Adigun, Abbas Bolaji. Geostatistical models of malaria and associated morbidity among preschool-aged children in Nigeria. 2016, Doctoral Thesis, University of Basel, Faculty of Science.
|
PDF
2871Kb |
Official URL: http://edoc.unibas.ch/diss/DissB_12446
Downloads: Statistics Overview
Abstract
Malaria remains a threat to the lives of millions of children in tropical and subtropical countries. It is still a disease of public health significance, because of its role as a major cause of morbidity and mortality among the vulnerable group, specifically children under the age of five in the endemic countries. Although, substantial progress has been made in the control and prevention of the disease especially during the past 15 years due to multilateral commitment to malaria control, and this has led to reduction in the burden attributed to the disease. During the same period, financial resources for malaria prevention and control have been like up to twenty-fold increase, which led to widespread scale-up of coverage of the core malaria control interventions: insecticide-treated nets (ITNs), indoor residual spraying (IRS), and prompt treatment of clinical malaria cases with artemisinin-based combination therapy (ACT).
High resolution disease risk distribution is essential information in successful control activities, because of its versatility in cost effective planning, surveillance, and evaluation of such activities. Spatial statistical modelling provides rigorous inferential framework for high resolution disease risk mapping. It is a data-driven approach, which is used to build mathematical relationship between geo-referenced disease data and potential predictors (environmental and socio-demographic factors). Such model always includes the location specific random effect to explain the spatial correlation in the disease data that are due to common exposure in neighbouring locations. Geostatistical model are highly parameterized, nevertheless, a Bayesian geostatistical framework provides flexible and rigorous inferential methods for modelling such data. Computation tools such as simulation based Markov chain Monte Carlo (MCMC) or numerical approximation approach as integrated nested Laplace approach (INLA) are mostly engaged for such model fit.
Nigeria is one of the countries in sub-Sahara Africa with high prevalence of malaria and its related morbidity and mortality among children under the age of five years. Contemporary high resolution estimates of malaria prevalence needed for control activities are lacking. Also the precise nature of malaria transmission and all-cause mortality remains unclear. Furthermore, spatial analysis of the effect of malaria intervention on the risk of the disease at the national and sub-national level is not yet done. Moreover, anaemia prevalence in Nigeria is high; however, its relationship with malaria burden among children under the age of five is not fully understood, coupled with lack of high resolution estimates of the spatial distribution of the risk in the country.
This thesis aims to address these knowledge gaps by developing data driven Bayesian geostatistical models for analyzing spatially referenced data and also to provide tools for malaria and its related morbidity control programmes in the country. The analysis in this work is based on data from the contemporary nationwide survey which are malaria indicator survey (MIS) and demography and health survey (DHS). Roll back malaria initiative in its global effort of coordinating malaria control developed the MIS to collect malaria related burden data on children under the age of five, and it is always conducted during the high transmission season. MIS is standardized in terms of survey design, questionnaire and implementation time.
In chapter 2, we implemented a Bayesian geo-statistical model to analyze the first nationally representative malaria parasitaemia prevalence data in Nigeria to produce high resolution risk estimates of spatial distribution of malaria prevalence in the country, and also derived number of infected children at the sub-national level. Rigorous Bayesian variable selections were incorporated in the spatial models in order to select the best environmental predictors of malaria and its functional form. The approach identifies important risk factor to build Bayesian model of malaria risk in Nigeria. Also, various interventions coverage indicators were derived to assess their effect on malaria risk. The high resolution estimates show that malaria risk varies between 19.6% and 47.7% in Lagos and Osun state, respectively. However, household coverage indicators of intervention did not indicates association with malaria risk.
Chapter 3, present the assessment of the spatial effect of ITN use by children less than five years on the malaria parasitaemia prevalence at the first administrative, after adjusting for climatic and socio-demographic factors. Bayesian geostatistical model with spatial varying coefficient at the sub-national level was used to explore the malaria risk-intervention relationship. Smooth map of intervention effect was produced based on the parameter estimates of ITN use at the first administrative level.
In chapter 4, we employed a joint Bayesian geo-statistical Cox model with log constant baseline hazard and binomial geostatistical logistic regression models to relate mortality with malaria prevalence, and take into account spatial misalignment between DHS and MIS datasets, to evaluate the contribution of malaria prevalence to all-cause mortality among children less than five year of age. The mortality model was implemented separately for infant 0-6 months, 7-11 months, and older children. The model adjusted for socio-demographic factors known to be associated with risk of death among this vulnerable group. We also produced smooth map of residual variation not accounted for by the factors in our model.
Chapter 5 presents the geostatistical analysis of haemoglobin level/anaemia risk. The study assessed malaria burden on anaemia risk among the children after adjusting for helminthiasis and schistosomiasis, and socio-demographic factors. We make use of some of these factors as available at individual level, and also use the predicted prevalence of those that were not directly obtained with the haemoglobin data, which led to the implementation of Bayesian geostatistical models (Gaussian and logistic) with measurement error, to incorporate the uncertainty in the predicted estimates. The predictive models were used to obtain high resolution estimates of geographical distribution of anaemia risk/haemoglobin level concentration in the country. The population adjusted prevalence show that approximately every 7 out of 10 children under the age of five years are anaemic in the country.
The work in this thesis contributes improved Bayesian statistical methods for generating reliable estimate of disease burden (malaria parasitaemia prevalence, anaemia prevalence and number of infected children) at high spatial resolution. It also adds to the evidence of improve method of evaluating the effect of malaria interventions on disease prevalence. Furthermore, the generated model based risk maps constitute important information to national malaria control programme, because of its resourcefulness in right targeting of high risk area to achieve disease reduction, and eventually elimination. Finally, our work provides essential yardstick on which newer estimates could be compared as new data becomes available and control efforts continue.
