El impacto de COVID-19 en la salud mediante un análisis de herramientas de Estadística y Machine Learning.
Loading...
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
Esta tesis desarrolla una estrategia analítica progresiva para estudiar cambios observados durante la pandemia de COVID-19 en indicadores relacionados con deterioro cognitivo y demencia. Se integraron cuatro niveles: vigilancia poblacional, predicción temporal, simulación estocástica de escenarios y trayectorias longitudinales hospitalarias. En el Capítulo II se analizaron 11.444 observaciones agregadas de cuatro indicadores del CDC. Los cambios fueron heterogéneos; Q30 mostró la señal temporal más consistente, aunque pequeña y no monotónica. La validación 2015–2018 hacia períodos posteriores mostró transportabilidad limitada y ningún algoritmo fue universalmente superior. En el Capítulo III se aplicó un modelo de Markov de siete estados, parametrizado mediante síntesis estructurada de evidencia, a una cohorte sintética de 10.000 personas durante diez años y 100 réplicas. Bajo los supuestos especificados, la disrupción prolongada proyectó mayor carga de enfermedad avanzada y mortalidad, mientras una intervención focalizada atenuó las diferencias. En el Capítulo IV se estudiaron 2.168 pacientes y 2.629 episodios del sistema público de la Región del Biobío. Los estados de utilización hospitalaria mostraron persistencia y heterogeneidad entre los períodos pre-COVID, COVID y post-COVID. En conjunto, la tesis aporta una arquitectura trazable que distingue descripción, asociación, predicción, proyección y simulación. Los resultados corresponden a cambios observados, asociaciones y consecuencias potenciales condicionadas por los modelos; no identifican un efecto causal único de la pandemia ni de políticas específicas.
This thesis develops a progressive analytical strategy to study changes observed during the COVID-19 pandemic in indicators related to cognitive decline and dementia. Four complementary levels were integrated: population surveillance, temporal prediction, stochastic scenario simulation, and longitudinal hospital trajectories. Chapter II analyzed 11,444 aggregated observations for four CDC indicators. Changes were heterogeneous; Q30 showed the most consistent temporal signal, although its magnitude was small and its trajectory was non-monotonic. Temporal validation from 2015–2018 to later periods showed limited transportability, and no algorithm was universally superior. Chapter III applied a seven-state Markov model, parameterized through a structured synthesis of evidence, to a synthetic cohort of 10,000 people over ten years and 100 Monte Carlo replications. Under the specified assumptions, prolonged disruption projected a greater burden of advanced disease and mortality, whereas a targeted intervention attenuated these differences. Chapter IV examined 2,168 patients and 2,629 episodes from the public health system of the Biobío Region. Hospital-utilization states showed persistence and heterogeneity across the pre-COVID, COVID, and post-COVID periods. Overall, the thesis contributes a traceable analytical architecture that distinguishes description, association, prediction, projection, and simulation. The findings represent observed changes, associations, and model-conditioned potential consequences; they do not identify a single causal effect of the pandemic or of specific policies.
This thesis develops a progressive analytical strategy to study changes observed during the COVID-19 pandemic in indicators related to cognitive decline and dementia. Four complementary levels were integrated: population surveillance, temporal prediction, stochastic scenario simulation, and longitudinal hospital trajectories. Chapter II analyzed 11,444 aggregated observations for four CDC indicators. Changes were heterogeneous; Q30 showed the most consistent temporal signal, although its magnitude was small and its trajectory was non-monotonic. Temporal validation from 2015–2018 to later periods showed limited transportability, and no algorithm was universally superior. Chapter III applied a seven-state Markov model, parameterized through a structured synthesis of evidence, to a synthetic cohort of 10,000 people over ten years and 100 Monte Carlo replications. Under the specified assumptions, prolonged disruption projected a greater burden of advanced disease and mortality, whereas a targeted intervention attenuated these differences. Chapter IV examined 2,168 patients and 2,629 episodes from the public health system of the Biobío Region. Hospital-utilization states showed persistence and heterogeneity across the pre-COVID, COVID, and post-COVID periods. Overall, the thesis contributes a traceable analytical architecture that distinguishes description, association, prediction, projection, and simulation. The findings represent observed changes, associations, and model-conditioned potential consequences; they do not identify a single causal effect of the pandemic or of specific policies.
Description
Tesis presentada para optar al grado de Doctor/a en Ingeniería Industrial.
Keywords
COVID-19 (Enfermedad), Estadísticas, Machine learning