Assessing malaria attributed mortality in west and southern Africa

Malaria has persistently remained a serious health and socio-economic problem in developing nations particularly in Sub-Saharan Africa (SSA). There are approximately 500 million cases of malaria each year and close to one million deaths occurring mainly among children under five years. Developing countries spend a reasonable proportion of their gross domestic product (GDP) on malaria which in the end hinders their levels of development.
World Health Organizations (WHO) and partners through the Roll Back Malaria initiative (RBM) have targeted vector control, health promotion and case management (using rapid diagnostic tests and treatment with Artemisinin combination therapy) in order reduce malaria morbidity and mortality cases. Since 2002, funds for promoting malaria control activities have increased exponentially in SSA. Major donors include presidential malaria initiative (PMI) and Global fund to fight AIDS, tuberculosis and malaria (GFATM). Countries which have scaled up the recommended malaria control strategies such as insecticides-treat net (ITN) and treatment of confirmed cases have reported a decline in both morbidity and mortality especially among children. However, these statistics are based on health facilities data and yet in most developing countries many deaths occur at home and are never recorded due to inefficient vital registration systems. Monitoring the progress of such interventions requires reliable sources of data on both the transmission and infection outcome.
In malaria endemic areas, people acquire natural immunity during the early years of their life after getting exposed to repeated infections. This is observed from the reductions in the number of severe malaria-related morbidity and mortality cases especially in children >5 years. Due to the current undertakings that are aimed at reducing malaria exposure, there are concerns about shifting the disease burden to older children but the required to data to monitor this are not readily available in SSA. Low income countries have resorted to health and demographic surveillance systems (HDSS) to monitor routinely population changes and health outcomes within a defined geographical area.
In 2000, the INDEPTH, a network of HDSS integrated the Malaria Transmission Intensity and Mortality Burden Across Africa (MTIMBA) project into selected sites’ routine activities in order to assess the transmission-malaria mortality relationship taking into account the current interventions. Mortality data and other demographic characteristics were extracted from routinely collected HDSS databases. The entomological data were collected every fortnight from randomly sampled compounds over the 3 years MTIMBA period.
The MTIMBA project generated large geostatistical data that are correlated in space and time. Furthermore, the project captured longitudinal mosquito data that were characterized by many zeros especially during the dry periods. The zeros are due empty traps from a compound or when all the captured mosquitoes are not infectious. Appropriate data analysis therefore should apply models that account for spatial-temporal correlation and the excess zeros in order to avoid over or underestimation of parameters. Zero-inflated geostatistical models account for spatial-temporal correlation by introducing location-specific and time interval random effects which creates more parameters to estimate. Bayesian models implemented via Markov chain Monte Carlo simulation (MCMC) addresses fit of highly parameterized models.
This work applied zero-inflated Bayesian models to estimate malaria attributable mortality across all age-groups using large, correlated and sparse data collected from Navrongo and Manhiça HDSS between 2001 and 2004. The contributions of this thesis were (i) the description of the HDSS data characteristics and relevant methods for analysis; (ii) the spatially explicit estimates of malaria transmission intensity at monthly intervals; and (iii) the relationship between all-cause mortality and malaria transmission intensity across all age categories.
Chapter 2 described the characteristics of the MTIMBA data. These are large geostatistical, temporal, seasonal and zero-inflated data. The mortality and mosquito data were misaligned because they were captured at different compounds and time periods. Zero-inflated Bayesian spatio-temporal models are the state-of-art in handling such data. The rigorous statistical process was demonstrated by modelling sporozoite rate (SR) data from Manhiça HDSS. The analysis of the MTIMBA data was used as an avenue for building SSA capacity through course work, seminars and mentorship. Site-specific analyses are still on-going. However, the project generated data that is relevant for assessing within and between site malaria transmission heterogeneity.
The Navrongo malaria exposure surfaces described in chapter 3 were obtained from zero-inflated geostatistical models fitting separately the binomial SR data and negative binomial count data by mosquito species. All the models included space and time correlation in addition to the Climate, environmental and seasonality covariates. The entomological inoculation rate (EIR) estimates were derived as a product of predicted man biting rate and SR. Observed EIR in this district was >100 infective bites/person/year. Distance to water to bodies, day temperatures and vegetation were the main predictors of mosquito densities for the two species. The EIR maps clearly indicated that the temporal heterogeneity was stronger than the spatial variation in this area. The same situation was also observed from the analyses of the two MTIMBA sites of Rufiji (Tanzania) and Kisumu (Kenya).
Monthly malaria exposure surfaces (chapter 3) were linked to the nearest compounds where mortality was observed as described in chapter 4. Time to death data were split at monthly intervals in order to generate Bernoulli and binomial data that were modelled via logistic regression formulations. Spatio-temporal models were fitted to obtain age-specific mortality risk estimates. The model considered 2 covariates; natural logarithm transformed EIR estimates with their measurement errors and age. ITN variable was only included in neonates, post-neonates and child models. The analysis showed a positive log-linear relationship between all-cause mortality and malaria exposure in all the age groups but the association was only important among children (1-4 years) and people >= 60 years. ITN use showed a protective effect among all the under five children, confirming what was observed in Rufiji and Kisumu HDSS.
The methods used in estimating malaria exposure surfaces and mortality risks in chapters 3 and 4 were extended to Manhiça HDSS (Mozambique) data to describe the mortality-malaria transmission relationship for this area (chapter 5). The spatio-temporal age-specific models considered EIR estimates with their measurement errors (to account for the predictive uncertainty) and age as model covariates.
The distance to the nearest water bodies was the only important common predictor of An. funestus and An. gambiae mosquito densities. Malaria transmission intensity declined consistently in this area. The Model-based results indicated a positive log-linear relationship between all-cause mortality and malaria exposure across all age groups namely; the neonates (0-28 days), post-neonates (1-11months), children (1-4years), young people (5-14 years), adults (15- 59years) and old age (>=60 years).
This work contributes to further understand of malaria-mortality relationships. A positive association between mortality and malaria exposure among the under fives is consistent with what was reported from the MTIMBA sites of Rufiji and Kisumu. Completion of the remaining site-specific analyses followed by a meta-analysis will make a great contribution to malaria epidemiology. Further work however, should consider cohort analysis in order to ascertain whether malaria control interventions have caused a shift in the age of acquired immunity.