Pre-Conference Workshops
Full Day Workshops (08:00 - 15:30)
N. Nikolaou, D. Cea, K. Wolf
Helmholtz Munich, Neuherberg, Germany
Background: Human health is shaped by a complex interplay of individual characteristics, environmental exposures, socio-economic and neighborhood contexts. While machine learning (ML) offers powerful tools to model such high-dimensional and correlated data, many researchers face barriers related to methodological complexity, reproducibility, and interpretability. There is a growing need for transparent, user-friendly ML workflows that integrate explainability and can be readily applied to epidemiological research.
Workshop aim: This full-day pre-conference workshop introduces MLEE, an open and generalizable ML framework designed to model associations between environmental, individual, and socio-economic factors and health outcomes. The workshop combines theory, pipeline design, and hands-on practice using open-source data and Python programming.
Format and Content: The morning session consists of a lecture on ML methods commonly used in epidemiology (including regularized regression, distance-based methods, tree-based algorithms, and neural networks), followed by an introduction to the MLEE pipeline architecture. Topics include data preprocessing, feature engineering, model selection, hyperparameter optimization, validation strategies, and explainability approaches such as SHapley Additive exPlanations (SHAP). The afternoon session focuses on practical implementation. Participants will work hands-on with Python to apply the MLEE pipeline to open-source environmental and health-related data. Guided exercises will cover model training, performance comparison, interpretation of results, and subgroup analyses. The instructors will provide step-by-step support and discussion to allow sufficient time for Q&A.
Outcome: Participants will leave with introductory knowledge in ML, and practical skills to implement an explainable ML pipeline, supporting open, reproducible and interpretable research.
M. Chadeau-Hyam1, R. Vermeulen2, A. Peters3
1 Imperial College, London, United Kingdom
2 Utrecht University, Utrecht, Netherlands
3 Helmholtz Zentrum München, Munich, Bavaria, Germany
Since the introduction of the Exposome concept in the early 2010’s, several generations of research projects have been funded globally. These have given rise to different interpretations, definitions and implementations of Exposome studies, borrowing from a range of scientific fields. In the present workshop, leveraging our >15 years involvement in Exposome projects, we propose to define, explore, illustrate, and discuss the future of Exposome sciences.
The morning session will be dedicated to the key concepts and methodological challenges of exposome research.
We will first provide an illustrated historical overview of Exposome research globally, highlighting the key steps, the main challenges, established resources for exposome science (R Vermeulen, Chair). We will subsequently focus on the main approaches to provide valid, reproducible and interpretable measures of the external (K de Hoogh) and internal (C Chatziioannou) exposome in large and diverse populations. Finally, we will provide an overview of the methodological needs and available tools for Exposome Analytics (M Chadeau-Hyam, co-chair). We will conclude with an overview of the key priorities that need to be addressed in the next generation of Exposome research projects.
The afternoon session (4 hours) will consist in several real-world examples of exposome research from the young Exposome researcher community of the EXPANSE project. These will be delivered in a tutorial format to ensure and accelerate the uptake of the scientific questions, methods and results proposed. We will cover the fields of (i) Exposome-based risk prediction (J de Bont), (ii) molecular characterisation of the exposome (M Vogli, S Dagnino), (iv) exploration of the key characteristics of the social exposome (R Castagne), (iii) interpretable approaches for mediation analyses for high-dimensional data (R Colindres-Zuehlke), and (iv) the use of agent-based models to estimate the effect of the exposome on the population (T Sonnenschein).
