edoc

Spatio-temporal modelling of under-five mortality and associations with malaria-anaemia comorbidity and health interventions in sub-Saharan Africa

Papaioannou, Isidoros. Spatio-temporal modelling of under-five mortality and associations with malaria-anaemia comorbidity and health interventions in sub-Saharan Africa. 2020, Doctoral Thesis, University of Basel, Associated Institution, Faculty of Science.

[img]
Preview
PDF
10Mb

Official URL: https://edoc.unibas.ch/78999/

Downloads: Statistics Overview

Abstract

A remarkable reduction of the total number of under-five deaths was achieved between 1990 and 2018 in the African setting, as pre-school mortality fell to 5.3 million deaths compared to 12.5 million in 1990. The bulk share of this reduction is attributed to the Millennium Development Goals (MDGs) era, during which time the under-five mortality rate has been declining with an annual rate of 3.8% across Africa. Despite these important achievements, the sub-Sahara African region did not meet the fourth target of the MDGs and still has an unacceptably high under-five mortality rate. Crucially, limiting the under-five mortality rate to a maximum of 25 deaths per 1,000 live births by 2030 lies at the heart of the Sustainable Development Goals (SDGs) and a recent report from the United Nations has warned that based on current trends, the African continent will not meet the SDG target for under-five mortality. Hence, providing useful insights from the associations between under-five mortality, the leading causes of disease and preventative or curative health interventions could make available valuable information to decision makers in order the African countries to achieve the SDGs on pre-school mortality.
Malaria is a major contributor to under-five mortality in sub-Saharan Africa, accounting for 400,000 deaths, approximately 60% of which are in children below the age of five. At global scale, the disability-adjusted life-years for the malaria disease are 45 million. An important aspect of the disease is that infection by malaria parasites does not necessarily lead to mortality and it is rather conditions that follow infection or other comorbidities that produce severe forms of the disease with increased mortality risk. Apart from malaria, pneumonia and diarrhea account for the most frequent causes of pre-school deaths. An interesting feature of all these three leading causes of under-fives in Africa, i.e. pneumonia, diarrhea and malaria, is that they share febrile response as their main clinical manifestation. Against the leading causes of under-five mortality, preventative or curative health interventions have been widely adopted in Africa, with their spatial coverage being on a significant rise, particularly due to the so-called scaling-up of health interventions during the last five years of the MDGs. For instance, ownership of Insecticide-Treated nets against malaria rose from 50 to 80 percent between 2010 and 2015, while their utilization averted 663 million clinical malaria cases over the MDGs era. Yet, the coverage of health interventions and the subsequent reduction in under-five deaths has happened in an unequal way across sub- Saharan Africa, raising concerns about health inequities at sub-national level.
The overall aim of the present PhD thesis is to develop, implement and interpret Bayesian geostatistical models with spatially varying coefficients in order to analyze approximately one million, cross-sectional mortality related-data in Africa and associate under-five mortality with malaria and health interventions. The point-by-point objectives of our work are as follows:
1. To develop a novel indicator for quantifying malaria-related mortality for children under the age of five in sub-Saharan Africa, namely the malaria-anemia comorbidity prevalence indicator (chapter 2);
2. To identify health inequities experienced by sub-national populations due to the geographical variation in the association between curative or preventive health interventions and under-five mortality in sub-Saharan Africa (chapter 3);
3. To assess the contribution of the leading causes of under-five mortality in sub- Saharan Africa on febrile response by associating the prevalence of malaria parasitaemia, diarrhoea and ARI with fever. (chapter 4);
4. To estimate the association between health interventions and under-five mortality on changes in mortality risk between two time points across Africa (chapter 5);
5. To compare Bayesian variable selection methods for spatially varying coefficient models, given that these approaches are at the forefront of analyzing geolocated mortality data in Africa (chapter 6).
In chapter 2, we estimated the association of malaria parasitaemia, anemia, and malaria- anemia comorbidity with all-cause under-five mortality and evaluated the potential of malaria-anemia comorbidity prevalence to quantify malaria-related deaths in sub-Saharan Africa. Additionally, we estimated within-country variation of the association between comorbidity and under-5 mortality, using spatially varying coefficient models. We presented our results at high spatial resolution, including model-based risk maps of malaria, anemia, and malaria-anemia comorbidity.
