A carregar...
Publicação
Modelação de resultados em saúde usando dados de inquéritos, ajustando para o desenho amostral
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.25 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Os Inquéritos Demográficos e de Saúde, com representatividade a nível nacional, têm
vindo a ser cada vez mais utilizados por investigadores da área das ciências da saúde
especialmente nos países de África Subsariana (SSA). De entre as principais aplicações
destacam-se os estudos de prevalência e fatores associados, os estudos de modelação
espacial e mapas de risco, nomeadamente nas denominadas doenças da pobreza (HIV,
Malária, Tuberculose, outras Doenças Tropicais Negligenciadas). Embora sejam dados
secundários, a elevada cobertura e representatividade (nomeadamente por sexo, classe
etária, regional e provincial) tem tornado estes inquéritos muito populares, sobretudo na
ausência de outros estudos de base populacional. Assim, tem-se assistido nesta última
década a um aumento substancial de publicações em revistas indexadas que utilizam
estes dados nas mais diversas áreas da saúde.
No entanto, estes inquéritos geralmente utilizam delineamentos amostrais complexos,
onde as unidades amostrais são selecionadas com diferentes probabilidades, através de
múltiplas etapas; desta forma é necessário que as análises estatísticas reflitam este processo
de amostragem, não só atribuindo a cada indivíduo o correspondente peso, mas também
através da correta estimação das variâncias dos estimadores obtidos, sob pena de se
sobrestimar ou subestimar os erros padrão associados aos estimadores empregues. Assim
sendo, torna-se imprescindível o ajuste dos métodos estatísticos convencionais para obter
estimativas mais precisas. Só desta forma se poderá fornecer evidências de apoio à decisão
confiáveis, nomeadamente para a orientação de políticas de saúde pública que visam
prevenir e combater determinadas doenças e/ou atingir os objetivos do desenvolvimento
sustentável.
Usando dados de Inquéritos Demográficos e de Saúde de Moçambique e do Malawi,
esta tese modelou diversos resultados em saúde tendo em conta o desenho completo da
pesquisa, em particular a estrutura multinível dos dados, a amostragem não proporcional
(exemplo: os pesos) e a informação geográfica.
Mais especificamente foi ajustado a nível micro e macro (i.e., a nível do indivíduo
e a nível agregado) um modelo para determinar os fatores de risco associados ao uso
de redes mosquiteiras em mulheres em idade reprodutiva em Moçambique, como uma forma de prevenir a Malária. Foram usados Modelos Lineares Generalizados e Modelos
Lineares Generalizados Mistos. Adicionalmente, estimou-se a distribuição geográfica da
desnutrição infantil no Malawi ao nível distrital, e da febre e diarreia infantil ao nível
provincial em Moçambique. Por fim, foi feita a modelação conjunta a da prevalência
de duas doenças (febre e a diarreia em Moçambique) que se sabem ser interligadas,
recorrendo ao Modelo de Análise de Dados Espaciais Conjuntos.
Em geral, as estimativas pontuais de média, proporção e os coeficientes de regressão
foram semelhantes comparando-se amostragem complexa e amostragem aleatória simples.
Os erro padrão ficaram subestimados ao usar métodos convencionais não ajustando para
o plano amostral. Isto faz com que os coeficientes estimados para modelos de regressão
sejam considerados significativos quando na verdade não o são. Adicionalmente, os
métodos ajustados ao peso foram os mais adequados. Embora o padrão espacial estimado
tenha sido semelhante, esta consideração dos pesos amostrais produziu estimativas de
risco de doença que tiveram um nível de confiança maior, tanto na análise univariada
quanto na análise bivariada. O uso dos métodos que considerem os pesos amostrais no
mapeamento de doenças para estimar a distribuição espacial de riscos de doenças com
base em dados complexos de pesquisas de saúde forneceu estimativas mais precisas, e
também evidências de apoio confiáveis para orientar as políticas de saúde pública na
focalização de recursos nas áreas de maior necessidade.
