Project Abstract
Poor environmental conditions such as air pollution, and unsafe water and sanitation have been ranked among
the top risk factors for disability-adjusted years (DALYs) in children. The highest number of deaths per capita
attributable to environmental exposures have been observed in Sub-Saharan Africa (SSA) with the highest
disease burden noted among children. The overall goal of the proposed research is to harness data science
applications to establish the spatial variability in the impact of ambient PM2.5 exposure on children’s health in
SSA and further identify the explanatory and moderating factors. The overall goal of the project would be
achieved through the following specific aims: (1) Establish the spatial variability in the impact of ambient PM2.5
exposure on children’s health in SSA, and explore the effect modifying role of neighbourhood greenness and
nutrition, (2) Estimate ambient PM2.5 exposures at multi-temporal scales by integrating land use regression
(LUR) models, high-resolution ground monitoring data, and mobile monitoring data in Uganda and Ghana, and
(3) Identify area - (regional, district) and household-level factors that explain the spatial variability in ambient
PM2.5 – child health relationship and establish the temporal changes in these exposure risk profiles. The
proposed research seeks to create new knowledge and provide evidence on the potential of data science for
addressing children’s environmental health problems in SSA in alignment with the DSI-Africa program.
For Aim 1, we will leverage data science tools to combine geospatial PM2.5 exposures estimated using
satellite remote sensing with data on child undernutrition, acute respiratory infections, and neonatal and infant
deaths assembled from several waves of Demographic and Health Survey (DHS) and Multiple Indicator Cluster
Survey (MICS) data spanning several decades. We will use a spatial random coefficient model set in a Bayesian
framework to model the spatially varying relationship between ambient PM2.5 and the child health outcomes of
interest controlling for individual- and area-level confounders. For Aim 2, we would apply machine learning
techniques to develop a land use regression (LUR) model for Kampala and Accra leveraging mobile and fixed
monitoring data and compare the models between the two cities under the following data conditions; (1) using
only consistent data available in both cities and (2) using city-specific data to derive locally optimized models.
We will in addition evaluate transferability of the models from one city to another, and also, identify the most
important temporal and spatial predictors in both cities. For Aim 3, we will use Bayesian Profile Regression (BPR)
and leveraging the same datasets in Aim 1 to identify profile clusters that characterize high PM2.5 exposures
and determine which exposure profile clusters is associated with increase prevalence of adverse child health
outcomes. We would also explore the temporal changes in exposure profiles in the study countries.
The findings of the proposed resaerch should help trigger investment in air pollution control as well as
policy action for addressing area and household poverty to help improve child health and survival in SSA.