Spatio-temporal modelling of incidence in COPD emergency admissions in an area of Northwest England from 2012 to 2018
This paper is a part of a single PhD “Geostatistical Methods for Modelling Spatially Aggregated Data”
Chronic Obstructive Pulmonary Disease (COPD) is one of the leading causes of mortality worldwide with an estimated three million deaths in 2015, corresponding to 5% of all deaths globally.
Its exacerbation is a major contributor to the number of emergency admission and hospitalization in the UK.
COPD is the second most common cause (after a heart attack) of admission to a medical ward in the UK - ie. it’s a huge cost burden and there is a belief that many cases could be prevented, hence the interest in predictions.
In this study, we pursue two objectives:
-
To assess the relative contribution of socio-economic and environmental variables for forecasting COPD emergency admissions
-
To develop a reliable early warning system that triggers an alarm whenever COPD emergency admissions signal the likely exceedance of predefined incidence thresholds
We developed a predictive model using a class of generalised linear mixed model. We selected the best predictors using the root mean square error (RMSE). We developed an early warning system based on exceedance probabilities.
The resulting predictors from our model selection are: minimum temperature; PM10; income deprivation; proportion of males; and proportion of the population aged above 75 years.
We found that, overall, the selected predictor variables explain about 22% per cent of the variability in the residual random effects. Among these variables, income deprivation attained the largest relative variance reduction of about 14%.
Our results demonstrate how to develop a predictive model as well as an early warning system for COPD emergency admission.
Our model has the potential to predict correctly in most areas with high sensitivity and specificity. The early warning system can potentially help to: identify and notify areas of high incidence of COPD emergency admission; and inform resource allocation for the healthcare system.
Keywords: COPD; emergency admission; early warning system; variogram; geostatistics; generalised linear mixed model; prediction model.