Data analysis of a study of myocardial injuries
in newborns with congenital cardiopathies
using a Bayesian Poisson model with an excess
of zero-valued observations

Suleimy Cristina MAZIN [1]

Marcela Silva de OLIVEIRA[2]

Edson Zangiacomi MARTINEZ[3]

Jorge Alberto ACHCAR3

Simone Gusmão RAMOS2

§    ABSTRACT: Count data with many zeros are common in studies in the medical field. In this article, we present a Poisson model with excess of zeros for the analysis of data from a study conducted in the Department of Pathology, Faculty of Medicine of Ribeirão Preto (FMRP-USP), where there is a huge excess of counts equal to zero. This is a study of myocardial injury in newborns with congenital heart diseases undergoing complex surgery. To this end, we present an adaptation of the Bayesian model based on Poisson distribution with excess of zeros by Angers and Biswas (2003). Given that the medical study design predicted that ten data counts were obtained for each individual, the difference between the Bayesian model presented here and that proposed by Angers and Biswas (2003) is the inclusion of multiple observations by individual and the presence of random effects that identify the dependency relationships between the observations of each newborn. The model proposed here showed to be satisfactory for the analysis of these data, allowing evidence, for example, that non-operated newborns with congenital heart diseases tend to have a greater number of events of intracellular edema. We used Bayesian methods based on Monte Carlo Markov Chain (MCMC) algorithm with the aid of the program Winbugs in order to obtain estimates of the parameters of interest.

§    KEYWORDS: Poisson distribution; excess zeros; Bayesian methods; medical statistics.

 



[1] Centro de Métodos Quantitativos (CEMEQ), Departamento de Medicina Social, Faculdade de Medicina de Ribeirão Preto, Universidade de São Paulo, CEP: 14049-900, Ribeirão Preto, SP, Brasil. E-mail: sumazin@pop.com.br

[2] Departamento de Patologia, Faculdade de Medicina de Ribeirão Preto, Universidade de São Paulo, CEP: 14049-900, Ribeirão Preto, SP, Brasil. E-mail: marcelaoliveira9@hotmail.com / sgramos@fmrp.usp.br

[3] Departamento de Medicina Social, Faculdade de Medicina de Ribeirão Preto, Universidade de São Paulo, CEP: 14049-900, Ribeirão Preto, SP, Brasil. E-mail: edson@fmrp.usp.br / achcar@fmrp.usp.br