Bayesian hierarchical modelling for two-dimensional blood pressure data

Many real-world phenomena are naturally bivariate. This includes blood pressure, which comprises systolic and diastolic levels. Here, we develop a Bayesian hierarchical model that estimates these values and their interactions simultaneously, using sparse data that vary substantially between groups and over time. A key element of the model is a two-dimensional second-order Intrinsic Gaussian Markov Random Field (IGMRF), which captures non-linear trends in the variables and their interactions. The model is fitted using Markov chain Monte Carlo methods, with a block Metropolis-Hastings algorithm providing efficient updates. Performance is demonstrated using simulated and real data. Furthermore, IGMRFs can be used to induce conditional dependence in Bayesian hierarchical models. IGMRFs have both a precision matrix, which defines the neighbourhood structure of the model, and a precision, or scaling, parameter. Previous studies have shown the importance of selecting the prior of this scaling parameter appropriately for different types of IGMRF, as it can have a substantial impact on posterior results. The focus is on the two-dimensional case, where tuning of the parameter's prior is achieved by mapping it to the marginal standard deviation of a two-dimensional IGMRF. We compare the effects of scaling various classes of IGMRF, to the application of blood pressure data.