Bayesian modelling for spatially misaligned health areal data: a multiple membership approach

Marco Gramatica,Peter Congdon,Silvia Liverani

DOI: https://doi.org/10.48550/arXiv.2004.05334

2020-12-05

Abstract:Diabetes prevalence is on the rise in the UK, and for public health strategy, estimation of relative disease risk and subsequent mapping is important. We consider an application to London data on diabetes prevalence and mortality. In order to improve the estimation of relative risks we analyse jointly prevalence and mortality data to ensure borrowing strength over the two outcomes. The available data involves two spatial frameworks, areas (middle level super output areas, MSOAs), and general practices (GPs) recruiting patients from several areas. This raises a spatial misalignment issue that we deal with by employing the multiple membership principle. Specifically we translate area spatial effects to explain GP practice prevalence according to proportions of GP populations resident in different areas. A sparse implementation in Stan of both the MCAR and GMCAR allows the comparison of these bivariate priors as well as exploring the different implications for the mapping patterns for both outcomes. The necessary causal precedence of diabetes prevalence over mortality allows a specific conditionality assumption in the GMCAR, not always present in the context of disease mapping.

Applications

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the joint estimation of relative risks in datasets where there is a spatial misalignment between diabetes incidence and mortality. Specifically, the paper focuses on data in the London area. The data on diabetes incidence is sourced from general practitioner (GP) disease registries, while the data on diabetes mortality comes from census units in Middle - layer Super Output Areas (MSOA). Since residents may register with different GPs in different areas, this leads to the problem of spatial misalignment of data, that is, it is not possible to directly attribute all GP patients to their specific MSOA, and vice versa.

To solve this problem, the authors adopted the Multiple Membership (MM) method. By transforming the regional spatial effect into an explanation of GP practice incidence, the spatial misalignment problem is dealt with according to the proportion of the GP population in different areas. In addition, the paper also introduced multivariate versions of the Conditional Autoregressive (CAR) model (MCAR and GMCAR) and implemented sparse versions of these models in RStan to compare the different impacts of these bivariate priors and explore the map patterns of the two outcomes (incidence and mortality).

In particular, since diabetes incidence necessarily precedes mortality in terms of causality, the paper embedded this causal information in the GMCAR model, that is, estimating the spatial random effect of mortality given the spatial structure of incidence. This method not only solves the problem of spatial misalignment but also takes into account the potential causal relationship between diabetes incidence and mortality, thereby improving the accuracy of estimating the relative risks of these two health outcomes.