Abordagem bayesiana para tratamento de dados faltantes com aplicação em um modelo logístico

Abstract

Missing data often comes up in practical applications and may cause many problems. The impact of missing data on modeling and statistical inferences is eminently important, especially in the face of subjects with missing data who have response patterns that differ greatly from those with complete data. Inadequate treatment or non-treatment of missing data may also affect the overall results of the analysis. There are several approaches of addressing the missing information problem. In this work,methodologies for missing data treatment in predictive models through an application of the problem are discussed. For this, the logistic regression technique is used to develop a predictive tool for the risk of hemorrhagic transformation in patients with ischemic stroke in a public hospital in Fortaleza, Brazil, in which, among their covariates, some of them have a representative amount of missing data. The main objective of this study is to apply different techniques of missing data treatment for each variable according to its nature and to adjust a predictive model, and then compare such approaches with a more complete database obtained at another point of this research.