High resolution disease risk distribution is essential information in successful control activities, because of its versatility in cost effective planning, surveillance, and evaluation of such activities. Spatial statistical modelling provides rigorous inferential framework for high resolution disease risk mapping. It is a data-driven approach, which is used to build mathematical relationship between geo-referenced disease data and potential predictors (environmental and socio-demographic factors). Such model always includes the location specific random effect to explain the spatial correlation in the disease data that are due to common exposure in neighbouring locations. Geostatistical model are highly parameterized, nevertheless, a Bayesian geostatistical framework provides flexible and rigorous inferential methods for modelling such data. Computation tools such as simulation based Markov chain Monte Carlo (MCMC) or numerical approximation approach as integrated nested Laplace approach (INLA) are mostly engaged for such model fit.
Nigeria is one of the countries in sub-Sahara Africa with high prevalence of malaria and its related morbidity and mortality among children under the age of five years. Contemporary high resolution estimates of malaria prevalence needed for control activities are lacking. Also the precise nature of malaria transmission and all-cause mortality remains unclear. Furthermore, spatial analysis of the effect of malaria intervention on the risk of the disease at the national and sub-national level is not yet done. Moreover, anaemia prevalence in Nigeria is high; however, its relationship with malaria burden among children under the age of five is not fully understood, coupled with lack of high resolution estimates of the spatial distribution of the risk in the country.
This thesis aims to address these knowledge gaps by developing data driven Bayesian geostatistical models for analyzing spatially referenced data and also to provide tools for malaria and its related morbidity control programmes in the country. The analysis in this work is based on data from the contemporary nationwide survey which are malaria indicator survey (MIS) and demography and health survey (DHS). Roll back malaria initiative in its global effort of coordinating malaria control developed the MIS to collect malaria related burden data on children under the age of five, and it is always conducted during the high transmission season. MIS is standardized in terms of survey design, questionnaire and implementation time.
In chapter 2, we implemented a Bayesian geo-statistical model to analyze the first nationally representative malaria parasitaemia prevalence data in Nigeria to produce high resolution risk estimates of spatial distribution of malaria prevalence in the country, and also derived number of infected children at the sub-national level. Rigorous Bayesian variable selections were incorporated in the spatial models in order to select the best environmental predictors of malaria and its functional form. The approach identifies important risk factor to build Bayesian model of malaria risk in Nigeria. Also, various interventions coverage indicators were derived to assess their effect on malaria risk. The high resolution estimates show that malaria risk varies between 19.6% and 47.7% in Lagos and Osun state, respectively. However, household coverage indicators of intervention did not indicates association with malaria risk.
Chapter 3, present the assessment of the spatial effect of ITN use by children less than five years on the malaria parasitaemia prevalence at the first administrative, after adjusting for climatic and socio-demographic factors. Bayesian geostatistical model with spatial varying coefficient at the sub-national level was used to explore the malaria risk-intervention relationship. Smooth map of intervention effect was produced based on the parameter estimates of ITN use at the first administrative level.
In chapter 4, we employed a joint Bayesian geo-statistical Cox model with log constant baseline hazard and binomial geostatistical logistic regression models to relate mortality with malaria prevalence, and take into account spatial misalignment between DHS and MIS datasets, to evaluate the contribution of malaria prevalence to all-cause mortality among children less than five year of age. The mortality model was implemented separately for infant 0-6 months, 7-11 months, and older children. The model adjusted for socio-demographic factors known to be associated with risk of death among this vulnerable group. We also produced smooth map of residual variation not accounted for by the factors in our model.
Chapter 5 presents the geostatistical analysis of haemoglobin level/anaemia risk. The study assessed malaria burden on anaemia risk among the children after adjusting for helminthiasis and schistosomiasis, and socio-demographic factors. We make use of some of these factors as available at individual level, and also use the predicted prevalence of those that were not directly obtained with the haemoglobin data, which led to the implementation of Bayesian geostatistical models (Gaussian and logistic) with measurement error, to incorporate the uncertainty in the predicted estimates. The predictive models were used to obtain high resolution estimates of geographical distribution of anaemia risk/haemoglobin level concentration in the country. The population adjusted prevalence show that approximately every 7 out of 10 children under the age of five years are anaemic in the country.
The work in this thesis contributes improved Bayesian statistical methods for generating reliable estimate of disease burden (malaria parasitaemia prevalence, anaemia prevalence and number of infected children) at high spatial resolution. It also adds to the evidence of improve method of evaluating the effect of malaria interventions on disease prevalence. Furthermore, the generated model based risk maps constitute important information to national malaria control programme, because of its resourcefulness in right targeting of high risk area to achieve disease reduction, and eventually elimination. Finally, our work provides essential yardstick on which newer estimates could be compared as new data becomes available and control efforts continue.
Advisors: | Utzinger, Jürg and Vounatsou, Penelope and Sogoba, Nafomon |
---|---|
Faculties and Departments: | 09 Associated Institutions > Swiss Tropical and Public Health Institute (Swiss TPH) > Former Units within Swiss TPH > Health Impact Assessment (Utzinger) |
UniBasel Contributors: | Vounatsou, Penelope |
Item Type: | Thesis |
Thesis Subtype: | Doctoral Thesis |
Thesis no: | 12446 |
Thesis status: | Complete |
Number of Pages: | 1 Online-Ressource (119 Seiten) |
Language: | English |
Identification Number: |
|
edoc DOI: | |
Last Modified: | 09 Feb 2020 05:31 |
Deposited On: | 01 Feb 2018 14:08 |
Repository Staff Only: item control page