Spatial statistics play a pivotal role in the study and control of tropical diseases, offering a powerful tool to understand their distribution, transmission dynamics, and the factors that drive their prevalence. In addition, spatial statistics is also being used to evaluate the progress of control interventions, such as vaccination campaigns, to ensure equitable immunization coverage and safeguarding vulnerable populations from preventable diseases.
This session delves into the application of cutting-edge spatial statistical methods to address pressing public health challenges that significantly impact populations in low-middle income countries, and provide pathways for more effective disease control and public health interventions.
Speakers:
Alejandro Roza Pozada, Hasselt University, Belgium - "A novel spatio-temporal modelling framework for correlated time series with heterogenous patterns and its applications to malaria outbreaks detection"
Edson Utazi, Southampton University, UK - "Geostatistical approaches for mapping the coverage of childhood vaccination in low- and middle-income countries"
Penelope Vounatsou, Swiss and Tropical Public Health Institute, Basel, Switzerland - "Modelling the impact of climate and vector control coverage interventions to changes in malaria parasite prevalence in sub-Saharan Africa: a spatio-temporal analysis"
Organised by Emanuele Giorgi, Lancaster University
Alejandro Rozo is a doctoral candidate in Statistics at KU Leuven, Belgium. His research interests lie in developing and applying spatial and spatio-temporal methods, focusing on applications in infectious and vector-borne diseases. Alejandro holds a Bachelor’s degree in Statistics from the National University of Colombia and a Master’s in Statistics and Data Science from KU Leuven. Throughout his career, he has been involved in research projects involving academia and the industry, especially in the area of epidemiology.
Mozambique suffers from a high burden of malaria; in 2021, the country accounted for a substantial proportion of global malaria infections and malaria-related mortality (resp. 4.1% and 3.8%). Currently, the Mozambique National Malaria Control Program has put efforts into operationalising a surveillance system in which health specialists and policymakers use spatio-temporal malaria incidence predictions based on geostatistical analysis. The current geostatistical approaches impose a geographical correlation structure on the data, which is assumed to be similar throughout the study region. This is suboptimal for malaria in Mozambique since trends in malaria infections are strongly related to environmental and climatic conditions. These conditions vary considerably throughout Mozambique, suggesting that a modelling approach should allow region-specific variation in spatio-temporal processes.
Using monthly data from 2017 to 2021 from each of the 161 districts of the country, we propose a novel multivariate time series model that allows for a more flexible spatio-temporal analysis. The novelty lies in the fact that Gaussian processes are used to flexibly model temporal trends in residual information. At the same time, we allow spatial correlation across the districts through a conditional autoregressive structure imposed on the scale and smoothness parameters of the Matérn correlation function in these Gaussian processes. We estimate this model using a Bayesian hierarchical approach using Markov Chain Monte Carlo. Using data from 2022, we assess its prediction capability and compare it with traditional methods. Through this, we provide improved spatio-temporal malaria incidence maps for Mozambique.
Dr C. Edson Utazi holds a PhD in Statistics from Lancaster University, UK. His research interests are in developing novel spatial and spatiotemporal methodology and applying this to mapping health and development indicators. Currently, he is a lecturer in the School of Geography and Environmental Science at the University of Southampton, where he also leads the vaccination coverage mapping work of the WorldPop programme. He has published widely in this area of work and has received a number of awards/recognitions.
Over the past two decades, there has been a rapid increase in the production of spatially detailed estimates of health and development indicators (HDIs) such as childhood vaccination coverage. Boosted more recently by the launch of the Sustainable Development Goals (SDGs) in 2015 with the central goal of “leaving no one behind”, these estimates have been shown to be critical to uncovering the heterogeneities that exist in coverage and improving precise targeting of interventions to reach all missed communities and under-served populations. Different methodological approaches have been developed for producing gridded estimates of vaccination coverage using different geolocated input data sets, which are often integrated with geospatial covariate information to improve predictive performance. In this study, we undertake a statistical comparison of these modelling approaches, namely model-based geostatistics (MBG), machine-learning methods and a combination of MBG and machine learning methods. In particular, we compare the predictive performances of these methods under varying point level sample sizes and different assumptions about the functional relationships between the outcome and covariate information. We apply the methods to mapping the coverage of the first doses of the diphtheria-pertussis-tetanus (DTP1) and measles-containing (MCV1) vaccines using household survey data in Nigeria. The findings of our study provide guidance to the users of these approaches, especially in the wider context of mapping SDG indicators or other HDIs.