Modelos espacio-temporales bayesianos para estudiar la incidencia de dengue en el Perú

• Autor: Caro Ferreyra, Katia Alejandra
• Formato: tesis de maestría
• Fecha de Publicación: 2024
• Institución: Pontificia Universidad Católica del Perú
• Repositorio: PUCP-Tesis
• Lenguaje: español
• OAI Identifier: oai:tesis.pucp.edu.pe:20.500.12404/30062
• Enlace del recurso: http://hdl.handle.net/20.500.12404/30062
• Nivel de acceso: acceso abierto
• Materia: Dengue--Perú--Prevención; Estadística bayesiana; Enfermedades transmisibles--Investigaciones--Perú; https://purl.org/pe-repo/ocde/ford#1.01.03

The prevention of dengue requires a system to identify areas at higher risk, using epidemiological data with spatial and temporal structure. Bayesian approaches, which integrate prior information and handle hierarchical structures, provide a flexible and robust method that allows for more accurate uncertainty estimates, as well as capturing spatial and spatiotemporal correlation, accounting for this variability in disease risk estimates. These hierarchical Bayesian approaches often require sophisticated numerical methods to provide parameter estimates. In this context, methods such as Markov Chain Monte Carlo (MCMC) or Integrated Nested Laplace Approximation (INLA) can be applied, the latter being a more computationally efficient alternative for latent Gaussian models (LGM), including spatial models such as the hierarchical Besag, York, and Molli´e (BYM) model, which can be extended to spatiotemporal analyses, being very useful for evaluating the count of cases over time. In this framework, the present study evaluated three Bayesian models: a hierarchical model with a parametric linear trend, a hierarchical model dynamically modeled using a random walk, and a non-parametric dynamic trend model with spatiotemporal interaction. To demonstrate the contribution of this proposal, the three models were fitted to real data that included both dengue cases and their incidence. In the model selection procedure, not only was the suitability of the models compared, but also different count distributions were analyzed, adding climatic covariates to the analysis.