Bayesian binary regression model: an application to in-hospital death after AMI prediction

Abstract

A Bayesian binary regression model is developed to predict death of patients after acute myocardial infarction (AMI). Markov Chain Monte Carlo (MCMC) methods are used to make inference and to evaluate Bayesian binary regression models. A model building strategy based on Bayes factor is proposed and aspects of model validation are extensively discussed in the paper, including the posterior distribution for the c-index and the analysis of residuals. Risk assessment, based on variables easily available within minutes of the patients' arrival at the hospital, is very important to decide the course of the treatment. The identified model reveals itself strongly reliable and accurate, with a rate of correct classification of 88% and a concordance index of 83%.Keywords: mortality probability prediction; model selection in binary regression; model diagnostic and residual analysis. ResumoUm modelo bayesiano de regressão binária é desenvolvido para predizer óbito hospitalar em pacientes acometidos por infarto agudo do miocárdio. Métodos de Monte Carlo via Cadeias de Markov (MCMC) são usados para fazer inferência e validação. Uma estratégia para construção de modelos, baseada no uso do fator de Bayes, é proposta e aspectos de validação são extensivamente discutidos neste artigo, incluindo a distribuição a posteriori para o índice de concordância e análise de resíduos. A determinação de fatores de risco, baseados em variáveis disponíveis na chegada do paciente ao hospital, é muito importante para a tomada de decisão sobre o curso do tratamento. O modelo identificado se revela fortemente confiável e acurado, com uma taxa de classificação correta de 88% e um índice de concordância de 83%.Palavras-chave: predição de probabilidade de óbito; seleção de modelos em regressão binária; diagnósticos e análise de resíduos.

An Application of Bayesian Approach in Modeling Risk of Death in an Intensive Care Unit

“…As such, the Bayesian approach is underutilized in areas of prognostic modeling for general ICU mortality outcomes. Although there were several studies that applied Bayesian MCMC for prediction of in-hospital risk of death, their areas were limited to specific subgroups of patients with diseases such as trauma [ 11 ], cancer and AIDS [ 12 ], acute myocardial infarction [ 13 ] and malaria [ 14 ]. These studies were mostly focused on application of Bayesian MCMC approach in variable selection and model choice.…”

Bayesian regression model with application to a study of food insecurity in household level: a cross sectional study

“…This method considers the parameter of the model as random variables and data are considered as fixed, and the parameters have their prior distribution [ 19 ]. Bayesian estimation had better results than maximum likelihood estimation even under non-informative prior, especially for small samples on logistic regression model because it allows for probabilistic interpretations of the model coefficients [ 19 – 21 ]. The weakness of maximum likelihood estimation in small sample can be solved by Bayesian estimation as an alternative technique, and this estimation solves the challenge of assumption of classical approach since it is flexible [ 22 ].…”

Bayesian outlier analysis in binary regression

“…The variable selection procedures include the use of Bayes factors to discriminate among competitive models and measures of the predictive ability of the model, as presented in [17,18]. The model with the best predictive ability, from now on called M 0 , includes the variables age, sex, history of arterial hypertension, previous myocardial infarction, the habit of smoking, the Killip class on admission and the interactions age × hypertension, sex × hypertension and hypertension × previous infarction.…”

“…The motivation to explore the theoretical results presented in this paper is to develop a Bayesian binary regression model to predict death of patients after AMI [18]. In that application, the rate of correct classification was around 89%.…”