M. Willis1, D. Pacyga2, J. Casey3
1 Boston University, Epidemiology, Boston, Massachusetts, United States of America
2 University of North Carolina at Chapel Hill, Epidemiology, Chapel Hill, North Carolina, United States of America
3 University of Washington, Environmental and Occupational Health Sciences, Seattle, Washington, United States of America
Sponsored by the International Society for Environmental Epidemiology Communications Committee
Description: This intensive, hands-on workshop is designed to integrate generative artificial intelligence (Gen AI) into the complete environmental epidemiology research cycle. Participants will work with a public dataset and apply Gen AI tools at every stage, from initial research ideation and comprehensive literature review to drafting a final manuscript or grant section. We will define key concepts, survey the current landscape of tools (LLMs, code assistants, research agents), and highlight capabilities and limitations relevant to scientific research. The workshop will then proceed to cover advanced prompt engineering, automated code generation, statistical output verification, and data augmentation techniques. A primary focus will be on the critical evaluation of AI outputs, such as diagnosing errors, validating statistical results, and ensuring responsible use and reproducibility. By the end of the full day, participants will have a practical, repeatable ‘research workflow’ that positions Gen AI as a sophisticated, verifiable partner in their day-to-day research.
Target audience: Environmental epidemiologists, public health researchers, data scientists working in environmental health, and graduate students interested in the intersection of Gen AI and epidemiology. Please bring a laptop; no prior experience with Gen AI is required.
M.A. Haque1, M.K. Uddin2
1 Bangladesh Medical University, Public Health and Informatics, Dhaka, Bangladesh
2 University of Dhaka, Department of Psychology, Dhaka, Bangladesh
Understanding the psychological mechanisms underlying pro-environmental behavior is essential for designing effective interventions to address climate change and its public health impacts. This full-day workshop is designed for PhD students, early-career researchers, and practitioners interested in environmental psychology, behavioral science, and psychometrics. It introduces key theories, measurement tools, and applied analytical techniques for assessing climate-related risk perception and pro-environmental behavior using the R programming language.
The workshop comprises seven interactive sessions. It begins with an overview of major behavior change theories relevant to environmental and health-protective behaviors, including the Theory of Planned Behavior, Social Cognitive Theory, Health Belief Model, and Protection Motivation Theory. Participants are introduced to widely used psychometric instruments such as the General Ecological Behavior Scale, Pro-Environmental Behavior Scale, New Ecological Paradigm Scale, and Connectedness to Nature Scale, with particular emphasis on measuring climate-related risk perception, including perceived susceptibility, severity, and threat appraisal.
Subsequent sessions focus on measuring latent constructs such as environmental concern, climate change skepticism, willingness to adopt sustainable practices, and policy support. Participants explore survey-based and observational approaches, followed by an introduction to key psychometric frameworks, including Classical Test Theory and Item Response Theory. Core concepts of reliability, validity, and factor analysis are discussed, covering both exploratory and confirmatory methods.
The final sessions provide hands-on training in survey design, scale validation, and IRT modeling using R, with real-world environmental and behavioral health datasets. By the end of the workshop, participants will gain practical skills to develop, evaluate, and apply psychometric tools for climate and environmental health research.
Half Day Workshops (Morning, 08:00 - 11:30)
C. M. Howlett-Downing
South African Medical Research Council, School of Health Systems and Public Health, Environmental Health, Climate Change and Health, Pretoria, Gauteng, South Africa
Distributed lag methods are widely used in environmental epidemiology to evaluate delayed health effects of short-term exposure. However, most training and applied examples assume dense exposure monitoring, daily temporal resolution, and complete meteorological covariates—conditions that are rarely met in low- and middle-income country (LMIC) settings. As a result, researchers working with parsimonious datasets often face a gap between recommended analytic approaches and what is feasible in practice.
This pre-conference workshop presents an applied framework for adapting Distributed Lag Case-Crossover (DL-CCO) models to exposure-limited epidemiological datasets, including those based on sparse monitoring networks, weekly aggregation, or incomplete covariate information. The workshop situates DL-CCO within the broader family of case-crossover and distributed lag approaches, clarifying its role as a robust design for estimating short-term exposure–health associations under constrained data environments.
Participants will be guided through key design considerations relevant to LMIC contexts, including referent selection strategies, lag specification under temporal aggregation, exposure scaling, and interpretation of odds ratios when traditional Poisson-based distributed lag models are not viable. Emphasis is placed on transparency of assumptions, conservative interpretation, and explicit discussion of methodological trade-offs, rather than causal inference.
