Scale and Disease Map Surveillance: Improving disease map surveillance by scale expansion and particle filtering.

Abstract: Disease Map Surveillance is largely underdeveloped in the toolkit of the disease mapping community. However, Public Health surveillance is clearly of great and growing importance in a world where pandemics occur! Its clear that when health changes are occurring quickly, then it is important to be able to track and predict changes. To do this in the context of spatial disease monitoring, it is important to consider both the range of data formats available at any given time, but also the tools available to allow the timely and accurate prediction.

This talk will focus on two important linked aspects of this issue: neglected data linkages and timely prediction using sequential Monte Carlo. First, the spatial scale chosen for analysis is clearly important in monitoring disease variation. But for most purposes, there are many different scales of aggregation available at any given time. Choice of temporal scale is also clearly relevant as well. My focus here is to stress that multiple scales must contain information relevant to decision making. For example, it has been found that spatio-temporal infectious disease models are often better when considered linked across spatial scales. The second linked theme is to consider a trade-off between accuracy and timeliness. Bayesian disease mapping has been faulted for the inability to quickly estimate risk features. McMC can be slow in complex and large-scale datasets. Often, approximations are sought (such as Laplace). However, an alternative type of approximation is to consider an evolving parameter set where choices of parameters are made dynamically. In addition, a trade-off can be made between fidelity and speed can be to use state-space models on functions of risk. Particle filtering for state space models is straightforward and, with derived risk functions, can be very fast. Some examples will be provided.

Finally, the relevance of filtration with multiscale models of risk in surveillance will be stressed.

Short Bio: Dr Lawson is a Professor of Biostatistics in the Department of Public Health Sciences, College of Medicine, Medical University of South Carolina (MUSC). He is a MUSC Distinguished Professor Emeritus and ASA Life Fellow. His PhD is from the University of St. Andrews, UK. He has over 250 journal papers on the subject of spatial epidemiology, spatial biostatistics and related areas. In addition to a number of book chapters, he is the author of 10 books in areas related to spatial epidemiology and health surveillance. The most recent of these is Lawson, A. B. (2021) Using R for Bayesian Spatial and Spatio-temporal Health modeling, CRC Press. He has acted as an advisor in disease mapping for the World Health Organization (WHO) and is founding editor of the Elsevier journal Spatial and Spatio-temporal Epidemiology. Dr Lawson has delivered many short courses in different locations over the last 20 years on Bayesian Disease Mapping and more general topics.

Spatial data science for epidemic intelligence

Abstract: Spatial epidemiology is an interdisciplinary domain where a lot of spatial and spatiotemporal data come together. As in many such domains, a lot of problems, varying from dealing with obscure file formats to the proper handling of spatial and temporal coordinate reference systems, come together and are collected in what we call spatial data science. Statistical inference, predicting spatial patterns of past or future disease outbreaks, or advising on counter-measures during a disease outbreak, not only call for different data, but also for different statistical methods and tools, ranging from linear regression to machine learning and deep learning. We will also try to address the role of large language models and foundation models in the practice of epidemic intelligence research.

Short Bio: Edzer Pebesma is a professor at the Institute for Geoinformatics, University of Münster, Germany, where he leads the spatiotemporal modelling lab. He is the author of the 2013 book “Applied Spatial Data Analysis with R” and the 2023 book “Spatial Data Science: With applications in R”, and author of or contributor to several key R packages for spatial data analysis, including sf, stars, gstat, and s2. Beyond spatial statistics, his research interests go into how spatial data science languages (Python, R, Julia, …) compare and how they deal with common, underlying challenges of spatial data analysis, such as change of support, handling geographic coordinates, or fitting models to very large datasets.

Spatio-temporal modelling of emerging infectious diseases to strengthen resilience against climate change

Abstract: Extreme climatic events, environmental degradation, rapid urbanisation, and socio-economic inequalities are intensifying the risk of infectious disease emergence and spread. Mosquito-borne diseases such as dengue and malaria are particularly sensitive to climate variability and change, with warming trends extending transmission seasons and shifting geographical ranges into regions with little immunity or weak health systems. More frequent storms, floods, and droughts further disrupt disease dynamics, driving outbreaks of vector- and water-borne diseases. This talk will explore advances in spatio-temporal modelling to assess past, present, and future risks of emerging infectious diseases, and highlight the partnerships, data, and tools needed to build climate resilience, strengthen preparedness, and support adaptation in public health systems.

