25-11-2024
Table 1. Files and their descriptions within the CMHProgressMap_LMICs GitHub repository
| Name | Type | Description |
|---|---|---|
| covariates | Folder | This folder contains a demo of the format of the data extracted from geospatial covariates considered when modelling the health and development indicators. This file is required to run all R scripts in this repository. Examples of geospatial covariate datasets can be found from https://hub.worldpop.org/project/categories?id=14. |
| indicators | Folder | This folder contains a demo of subnational maternal and child health indicators to model. This file is required to run all R scripts in this repository. The indicators were extracted from the 2014 (DHS-7)[1] and 2022 (DHS-8) [2] for Kenya , which are publicly available after registration onto the Measure DHS website (www.dhsprogram.com). |
| raster | Folder | This folder contains the file structure for the geospatial covariates considered when modelling the health and development indicators. This file is required to run all R scripts in this repository. Examples of geospatial covariate datasets can be found from https://hub.worldpop.org/project/categories?id=14. |
| results | Folder | Folder to contain all prediction and uncertainty gridded datasets (raster files), county and subcounty aggregations and model information produced from the workflow file |
| shapes | Folder | This folder contains the shapefiles required to run all R scripts in this repository. These should be the administrative boundaries of the study area as polygons and the location of the clusters in the study area as points (lat/lon) These shapefiles can be obtained from the DHS program at www.dhsprogram.com. |
| raster_preparation_covariates.R | R | This file contains code for the creation of the covariates from the raster files |
| workflow | R | R script for modelling the subnational maternal and child health indicators. The files required to run this script are the covariates and indicator csv files and the files in the shp folder. |
| utilities | R | R script containing helper functions for the main workflow file |
This code was prepared for modelling indicators in Kenya however it has been presented in a general way to allow application to other contexts.
First covariates are then scaled and centred based on the associated rasters. The geospatial covariate selection is two-staged.In the first stage, we check for multicollinearity amongst the geospatial covariates.In the second stage, we employ the backward stepwise model selection method.
To check for multicollinearity, a Pearson correlation matrix for the geospatial covariates is created and any pairs with a Pearson correlation coefficient r>0.8 are flagged. The flagged covariates are then individually fitted in non-Bayesian binomial generalised linear models (GLMs). The Bayesian information criteria (BIC) of the models are then calculated. The covariate in the model with a lower BIC is retained while the covariate in the model with the greater BIC is omitted for the target indicator.
After checking for multicollinearity, a backward model selection algorithm is used to select the best (sub)set of geospatial covariates for the target indicator. The algorithm is as follows. The remaining geospatial covariates are fitted in a non-Bayesian binomial GLM and the BIC is calculated. A covariate is removed from the model and the BIC is recalculated. If the recalculated BIC is less than the previously calculated BIC, this subset of covariates is preferred. These steps are performed iteratively until the recalculated BIC is not less than the BIC calculated from the previous iteration. At this point, the best (sub)set of geospatial covariates have been attained and they will be used when constructing the Bayesian point-referenced spatial binomial GLM in INLA.
A Bayesian point-referenced spatial binomial GLM is then estimated, using R-INLA the model contains a random effect with Matern covariance based on a mesh constructed from the country boundary and the GPS points where the indicators are collected, random effects corresponding to the district and state in which the point is located are also included [4,5]. Fixed effects based on the covariates identified in the model selection outlined are also included. More information on this model can be found in https://www.nature.com/articles/s41597-023-01961-2 or for a wider introduction to INLA please see https://www.paulamoraga.com/book-geospatial/sec-inla.html.
Additional components must be constructed before fitting the model in INLA. First a mesh of the study domain is constructed with the shape file and coordinates within the target indicator file. Using this mesh object, a stochastic partial differential equation (SPDE) object is defined with functions in INLA where the priors of the spatial decay parameter and spatial variance parameter is defined. With the mesh object, INLA stack “A” matrices are created and stacked with the INLA stack functions. Finally, these components, along with the model are fitted into the INLA function.
