B0935
Title: Spatiotemporally modelling opportunistically sampled epidemiological data: pitfalls and solutions
Authors: Alejandro Rozo Posada - KU Leuven (Belgium)
Arne Janssens - KU Leuven (Belgium)
Christel Faes - Hasselt University (Belgium)
Pieter Libin - Vrije Universiteit Brussel (Belgium)
Jonas Crevecoeur - KU Leuven (Belgium)
Thomas Neyens - Hasselt University (Belgium) [presenting]
Abstract: Large-scale survey databases obtained through the voluntary participation of data collectors and/or providers have become popular platforms for collecting timely data on epidemiological phenomena. Especially for diseases that occur within a population with considerable spatio-temporal variability, these databases are useful for surveillance that supports policymakers in planning local health interventions. Such surveillance systems typically use spatial or spatiotemporal statistical models to detect locations that show elevated disease risk. However, the voluntary nature of participation in the studies from which these databases originate leads to opportunistically sampled data that often showcase spatiotemporal imbalance and varying reporting efforts, among other problems. Although many have voiced concerns about these issues, it remains unclear if and how they invalidate model-based spatiotemporal insights into disease risk, especially in the context of large datasets. Via case and simulation studies using data from the Belgian Great Corona study, a public, online, weekly COVID-19 survey, and INTEGO, a database collected by voluntary general practitioners, the effects of spatiotemporal sample imbalance is investigated, varying reporting efforts, and sample size, on the performance of spatially discrete spatiotemporal statistical models. Pitfalls in and solutions for designing such studies and analyzing the data they provide are discussed.
