and stuck with me was his candid approach to discussing the use and limitations of causal inference methods in his work. The goal of using causal inference is to reduce bias in estimates, but it simply isn’t possible to prove that we have achieved this with real datasets. Brian emphasised the importance of explaining our assumptions clearly, testing these methods with simulation studies and trying to contextualise our findings within the literature. We know that these approaches aren’t perfect, but we are trying to do better than traditional approaches.
What do you hope that everyone took away from this series?
Pierre: My biggest takeaway from the series relates to the challenges of applying causal inference in environmental epidemiology. It is apparent that applying the framework is not straightforward for a range of reasons, especially since the idea of “treatment” in this context is less clear. It is not binary, and not necessarily observed directly. There is probably a framework that needs to be built for environmental epidemiology and lots of research to be done.
Max: My main takeaway is that this is a rapidly growing area with major opportunities and gaps. Creating spaces like this, where methods can be shared and discussed across disciplines, will be key to adapting and extending causal inference methods to the specific challenges of environmental epidemiology research.
For me, I hope this seminar series gave attendees who were less familiar with causal inference a chance to learn about a range of approaches that could be valuable in their environmental epidemiology research. Ultimately, we saw this seminar series as an opportunity to use DASH as a platform for connecting environmental epidemiologists and causal inference researchers at LSHTM - building shared understanding and sparking future collaborations.
**
A list of all the seminars the environmental epidemiology seminar series can be found below, with links to recordings for each seminar if you're interested:
If you enjoyed this article and would like to build a career in global health, we offer a range of MSc programmes covering health and data, infectious and tropical diseases, population health, and public health and policy.
Available on campus or online, including flexible study that works around your work and home life, be part of a global community at the UK's no.1 public health university.