Análise de dados de sobrevivência espacialmente correlacionados

• Ano de defesa: 2014
• Autor(a) principal: Calisto Manuel Máquina
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Modelo de riscos proporcionais; Modelos de fragilidade; Dados de sobrevivência; Análise de sobrevivência (Biometria); Estatistica; Teoria bayesiana de decisão estatistica
• Link de acesso: http://hdl.handle.net/1843/BUBD-9GXFB8

In this work the analyses of two real data sets with spatial information available are presented. The first data set corresponds to an application of the medical field in which the objective is to study the life of patients with nasopharyngeal cancer in Connecticut, USA, and the second data set comprises a sample of individuals living in municipalities belonging to the metropolitan region of Belo Horizonte, who were victims of attempted murder. Considering a Bayesian approach and assuming that the failure times follow Weibull distribution, the gamma and lognormal frailty models, which do not take into account the spatial dependence structure present in the data, and the intrinsic CAR model, which accommodates such dependence structure, were adjusted for both data sets. In order to evaluate the effect of different prior specifications on the fitted models a sensitivity analysis was conducted. The best fitted models were then compared, and maps displaying the estimated frailties for each area were built. Based on such maps it was possible to observe the existence of a risk pattern of the occurrence of the event interest over the areas under study , indicating the presence of spatial dependence structure in both databases analyzed.