Criação de um modelo matemático baseado em múltiplas variáveis ambientais para estimar mortalidade em saúde pública

• Ano de defesa: 2022
• Autor(a) principal: Lopes, Anaísa Filmiano Andrade
• Tipo de documento: Tese
• Tipo de acesso: Acesso embargado
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Ciências da Saúde
• Palavras-chave em Português: Saúde Ambiental; Saúde Pública; Análise Fatorial Exploratória; Regressão Linear Múltipla; Análise Estatística Multivariada; Environmental health.; Public health; Exploratory Factor Analysis; Multiple Linear Regression; Multivariate Statistical Analysis; CNPQ::CIENCIAS DA SAUDE; Ciências médicas; Saúde ambiental; Saúde pública - Pesquisa
• Link de acesso: https://repositorio.ufu.br/handle/123456789/37195; http://doi.org/10.14393/ufu.te.2023.30

One of the challenges that public management has been facing is to structure surveillance systems aimed at changing environmental contexts that represent risk situations and critical outcomes for human health. That said, this research aimed to create a mathematical model based on multiple environmental variables capable of estimating mortality in public health.To this end, a survey was carried out, selection and organization of multiple variables was carried out based on the Driving Force/Pressure/Situation/Exposure/Effect (FPSEE) model recommended by the World Health Organization. From the choice of environmental variables, the following statistical methods of multivariate analysis were used: Exploratory Factor Analysis (EFA), in order to find its latent structure and marker variables that were, finally, used to estimate the best mortality predictor model in public health, using the Stepwise Multiple Linear Regression Analysis technique. All statistical analyzes were processed using the IBM-SPSS Statistics software, version 22.0. The original database consisted of 853 observations that refer to the municipalities of Minas Gerais, southeastern region, Brazil and the data were obtained from public virtual information systems for the year 2017. Based on the underlying theoretical foundations, 130 variables were initially selected for analysis, grouped into 14 groups. From the FPSEE model, it was identified that 19.23% of the variables were classified as Driving Force; 6.9% as Pressure; 14.6% as Status; 21.5% as Exposure and 37.7% as Health Effect. After reviewing the literature and verifying the theoretical and statistical assumptions of the AFE, 54 variables were excluded as a result of repeated information, nature of the scale, missing cases and xaz\Saxz\partial correlations, leaving 76 variables suitable for factor analysis. The Spearman correlation matrix (ρ) showed 54.73% of significant linear correlations (α < 0.05), a percentage that increases to 59.17% when considering significant correlations at the level α < 0.10. The factorability of the variables was confirmed by the Bartlett sphericity test (p-value < 0.001) and the Kaiser-Meyer-Olkin (KMO) sample adequacy measure with a result equal to 0.952. From the rotated factor loading matrix (varimax) and based on the convergent results of the Scree Plot tests and the percentage of explained variance, 5 factors were extracted that, together, explain 59.78% of the total variance of the data. The first factor was labeled as socioenvironmental; the second as social vulnerability; the third as air quality; the fourth as mortality and the fifth as agrilivestock. The marker variables were, respectively: number of deaths from cancer; percentage of people enrolled in the Single Registry without adequate water supply; NO2 concentration; homicide mortality rate and finally, planted forest cover and natural vegetation cover. The variable with the highest factor loading in each factor and the variable with the second highest factor loading in the fifth factor were selected for the estimation of the best predictor model of mortality through Stepwise Multiple Linear Regression. The best model mathematical found by the RLM method ( = 0.126, p-value < 0.001) was Y= 7.655 + (-0.289 X1) + (0.132 X2) + (-0.109 X3), in which the variation of the variable dependent (gross mortality rate) is predicted by environmental variables: X1= percentage of natural vegetation cover ( = -0.289; p-value = 0.000), X2= homicide rate ( = 0.132; p-value = 0.000) and X3= percentage of coverage by planted forest ( = -0.109; p-value = 0.001). Through the EFA, 5 factors were identified and from them 6 marker variables were obtained capable of representing the entire initial set of variables with the least loss of information. From the variables selected by the AFE, it was possible to obtain a predictor model of mortality and determine which environmental variables best explain the behavior of mortality in public health. By clarifying the interrelationships between environmental variables and public health, it is possible to support decision-making in public management and mitigation of critical outcomes in human health.