In chapter 3, we modeled the geographical variation in the association between health interventions and all-cause, under-five mortality in order to identify health inequities experienced by sub-national populations within a given country. To achieve that, we developed Bayesian geostatistical Weibull survival models with spatially varying coefficients for the effect of health interventions on mortality. Our approach allowed us to calculate the number of statistically important associations between interventions and mortality at regional level and hence to assess if health equity of interventions exists across the regions of a given country.
In chapter 4, we assessed the contribution of the leading causes of under-five mortality in sub-Saharan Africa on febrile response by associating the prevalence of malaria parasitaemia, diarrhoea and ARI with fever. Our flexible Bayesian spatial modelling approach allowed evaluating the geographical distribution of disease-exposure effect on fever in space (Administrative level 1). We also calculated the Potential Attributable Fraction (PAF) in order to quantify the contribution of childhood diseases on fever.
In chapter 5, we developed a novel methodology to statistically model the effect of health interventions on the changes in under-five mortality risk between two DHS survey time-points for 21 countries in Africa. We used a Bayesian geostatistical Weibull survival modeling approach and implemented rigorous Bayesian variable selection procedures in order to identify the most suitable set of health interventions for subsequent model fit.
In chapter 6, we assessed the performance of stochastic search variable selection (SSVS) for the fixed effects of geostatistical models, we compared three different Bayesian variable selection (BVS) methods for conditionally autoregressive (CAR) structured spatially varying coefficient models and finally we assessed the sensitivity of SSVS for the fixed effects when is co-implemented with a BVS procedure. We conducted a simulation study and applied the methods to the Burundi DHS in order to assess the aforementioned selection procedures. The present PhD thesis has contributed to the scientific fields of Epidemiology and Statistics by committing to the spatio-temporal modelling of under-five mortality data in the African setting, using primarily routinely collected, cross-sectional, household-based survey data coming from the Demographic and Health surveys program. The key outcomes of the research conducted in this thesis are as follows:
1. Our work contributed to the development, proposal and validation of a novel indicator for quantifying malaria-mortality using survey data, i.e. the malaria-anemia comorbidity indicator. Our main conclusions were that malaria burden in sub-Saharan Africa is considerably underestimated when anemia in not taken into account and that the malaria-anemia comorbidity prevalence provides a useful measure of the malaria-related deaths;
2. We presented the first study to assess sub-national health inequities, across most countries in Africa, by employing a spatial statistical modelling approach and routinely collected survey data coming from the DHS and MIS. Our results demonstrated strong sub-national health inequities across various regions for 28 African countries;
3. Our estimates confirmed the strong contribution of diarrhoea and acute respiratory infection on febrile response and accounted only one out of five cases to malaria;
4. Our work concluded that the health interventions that are mostly associated with changes in all-cause, under-five mortality risk in sub-Saharan Africa were Bacillus Calmette–Guérin (BCG) immunization, vitamin A supplementation and deworming medication;
5. Our analysis showed that the SSVS method is able to accurately identify the statistically important predictors for the fixed effects of geostatistical models and that SSVS is not sensitive to co-implementation with a BVS procedure for CAR- structured, spatially varying coefficients. We also concluded that one of the three BVS methods for varying coefficients, namely the Global selection method, is able to identify true varying coefficients with 70% success rate.
Advisors:Vounatsou, Penelope
Committee Members:Stensgaard, Anna-Sofie
Faculties and Departments:09 Associated Institutions > Swiss Tropical and Public Health Institute (Swiss TPH) > Department of Epidemiology and Public Health (EPH) > Biostatistics > Bayesian Modelling and Analysis (Vounatsou)
UniBasel Contributors:Vounatsou, Penelope
Item Type:Thesis
Thesis Subtype:Doctoral Thesis
Thesis no:13763
Thesis status:Complete
Number of Pages:215
Language:English
Identification Number:
  • urn: urn:nbn:ch:bel-bau-diss137636
edoc DOI:
Last Modified:28 Jan 2021 05:30
Deposited On:27 Jan 2021 16:09

Repository Staff Only: item control page