Guilherme Oliveira (McGill University)

Title: A Zero-Inflated Spatiotemporal Model for Underreported Infectious Diseases Counts

Abstract

Underreporting of disease cases is a recurring challenge in epidemiology, which introduces bias into the statistical estimation of disease rates. Although many approaches for modeling underreported count data have been proposed in recent years, there remains a lack of methods that address data correction within a spatiotemporal framework. This limitation is especially pronounced in analyses based on less aggregated time periods and small geographic areas, where excess zeros are frequently observed. Zero-inflation can be caused by both the absence of the disease and underregistration. In this talk, after briefly revisiting some existing approaches for modeling underreported count data, I will introduce a zero-inflated model that explicitly accounts for both the absence of the disease (true zeros) and an imperfect counting process. Conditional on disease presence, the observed count follows a Binomial thinned zero-truncated negative binomial distribution, which may lead to the observation of zeros even when the disease is present but goes undetected. We consider a spatiotemporal setting, and inference follows the Bayesian paradigm. By taking into account underreporting, excess zeros, and spatiotemporal heterogeneity, the proposed modeling strategy aims to provide more realistic estimates for associated disease rates. In this way, decision-makers can make more informed and accurate decisions for disease control and prevention. Simulation studies are performed to explore the model's behavior under different levels of presence and underreporting, as well as in distinct data generation processes. We apply the model to the cases of chikungunya infection in Rio de Janeiro, Brazil.

Speaker bio

Guilherme Oliveira is an Associate Professor of Statistics at the Federal Center for Technological Education of Minas Gerais (CEFET-MG), Department of Computer Sciences, in Belo Horizonte, Brazil. He received his PhD in Statistics from the Federal University of Minas Gerais (UFMG) in 2020. From May 2025 to April 2026, he is on sabbatical leave as a visiting professor at EBOH, McGill University. His research and funded projects have focused on Bayesian methods for analyzing underreported data, with applications in Public Health, Epidemiology, and the Social Sciences. Areas of interest include spatiotemporal modeling, disease mapping, measurement error, and machine learning. For more information, please visit: https://sites.google.com/view/guilherme-deoliveira/.