Once the model is fitted 1000 posterior samples are taken from the INLA model. The script then reads in the raster files corresponding to the geospatial covariates of the model for the target indicators and compiles it as a prediction data frame. Finally, the predicted values are computed from the prediction data frame, INLA mesh objects, INLA posterior sample objects and the random effects corresponding to the district and state. This results in 1000 samples from the modelled distribution for each 1x1km grid square.
This matrix of samples is produced for the indicator both round 1 corresponding to the 2014 DHS survey [1] and round 2 corresponding to the 2022 DHS survey[2]. A matrix of samples for the change in the indicator is then produced for the change in the indicator between the two surveys.
These matrices of samples at the 1x1km level are then aggregated to both subcounty and county administrative boundaries using population weighted aggregation using population data is from Worldpop unconstrained population raster 1km resolution [6]. The mean, standard deviation 95% credible intervals are produced directly from the 1000 samples. In the case of the aggregation of the change in the indicator a further column is included corresponding to the probability that the true change is greater than 0. The aggregated ouputs are availible at https://doi.org/10.5258/SOTON/WP00791
The script then outputs raster tif files corresponding to the mean, median, standard deviation and 95% credible interval of the indicators at the 5x5km level. These high resolution outputs are availible at https://doi.org/10.5258/SOTON/WP00790
The authors acknowledge the support of the PMO Team at WorldPop. Moreover, the authors would like to thank the DHS Program staff for their input on the construction of some of the indicators. This work was approved by the ethics and research governance committee at the University of Southampton.
Dorey, P.,Chan, H.M.T., Tejedor-Garavito, N., Priyatikanto, R., Bonnie, A., Williams, E.M., Johnson, M., Tatem, A.J., Pezzulo, C. 2024. CMHProgressMap_LMICs: Child and Maternal Health Indicator Mapping from DHS Data in LMICs, version 1.1. WorldPop, University of Southampton. DOI: https://doi.org/10.5281/zenodo/14217826
-
2014-15 (DHS-7) Kenya DHS: Kenya National Bureau of Statistics, Ministry of Health/Kenya, National AIDS Control Council/Kenya, Kenya Medical Research Institute, National Council for Population and Development/Kenya, and ICF International. 2015. Kenya Demographic and Health Survey 2014 [DATASETS]. Rockville, MD, USA: Kenya National Bureau of Statistics, Ministry of Health/Kenya, National AIDS Control Council/Kenya, Kenya Medical Research Institute, National Council for Population and Development/Kenya, and ICF International.
-
2022 (DHS-8) Kenya DHS: KNBS and ICF. 2023. Kenya Demographic and Health Survey 2022: [DATASETS]. Nairobi, Kenya, and Rockville, Maryland, USA: KNBS and ICF. http://dhsprogram.com/pubs/pdf/FR339/FR339.pdf. 2017
-
The DHS Program Code Share Project, Code Library, DHS Program. DHS Program GitHub site. https://github.com/DHSProgram., in DHS Program GitHub site. 2022.
-
Rue, H., Martino, S. and Chopin, N., 2009. Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. Journal of the Royal Statistical Society Series B: Statistical Methodology, 71(2), pp.319-392.
-
Lindgren, F., Rue, H. and Lindström, J., 2011. An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach. Journal of the Royal Statistical Society Series B: Statistical Methodology, 73(4), pp.423-498.
-
WorldPop (www.worldpop.org - School of Geography and Environmental Science, University of Southampton; Department of Geography and Geosciences, University of Louisville; Departement de Geographie, Universite de Namur) and Center for International Earth Science Information Network (CIESIN), Columbia University (2018). Global High Resolution Population Denominators Project - Funded by The Bill and Melinda Gates Foundation (OPP1134076). https://dx.doi.org/10.5258/SOTON/WP00670