Demographic and Health Surveys, with national representativeness, have been increasingly used by health science researchers, especially in sub-Saharan African (SSA) countries. Among the main applications, stands out the prevalence studies and associated factors, spatial modeling studies and risk maps, particularly in the so-called poverty diseases (HIV, Malaria, Tuberculosis, other Neglected Tropical Diseases). Although secondary data are secondary data, high coverage and representativeness (mainly by gender, age group, regional and provincial) has made these surveys very popular, especially in the absence of other population-based studies. Thus, there has been a substantial increase in publications in indexed journals that use these data in various areas of health. However, these surveys generally use complex sample designs, where sample units are selected with different probabilities, through multiple steps; thus, it is necessary that the statistical analyses reflect this sampling process, not only attributing to each individual the corresponding weight, but also through the correct estimation of the variances of the estimators obtained, under penalty of overestimating or underestimating the standard errors associated with the estimators used. Therefore, it is essential to adjust conventional statistical methods to obtain more accurate estimates. Only in this way can reliable evidence of decision support be provided, in particular for the guidance of public health policies aimed at preventing and combating certain diseases and/or achieving the objectives of sustainable development. Using data from Demographic and Health Surveys of Mozambique and Malawi, this thesis modeled several health outcomes taking into account the complete design of the research, in particular the multilevel structure of the data, non-proportional sampling (e.g., weights) and geographic information. More specifically, a modelwas adjusted at micro and macro level (i.e., at the individual level and at the aggregate level) to determine the risk factors associated with the use of mosquito nets in women of reproductive age in Mozambique as a way to prevent Malaria. Generalized Linear Models and Mixed Generalized Linear Models were used. In addition, the geographical distribution of child malnutrition in Malawi at the district level and of fever and child diarrhea at the provincial level in Mozambique was estimated. Finally, the joint modeling of the prevalence of two diseases (fever and diarrhea in Mozambique, which are known to be interconnected) was carried out, using joint model of spatial data analysis. In general, the point estimates of mean, proportion and regression coefficients were similar comparing complex sampling and simple random sampling. The standard errors were underestimated when using conventional methods not adjusting for the sampling plan. This makes the estimated coefficients for regression models considered significant, when in fact they are not. Additionally, the weight-adjusted methods were the most appropriate. Although the estimated spatial pattern was similar, the use of sample weights produced estimates of disease risk that had a higher level of confidence, both in the univariate and bivariate analysis. The use of methods that consider sample weights in disease mapping to estimate the spatial distribution of disease risks based on complex data from health research provided more accurate and reliable estimates that can be used to guide public health policies to focus resources on areas of the greatest need.
Demographic and Health Surveys, with national representativeness, have been increasingly used by health science researchers, especially in sub-Saharan African (SSA) countries. Among the main applications, stands out the prevalence studies and associated factors, spatial modeling studies and risk maps, particularly in the so-called poverty diseases (HIV, Malaria, Tuberculosis, other Neglected Tropical Diseases). Although secondary data are secondary data, high coverage and representativeness (mainly by gender, age group, regional and provincial) has made these surveys very popular, especially in the absence of other population-based studies. Thus, there has been a substantial increase in publications in indexed journals that use these data in various areas of health. However, these surveys generally use complex sample designs, where sample units are selected with different probabilities, through multiple steps; thus, it is necessary that the statistical analyses reflect this sampling process, not only attributing to each individual the corresponding weight, but also through the correct estimation of the variances of the estimators obtained, under penalty of overestimating or underestimating the standard errors associated with the estimators used. Therefore, it is essential to adjust conventional statistical methods to obtain more accurate estimates. Only in this way can reliable evidence of decision support be provided, in particular for the guidance of public health policies aimed at preventing and combating certain diseases and/or achieving the objectives of sustainable development. Using data from Demographic and Health Surveys of Mozambique and Malawi, this thesis modeled several health outcomes taking into account the complete design of the research, in particular the multilevel structure of the data, non-proportional sampling (e.g., weights) and geographic information. More specifically, a modelwas adjusted at micro and macro level (i.e., at the individual level and at the aggregate level) to determine the risk factors associated with the use of mosquito nets in women of reproductive age in Mozambique as a way to prevent Malaria. Generalized Linear Models and Mixed Generalized Linear Models were used. In addition, the geographical distribution of child malnutrition in Malawi at the district level and of fever and child diarrhea at the provincial level in Mozambique was estimated. Finally, the joint modeling of the prevalence of two diseases (fever and diarrhea in Mozambique, which are known to be interconnected) was carried out, using joint model of spatial data analysis. In general, the point estimates of mean, proportion and regression coefficients were similar comparing complex sampling and simple random sampling. The standard errors were underestimated when using conventional methods not adjusting for the sampling plan. This makes the estimated coefficients for regression models considered significant, when in fact they are not. Additionally, the weight-adjusted methods were the most appropriate. Although the estimated spatial pattern was similar, the use of sample weights produced estimates of disease risk that had a higher level of confidence, both in the univariate and bivariate analysis. The use of methods that consider sample weights in disease mapping to estimate the spatial distribution of disease risks based on complex data from health research provided more accurate and reliable estimates that can be used to guide public health policies to focus resources on areas of the greatest need.
Descrição
Palavras-chave
Modelos Lineares Generalizados Amostragem Complexa Modelos Multinível Mapas de Risco Modelos Espaciais Malária