Statistical modelling of cardiovascular disease patients using Bayesian approaches

Abstract

This study focuses on statistical modelling on cardiovascular disease (CVD) patients in Malaysia. A secondary dataset from the National Cardiovascular Disease Database-Acute Coronary Syndrome (NCVD-ACS) registry for the years 2006 to 2013 is utilised. Studies have shown that CVD affects males and females differently. Thus, a gender-specific analysis with regard to the risk factors and mortality among ST-Elevation Myocardial Infarction (STEMI) patients is needed. Initially, this study performed the standard multivariate logistic analysis where the aims are to identify risk factors associated with mortality for each gender and to compare differences, if any, among STEMI patients. The results showed that gender differences existed among STEMI patients. Even though females share the same risk factors as males, there are risk factors that relate only to females which may have increased their tendency to develop and increase the risk of mortality of CVD patients. An important contribution of this analysis is that it gives an understanding of possible gender-based differences in baseline characteristics, risk factors, treatments and outcomes which will help cardiac care specialists in improving current management of patients with CVD. Next, Bayesian analysis is proposed to develop a prognostic model of the STEMI patients. Bayesian Markov Chain Monte Carlo (MCMC) simulation approach is applied. Beside that, comparisons of the parameter estimates from the proposed Bayesian and frequentist models are made. The results showed that the proposed Bayesian modelling can deal correctly with the probabilities and provides parameter estimates of the posterior distribution which have natural clinical interpretations. In doing so, several programming codes for the Bayesian model development and convergence diagnostics in the Just Another Gibbs Sampler (JAGS) software in R interface are developed. In the final part of this study, a graphical probabilistic model framework defined using a Bayesian Network (BN) is proposed to identify and interpret the dependence structure between the predictors and health outcomes of STEMI patients. In doing so, the two learning processes are involved in obtaining the BN model from the data namely the structural learning and parameter learning. From the structural learning, 25 and 20 arcs were considered significant for males’ and females’ BN respectively. A few variables namely, Killip class, renal disease and age group were classified as key predictors as they were the most influential variables directly associated with the outcome of patients’ status. Moreover, conditional probabilities for each feature were obtained. The novelty of this study is that it provides an indication on the strength of each arc in the network by exploiting the bootstrap resampling method in the structural learning. A graphical model is developed where the relationships in a diagrammatical form is capable to be displayed and the cause-effect relationships can be illustrated. An important implication of this model is that it identifies dependencies based on the different features of variables. It can also include expert knowledge to improve predictability for data driven research when information or resources regarding the variables are limited.