Temporal variations in malaria risk in Africa

In sub-Saharan Africa, malaria is a major cause of morbidity and mortality especially
among children less than five years of age and pregnant women. Malaria situations are
very diverse because of many factors involved in malaria transmission and the great
variety of their local combinations. These include climatic, ecologic, social, economic
and cultural factors. A number of epidemiological approaches have been used to try and
reduce malaria situations to a manageable number of types and classes for efficient
planning and targeting of appropriate malaria control strategies. Modelling and mapping
of malaria have long been recognized as important means to developing empirical
knowledge of this kind. Recently, the availability of new data sets, innovative analytical
tools and statistical methods has resulted in the development of more comprehensive
malaria maps for east, west and central Africa. However, most risk maps that have been
produced so far do not take into account seasonal variation in malaria transmission.
Seasonality affects the dynamic relationship between vector mosquito densities,
inoculation rate, parasite prevalence and disease outcome. Quantitative description and
mapping of malaria seasonality is therefore important for modelling malaria transmission
dynamics and for timely spatial targeting of interventions.
This thesis is part of an on going effort within the MARA/ARMA (Mapping Malaria Risk
in Africa/Atlas du risqué de la Malaria en Afrique) collaboration towards the
development of improved malaria risk maps for Africa. The main objective is the
development of an empirical model of malaria seasonality by fitting classical and modern
statistical models to clinical and / or entomological indices where available. This work
also intended to identify important determinants of between-year and between-area
variation that may be useful for developing climate based seasonal forecasting models for
malaria epidemics.
Chapter 1 gives an overview of the transmission and epidemiology of malaria in Africa
and set the rational for this work. The initial focus of the analysis was on southern Africa,
until recently this was the only region with reasonably comprehensive clinical malaria
case data in the continent and therefore offered an ideal starting point. This region has a
long history of successful malaria vector control by indoor residual spraying (IRS) with
insecticides and this may have an impact on the level of malaria endemicity and
consequently what we are modelling. Chapter 2 therefore reviews the historical impact of
IRS in southern Africa. Chapters 3 evaluate the impact of the El Nino Southern
Oscillation (ENSO) phenomenon on annual malaria incidence in Southern Africa. This is
the main driver of inter-annual and seasonal variability in climate in most regions in
Africa, and is important because ENSO events alter seasonality in climate in a way that
influences malaria seasonality. Chapter 4 uses Zimbabwe to examine the spatio-temporal
role of climate on year to year variation of malaria incidence. This country has a
heterogeneity of climatic suitability for malaria transmission and reflects varying
epidemiological profiles that occur in Southern Africa. Chapter 5 uses Zimbabwe as an
example towards the development of an empirical model of malaria seasonality based on
clinical malaria case data. Chapter 6 assesses the potential for use of the entomological
inoculation rate (EIR) to describe malaria seasonality in Africa. Chapter 7 improves on
work done in chapter 6 by modelling and mapping seasonal transmission of malaria
transmission using an approximation based on discrete Fourier transformations which
remove noise in the original time series and allows for the description important / main
seasonal components in EIR in relation to those of meteorological covariates. The work
described in these chapters culminated in five scientific publications and one working
paper
Chapter 2 showed that Southern African countries that sustained the application of IRS
reduced the level of transmission from hyper- to meso-endemicity and from meso- to
hypo-endemicity. This means that in instances where pre-control malariometric indices
are not available one can not assume to be modelling baseline endemicity. Preferably,
where the data are available the ideal situation will be to develop pre- and post-control
models to evaluate changes in the malaria risk pattern over time.
Chapter 3 found that contrary to east Africa where ENSO events and in particular El
Nino has been linked to changes in climatic condition and increase in epidemic risk, in
Southern Africa, ENSO has the opposite effect during El Nino years, with heightened
incidence during La Nina years. However, the impact of ENSO also varies over time
within countries, depending on existing malaria control efforts and response capacity.
From this analysis it is clear that in order to lay an empirical basis for epidemic
forecasting models there is a need for spatial-temporal models that at the same time
consider both ENSO driven climate anomalies and non ENSO factors influencing
epidemic risk potential.
Chapter 4 confirmed that there is considerable inter-annual variation in the timing and
intensity of malaria incidence in Zimbabwe. The modelling approach adjusted for
unmeasured space-time varying risk factors and showed that while year to year variation
in malaria incidence is driven mainly by climate the resultant spatial risk pattern may to
large extent be influenced by other risk factors except during high and low risk years
following the occurrence of extremely wet and dry conditions, respectively. It is likely
therefore that only years characterized by extreme climatic conditions may be important
for delineating areas prone to climate driven epidemics, and for developing climate based
seasonal forecasting models for malaria epidemics.
Chapter 5 employed the Bayesian spatial statistical method to quantify the relative
amount of transmission in each month. This method smoothed for unobserved or
unmeasured residual variation in malaria case rates while adjusting for environmental
covariates enabling us to interpret the spatial pattern of malaria in seasonality. This work
also demonstrated the feasibility of using Markham’s seasonality index (previously used
for rainfall) to describe malaria seasonality. In this analysis the index was used to
summarize the spatial pattern of the modelled seasonal trend by displaying the
concentration of malaria case load during the peak season across, which is important for
malaria control.
Chapter 6 adopted Markham’s seasonality index to characterize seasonality in EIR in
relation to environment covariates. This work successfully identified rainfall seasonality
and minimum temperature as predictors of malaria seasonality across a number of sites in
Africa. However, model predictions were poor in areas characterized by two rainfall
peaks and irrigation activities. The seasonality concentration index performed better in
areas with a unimodal seasonal pattern, and this might have had an adverse effect in the
analysis in areas with a bimodal seasonal pattern. This highlighted the need for an
improved quantification of malaria seasonality to model the complex and varied seasonal
dynamics across the continent.
Chapter 7 used an approximation of the discrete Fourier transform to the model
relationship between seasonality in EIR and meteorological covariates. This was used to
predict the seasonal average as well as the magnitude and timing of the main seasonal
cycles. This allowed for the estimation of the overall degree and timing malaria
seasonality and the duration of transmission across sub-Saharan Africa. Model
predictions can be used to estimate the average seasonal pattern of malaria transmission
across the continent. This analysis presents the first step towards the development of
improved models of malaria seasonality, and as more data become available the models
can be further refined.
In conclusion the Bayesian analytical framework used in this study enhanced our ability
to evaluate the relationship between malaria and climatic / environmental factors, and
improved considerably the identification of important associations and covariates.
Climatic and associated environmental determinants of seasonal and between yearvariation
in malaria, including the impact of ENSO identified in this work, provide
valuable information for the development of climate based seasonal forecasting models
for malaria. Furthermore, an approximation of the discrete Fourier transformation of the
data enabled us for the first time to develop empirical models and maps of the seasonality
of transmission of malaria at a continental level. These are positive developments for the
malaria modelling, mapping and control community in general.