A simulation study of disaggregation regression for spatial disease mapping

Abstract

Disaggregation regression has become an important tool in spatial disease mapping for making fine‐scale predictions of disease risk from aggregated response data. By including high resolution covariate information and modeling the data generating process on a fine scale, it is hoped that these models can accurately learn the relationships between covariates and response at a fine spatial scale. However, validating these high resolution predictions can be a challenge, as often there is no data observed at this spatial scale. In this study, disaggregation regression was performed on simulated data in various settings and the resulting fine‐scale predictions are compared to the simulated ground truth. Performance was investigated with varying numbers of data points, sizes of aggregated areas and levels of model misspecification. The effectiveness of cross validation on the aggregate level as a measure of fine‐scale predictive performance was also investigated. Predictive performance improved as the number of observations increased and as the size of the aggregated areas decreased. When the model was well‐specified, fine‐scale predictions were accurate even with small numbers of observations and large aggregated areas. Under model misspecification predictive performance was significantly worse for large aggregated areas but remained high when response data was aggregated over smaller regions. Cross‐validation correlation on the aggregate level was a moderately good predictor of fine‐scale predictive performance. While these simulations are unlikely to capture the nuances of real‐life response data, this study gives insight into the effectiveness of disaggregation regression in different contexts.

A Downscaling Approach to Compare COVID-19 Count Data from Databases Aggregated at Different Spatial Scales

“…Assuming a well‐specified model, the downscaling model is likely to have provided an accurate representation of the uncertainty (Arambepola et al., 2022), which has allowed us to quantify deviations of the counts from the disaggregated databases with regard to the estimated mean and 95% credible intervals of the predicted counts from the downscaling model. To prevent the risk of overfitting, we built a relatively parsimonious model by selecting a few theoretically‐relevant covariates and constraining them to be linearly associated with COVID‐19 counts to keep the model as simple as possible.…”

“…A recent study that investigates the predictive performance of downscaling models exploring different contexts (various point data, aggregated areas sizes, and types of model misspecifications) suggests that predictive performance is likely to improve with a high number of data points and small polygon areas. If these conditions are not satisfied, predictions should remain accurate enough if the model is well‐specified (Arambepola et al., 2022). With a total of 33 polygons used to fit a relatively parsimonious model, we are confident that the predicted mean and uncertainty (95% credible intervals) should be accurate enough to provide reliable estimates of COVID‐19 counts at the district level.…”

“…We use the R package disaggregation which implements a Bayesian approach to carry out spatial disaggregation modelling, also called downscaling (Nandi et al., 2020). Similar approaches have been recently used to estimate fine‐scale (grid‐cell) risk of disease based on spatially aggregated (polygon‐level) data (Arambepola et al., 2022; Diggle et al., 2013; Li et al., 2012; Sturrock et al., 2014; Weiss et al., 2019; Wilson & Wakefield, 2020). We apply a downscaling method to estimate cases of COVID‐19 and their uncertainty within fine grid cells (about 5 km spatial resolution) in China using JHU data aggregated at the province level (Dong et al., 2020).…”

PriorVAE: encoding spatial priors with variational autoencoders for small-area estimation

“…This assumption is unrealistic for many settings as heterogeneity might be present within each B i , as long as the size of the area is non-negligible. Areal data can be viewed as an aggregation of point data and a series of approaches began to emerge recently to address this issue [ 42 , 43 ]. Hence, it is reasonable to use a data augmentation step to sample from the posterior distribution of the exact locations, and then to aggregate results.…”

An integrated species distribution modelling framework for heterogeneous biodiversity data

“…There are many methodological advances in terms of how different data can be integrated into models. For example, in a public health problem Arambepola et al have developed methods to combine polygon and point estimates via disaggregation regressions so as to downscale critical health related indicators in the absence of finer resolved information (Arambepola et al, 2022).…”