Análisis espacial jerárquico bayesiano sobre la asociación entre la letalidad de las personas diagnosticadas con cáncer y la exposición ambiental a plaguicidas en Costa Rica, 2011-2015
Fecha
2024
Tipo
tesis de maestría
Autores
Quesada Leitón, Hazel Paola
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
El objetivo principal de esta investigación fue identificar un método unificado para analizar datos relacionados con mortalidad por cáncer de mama y próstata en Costa Rica, considerando aspectos jerárquicos, espaciales y ecológicos. Esto se basó en un modelo lineal jerárquico binomial, cuya eficacia se evaluó a través de simulaciones. Inicialmente, se llevó a cabo un estudio empírico para comprender la relación entre las variables y construir un índice de exposición a plaguicidas. Los resultados mostraron que los modelos cuya unidad geográfica corresponde a los conglomerados (unión de distritos dentro de provincias) mejoraron el rendimiento, en comparación con el agrupamiento original por distritos. Se generaron varios escenarios de simulación para identificar el modelo óptimo, variando tanto la unidad geográfica como el coeficiente de regresión asociado a la exposición a plaguicidas. La simulación demostró que los modelos jerárquicos y espaciales superaron al modelo simple en términos de precisión y ajuste. Aunque en los diferentes escenarios de simulación evaluados el modelo jerárquico fue ligeramente superior según la log verosimilitud marginal, los modelos jerárquicos espaciales fueron preferidos debido a su capacidad para controlar el sesgo en la estimación del error debido a la correlación de datos. La selección del mejor modelo se basó en medidas de bondad de ajuste como la log verosimilitud marginal, criterio de información de devianza y criterio de información de Akaike Watanabe, donde el agrupamiento de distritos dentro de provincias mediante análisis de conglomerados mostró el mejor ajuste. Los modelos espaciales con criterio de vecindad de reina fueron identificados como los más adecuados para el análisis de los datos. Además, se subraya la importancia de controlar el uso de plaguicidas, dada su posible relación con el cáncer. Se insta a las instituciones encargadas del registro de plaguicidas a implementar un control más estricto para mejorar la calidad de los modelos de análisis y, en última instancia, contribuir a la comprensión y prevención del cáncer.
The primary objective of this research was to identify a unified method for analyzing data related to breast and prostate cancer mortality in Costa Rica, considering hierarchical, spatial, and ecological aspects. This was based on a hierarchical binomial linear model, whose effectiveness was evaluated through simulations. Initially, an empirical study was conducted to understand the relationship between variables and to construct a pesticide exposure index. The results showed that models using conglomerates as the geographic unit (combination of districts within provinces) performed better compared to the original district-level grouping. Several simulation scenarios were generated to identify the optimal model, varying both the geographic unit and the regression coefficient associated with pesticide exposure. The simulation demonstrated that hierarchical and spatial models outperformed the simple model in terms of accuracy and fit. Although the hierarchical model was slightly superior according to the marginal log-likelihood in the different simulation scenarios evaluated, the hierarchical spatial models were preferred due to their ability to control for bias in error estimation due to data correlation. The best model selection was based on goodness-of-fit measures such as marginal loglikelihood, deviance information criterion, and Watanabe-Akaike information criterion, where clustering districts within provinces through cluster analysis showed the best fit. Spatial models using the queen contiguity criterion were identified as the most suitable for data analysis. Additionally, the importance of controlling pesticide use is emphasized, given its possible relationship with cancer. Institutions responsible for pesticide registration are urged to implement stricter control measures to improve the quality of analysis models and ultimately contribute to the understanding and prevention of cancer.
The primary objective of this research was to identify a unified method for analyzing data related to breast and prostate cancer mortality in Costa Rica, considering hierarchical, spatial, and ecological aspects. This was based on a hierarchical binomial linear model, whose effectiveness was evaluated through simulations. Initially, an empirical study was conducted to understand the relationship between variables and to construct a pesticide exposure index. The results showed that models using conglomerates as the geographic unit (combination of districts within provinces) performed better compared to the original district-level grouping. Several simulation scenarios were generated to identify the optimal model, varying both the geographic unit and the regression coefficient associated with pesticide exposure. The simulation demonstrated that hierarchical and spatial models outperformed the simple model in terms of accuracy and fit. Although the hierarchical model was slightly superior according to the marginal log-likelihood in the different simulation scenarios evaluated, the hierarchical spatial models were preferred due to their ability to control for bias in error estimation due to data correlation. The best model selection was based on goodness-of-fit measures such as marginal loglikelihood, deviance information criterion, and Watanabe-Akaike information criterion, where clustering districts within provinces through cluster analysis showed the best fit. Spatial models using the queen contiguity criterion were identified as the most suitable for data analysis. Additionally, the importance of controlling pesticide use is emphasized, given its possible relationship with cancer. Institutions responsible for pesticide registration are urged to implement stricter control measures to improve the quality of analysis models and ultimately contribute to the understanding and prevention of cancer.
Descripción
Palabras clave
ANÁLISIS ESPACIAL, CÁNCER, DIAGNÓSTICO, ESTADÍSTICA, PLAGUICIDAS, REGRESIÓN LOGÍSTICA, AUTOCORRELACIÓN ESPACIAL, COSTA RICA