Two-stage Bayesian model to evaluate the effect of air pollution on chronic respiratory diseases using drug prescriptions

Abstract

Exposure to high levels of air pollutant concentration is known to be associated with respiratory problems which can translate into higher morbidity and mortality rates. The link between air pollution and population health has mainly been assessed considering air quality and hospitalization or mortality data. However this approach limits the analysis to individuals characterized by severe conditions. In this paper we evaluate the link between air pollution and respiratory diseases using general practice drug prescriptions for chronic respiratory diseases, which allow to draw conclusions based on the general population.We propose a two-stage statistical approach: in the first stage we specify a space-time model to estimate the monthly NO 2 concentration integrating several data sources characterized by different spatio-temporal resolution; in the second stage we link the concentration to the β 2 -agonists prescribed monthly by general practices in England and we model the prescription rates through a small area approach.

Quantifying the impact of current and future concentrations of air pollutants on respiratory disease risk in England

“…Therefore the uncertainty in its value should be accounted for when estimating its health effects. One approach to achieving this was proposed by [45], who created an exposure distribution for each areal unit and time period, and averaged the estimated pollution-health effects over this exposure distribution. However, a number of approaches are possible for accounting for this exposure uncertainty, and a rigorous comparison of the different approaches is much needed.…”

Multivariate space‐time modelling of multiple air pollutants and their health effects accounting for exposure uncertainty

“…Here,̂k t is the residual variation in pollutant 2 not accounted for by pollutant 1 in area k and time t. However, in line with the previous section, we extend (8) to allow for uncertainty in̂k t using a similar measurement error approach. Specifically, given our h samples of each pollutant at each area and time period, we fit model (7) for each set of samples, yielding h sets of residuals {̂i kt }. These are then used in the following measurement error model:̂i…”

“…To this end, 2 variants of the disease model proposed in Section 3 are compared in this study. The first is the baseline model ignoring measurement error given by (5) and augmented by (7) and (8). Hereafter, this model is referred to as BM.…”

“…To generate each simulated data set initially, model BM is fitted to the real data with the spatial maximum aggregation metric. Here, (7) was used to estimate the residual variation in PM 10 after NO 2 adjustment. This yielded realistic values of the model parameters which are then fixed for all simulated data sets.…”

“…Such long-term effects can be estimated by cohort studies such as Laden et al, 4 but they are expensive and time consuming to implement due to the long-term follow up required for the cohort. Therefore, a small-area spatio-temporal ecological-level study design can also be used, which uses routinely available data, and examples include Elliott et al, 5 Lee et al, 6 and Blangiardo et al 7 The disease data for this study design are counts of the total disease burden from the populations living in nonoverlapping areal units for consecutive time periods, and Poisson log-linear models accounting for spatio-temporal autocorrelation are typically used for the analysis. The pollution data available to characterise exposure comprise point-level measurements from a network of monitors and output from atmospheric dispersion models, and both of these data types have been used to estimate pollution concentrations in existing epidemiological health studies.…”