The workshop combines conceptual instruction with hands-on implementation using example datasets that reflect real-world LMIC data structures. Practical sessions will demonstrate model specification, diagnostics, common failure modes, and sensitivity analyses, including comparison with alternative approaches such as pseudo-case-crossover and simplified time-series models. Guidance on reporting practices, reviewer expectations, and defensible language for publication will also be provided.
This workshop is intended for early- and mid-career researchers, analysts, and practitioners seeking to apply distributed lag methods responsibly in data-constrained settings. By focusing on methodological adaptation rather than idealised conditions, the workshop directly supports the ISEE 2026 theme of understanding and responding to global and local environmental health challenges.
H. Hu1, C. Tao2, J. Bian3, S. Niu2, X. He3, C. Leiser1
1 Brigham and Women’s Hospital, Channing Division of Network Medicine, Boston, Massachusetts, United States of America
2 Mayo Clinic, Department of Artificial Intelligence and Informatics, Jacksonville, Florida, United States of America
3 Indiana University, Department of Biostatistics & Health Data Science, Indianapolis, Indiana, United States of America
Environmental epidemiology increasingly leverages rich spatial and contextual data sources – such as air pollution, greenness, climate, built environment, and neighborhood social context – to characterize the exposome. However, these data often exist on different spatial and temporal scales, lack a unified ontology, and require complex linkage to epidemiologic or clinical cohorts before downstream analyses can be performed. This half-day workshop will introduce SPACESCANS (spatial and contextual exposome semantic data integration system), a framework and toolset designed to support ontology-driven organization of spatial and contextual exposome data, spatiotemporal linkage, and exposome-wide association and predictive modeling pipelines.
The outline of the workshop is as follows:
- Background on the spatial and contextual exposome and presentation of an ontology to organize natural, built, and social environmental domains, highlight metadata standards, and support harmonization across datasets.
- Overview and live demonstration of the SPACESCANS spatiotemporal linkage workflow, including defining geographic units and buffers, aligning spatial grids and time windows, and linking multiple environmental datasets to cohort data. Participants will be guided through hands-on exercises using a demo dataset.
- Demonstration of downstream analysis pipelines using linked SPACESCANS outputs, including exposome-wide association studies (ExWAS) and predictive modeling. We will discuss approaches for handling high-dimensional, correlated exposures, feature engineering, and model validation.
- Question and answer session and discussion of future directions for spatial/contextual exposome tools and standards.
Following completion of the workshop, attendees will have a better understanding of: (1) ontology-based organization of spatial and contextual exposome data; (2) practical steps to perform spatiotemporal linkage using SPACESCANS; and (3) strategies for conducting ExWAS and predictive modeling using linked exposome data.
A personal computer with R installed is suggested. Participants will be able to follow along with the slides and explore the SPACESCANS tools in real time during the workshop.
Half Day Workshops (Afternoon, 12:00 - 15:30)
P. Masselot
London School of Hygiene & Tropical Medicine, Environment & Health Modelling (EHM) Lab, London, United Kingdom
Quantitative health impact assessment is becoming an important tool of modern environmental epidemiology. It aims to quantify the burden associated with environmental exposures under various counterfactual scenarios or pathways and is useful for policy and intervention evaluation. The quality and usefulness of health impact assessments depends on several factors including exposure data, baseline health case rates and, most importantly, available exposure-response functions. These functions are key to accurate health impact assessments but are not always readily available to researchers in the environment-health community, often due to technical and computational barriers.
The objective of this workshop is twofold: i) to advocate for wider availability of exposure-response functions in environmental epidemiology, and ii) to introduce techniques and methods for health impact assessments in practice. In a first part, we will discuss barriers and strategies to storing, exploiting and sharing exposure-response functions. This includes necessary information, formats, online solutions and retrieval of exposure-response functions. In a second part, I will show how exposure data and exposure-response functions are used for health impact assessment. This includes burden computation under different scenarios as well as uncertainty quantification.
