Comparison between suitable priors in Bayesian modelling of risk factor of Malaysian coronary artery disease among female patients

Abstract

Most adults at increased risk of coronary artery disease (CAD) have no symptoms or obvious signs especially among women. In this study, three types of Bayesian models, each with different prior distribution were considered to identify associated risk factors in CAD among female patients presenting with ST-Elevation Myocardial Infarction (STEMI) and to obtain feasible model to fit the data. Comparisons were made to find the best model. A total of 1248 STEMI female patients from the National Cardiovascular Disease Database-Acute Coronary Syndrome (NCVD-ACS) registry year 2006-2013 were analysed. Bayesian Markov Chain Monte Carlo (MCMC) simulation approach was applied in the univariate and multivariate analysis for the three models. Model performance was assessed through measures of discrimination and calibration. Bayesian model C which used both Beta and Dirichlet prior distributions was considered as the best model. The Bayesian model C consisted of six significant variables namely dyslipidaemia, myocardial infarction (MI), smoking, renal disease, Killip class and age group. The same set of variables that were observed to be significant in the Bayesian model C was also found to be significant in models A and B which used single prior distribution, respectively. Model C performed better than models A and B, with good discrimination and calibration. This study illustrated that posterior estimation was mainly affected by the available prior information. Model which has both Beta and Dirichlet prior distributions can deal correctly with the probabilities and improves the quality of the estimation.