Fusing areal and point level survey data to map childhood vaccination coverage
64th ISI World Statistics Congress - Ottawa, Canada
Format: CPS Abstract
Keywords: "bayesian, "spatial
Session: CPS 21 - Survey statistics II
Monday 17 July 4 p.m. - 5:25 p.m. (Canada/Eastern)
High resolution maps of health and development indicators are mainly produced using modelling approaches utilizing point-referenced data which are typically sourced from household surveys. However, in vaccination coverage estimation, spatially misaligned multiple input data sets can sometimes be available to model a coverage indicator of interest. Here, we focus on the case where both point-level data on vaccination coverage from a survey and routine areal data on disease counts are available for analysis and propose a fusion model that combines information from both datasets to map an indicator of vaccination coverage at a high resolution. The proposed model is a combination of a conditional autoregressive model for the areal data and a Gaussian process model for the point level data, with Poisson and binomial likelihoods specified for both outcomes, respectively. The melding of both spatial scales in the model is accomplished using the components of the linear predictor. The model is fitted in a Bayesian framework using the INLA-SPDE approach and applied to mapping the coverage of measles vaccination in Nigeria using the 2018 Demographic and Health Survey (DHS) data and measles case counts. The predicted coverage maps reveal that combining information from both data sources leads to better identification of low coverage areas compared to coverage maps produced using only geolocated survey data as is usually the practice.