Analyzing Bayesian hierarchical models for spatiotemporal dengue incidence prediction across the Brazilian cerrado municipalities
Journal of Mathematical Problems, Equations and Statistics · 01.07.2024 · 5 (2) · 217-221
DOI: 10.22271/math.2024.v5.i2c.313 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Mathematical Problems, Equations and Statistics, online erschienen am 01.07.2024. Verfasst von Marcelo Figueiredo und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Between 2015 and 2022, dengue fever notifications in the Brazilian Cerrado biome increased by 214%, yet predictive models that capture both spatial clustering and temporal seasonality across municipalities remain scarce. This research develops a Bayesian hierarchical model (BHM) for spatiotemporal prediction of weekly dengue incidence rates across 147 municipalities in the Cerrado region of central Brazil. The model comprises three layers: a Poisson observation layer for case counts, a latent Gaussian process layer that decomposes the log-relative risk into a spatially structured component (modelled via an intrinsic conditional autoregressive prior), a temporally structured component (modelled via a first-order random walk), and a space-time interaction term. Covariate effects for temperature, precipitation, vegetation index, and population density enter the linear predictor at the municipality level. Posterior inference was carried out through Markov chain Monte Carlo sampling using a No-U-Turn Sampler implemented in Stan, with four chains of 5,000 post-warmup iterations each. The research compared the proposed BHM-ST model against three alternatives: a spatial-only BHM (BHM-S), a temporal-only BHM (BHM-T), and a standard generalised linear model (GLM). Model comparison via the deviance information criterion (DIC) and the widely applicable information criterion (WAIC) showed that BHM-ST achieved a DIC of 2,847 and a WAIC of 2,903, outperforming all competitors by margins ranging from 18.4% to 32.5%. Leave-one-municipality-out cross-validation yielded a mean absolute prediction error of 11.3 cases per 100,000 population per week. The posterior predictive distributions correctly identified 83.7% of outbreak weeks across all municipalities, suggesting that the model can serve as a practical early-warning tool for public health authorities in the Cerrado.
Zitieren
APA
Figueiredo, M., & Campos, R. (2024). Analyzing Bayesian hierarchical models for spatiotemporal dengue incidence prediction across the Brazilian cerrado municipalities. Journal of Mathematical Problems, Equations and Statistics, 5(2), 217-221. https://doi.org/10.22271/math.2024.v5.i2c.313
Vancouver
Figueiredo M, Campos R. Analyzing Bayesian hierarchical models for spatiotemporal dengue incidence prediction across the Brazilian cerrado municipalities. Journal of Mathematical Problems, Equations and Statistics. 2024;5(2):217-221. doi:10.22271/math.2024.v5.i2c.313
BibTeX
@article{Figueiredo2024,
author = {Figueiredo, Marcelo and Campos, Renata},
title = {Analyzing Bayesian hierarchical models for spatiotemporal dengue incidence prediction across the Brazilian cerrado municipalities},
journal = {Journal of Mathematical Problems, Equations and Statistics},
year = {2024},
volume = {5},
number = {2},
pages = {217--221},
doi = {10.22271/math.2024.v5.i2c.313}
}
EUVIMED