This workshop will take the form of a tutorial focusing on the practical aspects of the process with the R software. We will see how to export epidemiological results from R to be used in health impact assessments and solutions for storing. We will also code a simple health impact assessment in R. The workshop will use a dataset of exposure-response functions for temperature-related mortality I have produced to illustrate the various aspects of health impact assessment.
T. Benmarhnia1, 2, Y. Ma1
1 University of California San Diego, Scripps Institution of Oceanography, La Jolla, California, United States of America
2 University of Rennes, Rennes, France
Quasi-experimental methods have been increasingly applied in environmental epidemiology to evaluate the effectiveness of health interventions or to assess the health impacts of extreme weather events. Capitalizing on the timing of natural experiments, difference-in-differences and synthetic control methods can be used to estimate causal effects under specific identification assumptions, especially when it is impractical or infeasible to conduct a randomized controlled trial. Such methods require one or more control groups not exposed to the event; however, suitable control groups are not always available. In this context, the interrupted time series (ITS) design offers a valuable alternative.
The ITS design is well suited to evaluating the population-level health effects of events or interventions that occur at a clearly defined time point in a single exposed or intervened population. This design extrapolates the time-series data from the pre-event phase to generate a counterfactual trend that represents what would have been observed (under specific assumptions) in the absence of the event. This method has been widely used in public health settings, including the evaluation of health impacts of environmental policies or extreme weather events such as floods or wildfires.
In this workshop, we will first provide a theoretical overview of both traditional and recently proposed ITS modeling approaches under the potential outcomes framework, spanning from standard one- and two-stage models to machine learning-integrated methods and extensions for staggered intervention scenarios. We will discuss the assumptions, model specifications, advantages, and potential methodological challenges of these approaches. We will focus on examples related to air quality warning systems and wildfire events. In the second part of this workshop, we will showcase multiple ITS modeling approaches using R programming language, with examples from environmental epidemiology to facilitate translation to real-world data analysis. We will guide attendees to work on hands-on programming exercises.
S. Baierl1, 2, M. Coenen1, 2
1 LMU Munich, Chair of Public Health and Health Services Research, Institute for Medical Information Processing, Biometry, and Epidemiology (IBE), Munich, Germany
2 Pettenkofer School of Public Health, Munich, Germany
Urban green spaces contribute to the mental well-being and health of the urban population while also providing habitats for biodiversity. The transdisciplinary project CitySoundscapes investigates how urban green spaces must be structured, equipped, and distributed in cities to foster biodiversity while simultaneously serving as a health resource for visitors. A particular focus is placed on the acoustic environment (soundscapes) of green spaces. Soundscapes not only influence mental well-being, but are also indicators of animal diversity (e.g., birdsong), environmental features (e.g., trees in the wind, water), and stress factors (e.g., traffic, noise). Thus, they enable quantifying the relationship between biodiversity and mental health.
In CitySoundscapes, participatory soundwalks are used to investigate the perception of soundscapes and its association with mental well-being and health. During the soundwalks, participants complete an online questionnaire at different green and control sites to assess the perceived soundscape quality based on the ISO 12913-2 standard, as well as their emotional affect using validated instruments such as Self-Assessment Manikin (SAM). The questionnaire further includes items on sociodemographic characteristics and nature connectedness. Additionally, audio recordings of the heard sounds are collected, and meteorological parameters are recorded for each walk. The study is conducted in three selected Munich districts that represent different urban environments and sociodemographic structures.
In this workshop, participants can actively take part in the CitySoundscapes study by joining one of the soundwalks. Participants follow a designated route through a green space near the Isar river and pause at several locations to listen to their surroundings for a few minutes. They share their impressions through the questionnaire. The soundwalk is accompanied by thematic inputs on the CitySoundscapes project, the soundwalk method, and preliminary findings on the relationships between soundscapes, biodiversity and mental health.
Meeting point: Kolumbusplatz station (subway exit D, Falkenstraße)