Short Bio: Rachel Lowe is an ICREA Research Professor and Global Health Resilience (GHR) Group Leader at the Barcelona Supercomputing Center (BSC). The GHR group co-develops policy-relevant methodological solutions to enhance surveillance, preparedness, and response to global health challenges, with a focus on climate-sensitive infectious diseases. Rachel coordinates two Wellcome Trust digital technology, climate, and health projects, HARMONIZE and IDExtremes, which aim to provide robust data and modelling tools to build local resilience against emerging infectious disease threats. Rachel is a member of the World Meteorological Organization World Weather Research Programme (WWRP) and the Lancet Commission for Strengthening the Use of Epidemiological Modelling of Emerging and Pandemic Infectious Diseases. Rachel was nominated as a lead author of the IPCC Seventh Assessment Report (WGII) chapter on health and well-being.

Statistical and Machine Learning for Geospatial Big Data

Abstract: Geospatial data, routinely encountered in environmental health, climate sciences, disease epidemiology, forestry, and ecology, have traditionally been analyzed using statistical models built on foundations of stochastic processes. Increasingly, practitioners are adopting machine learning methods for geospatial analysis. Should the decades of development in spatial statistics be abandoned for black-box machine learning? This short course demonstrates the pitfalls of naive machine learning for geospatial data and offers a tutorial on state-of-the-art hybrid machine learning methods for geospatial data that leverage well-established spatial statistics principles.

We review traditional geospatial approaches, such as the linear mixed model using a Gaussian process (GP), and non-linear machine learning methods, including random forests and neural networks. We then present hybrid approaches that embed non-linear machine learning within traditional geospatial mixed effect models, relaxing the stringent assumption of linearity, while preserving Gaussian process random effects and thereby retaining the interpretability, flexibility, and parsimony for estimation and prediction. We present recent methods with different choices of machine learning algorithms (random forests and neural networks) and different data types (continuous and binary), and provide demonstrations with published software on real and simulated data. The tutorial will equip practitioners with a suite of hybrid machine learning methods and software tailored for geospatial analysis.

Short Bio: Dr. Abhi Datta is a Professor in the Department of Biostatistics at the Johns Hopkins Bloomberg School of Public Health. His research develops statistical and machine-learning methods for geospatial data with applications to air pollution, forestry, ecology, and infectious‑disease modeling, and Bayesian models for combining multi‑source data to improve public‑health metrics such as cause‑specific mortality estimates. Dr. Datta serves as an associate editor for multiple statistical journals, including JASA, JRSSB, and Biometrics. He is an elected Fellow of the American Statistical Association and has received multiple early-career awards, including the Emerging Leader Award from COPSS, the Abdel El-Shaarawi Early Investigator’s Award from The International Environmetrics Society (TIES), and the Early Career Award from the International Indian Statistical Association (IISA).  

Spatial and Spatio-Temporal Modelling with inlabru

Abstract: The Integrated Laplace Approximation (INLA) has become an established Bayesian inferential framework for a wide range of complex models, including spatial and spatio-temporal models, as it enables accurate and computationally efficient estimation within the broad class of Latent Gaussian Models (LGMs). The continued popularity of INLA is driven not only by its methodological strengths but also by ongoing development and derivative projects that extend its functionality.

One notable example is inlabru, a user-friendly wrapper for R-INLA that facilitates model specification while adding further functionality to the existing R-INLA framework. Designed to improve accessibility, inlabru allows advanced modelling techniques to be applied more widely. Specifically, it supports the integration of observational processes and heterogeneous data sources into joint models and extends the range of model structures by accommodating non-linear relationships.

This workshop will introduce the fundamental principles of inlabru and provide an overview of its general capabilities. The sessions will combine illustrative examples with practical exercises to demonstrate the application of the software across a range of modelling contexts.

Short Bio: Janine B. Illian is the Chair in Statistical Sciences and Head of Statistics at Glasgow University. Her work focuses on developing realistically complex spatial and spatio-temporal modelling methodology. She is the author of “Statistical Analysis and Modelling of Spatial Point Patterns” (Wiley, 2008), which has been a standard work on point process modelling since its publication. Her research profile focuses on the development of modern statistical methodology that is computationally feasible, relevant to, and usable by end-users, especially in the context of integrated nested Laplace approximation, INLA. She has taken complex spatial modelling approaches from the theoretical literature into the real world and is encouraging statistical development by fostering strong relationships with the user community, in particular in the context of ecology. Her EPSRC funded work has led to the development of the software package inlabru that features increased flexibility of modelling approaches and observation processes combined with user-friendly implementations.