Biology Reference
In-Depth Information
inference and a Markov chain Monte Carlo (MCMC) algorithm produces
uncertainty metrics around the final predictions as well as the model inputs
in the form of predictive posterior distributions (
Patil et al., 2011
). Areas with
the least uncertainty are those with a large number of recent surveys with
relatively homogenous results, whereas greater uncertainty would be found in
places with sparse or old surveys with ranges of different observed PRs.
To model a global endemicity surface of
P. vivax
malaria, an MBG frame-
work that had been successfully employed for falciparum malaria (
Hay
et al., 2009
;
Gething et al., 2011a
) was used following modifications made to
accommodate biological features unique to
P. vivax
. The modelling methods
incorporate surveys from a wide time period such that older surveys are
given less weight than recent ones. The environmental covariates included
were those that had an a priori expectation to affect malaria transmission
intensity. These were urban areas defined by the GRUMP urban extent
product (
Balk et al., 2006
;
CIESIN/IFPRI/WB/CIAT, 2007
), a long-term
average vegetation index product used as a proxy for available moisture for
vector reproduction and survival (
Hay et al., 2006
;
Scharlemann et al., 2008
),
and the temperature suitability index derived from the model described
above, which identifies areas suitable for transmission based on the require-
ments of vector survival and sporogony (
Gething et al., 2011b
).
PR data were standardised by age because of variation in infection rates
observed in different age groups. It is often observed that malaria preva-
lence rises rapidly in infancy before reaching a plateau in early childhood
and declining through adolescence and adulthood. This phenomenon was
modelled using a previously described framework (
Smith et al., 2007
) to
standardise for the prevalence variability among age groups. The model was
informed with finely age-stratified
Pv
PR surveys to represent vivax-specific
age profiles (
Mueller et al., 2009b
;
Lin et al., 2010
) and was used to convert
all the observed survey prevalence values to standardised age-independent
values for use in the MBG modelling. Predictions were made for all-age
prevalence estimates for individuals aged one to 99 years (
Pv
PR
1-99
). Chil-
dren aged <1 year were not included because of the potential confounding
effect of maternal antibodies, but all other ages were included. This deviated
from the method of using the 2- to 10-year cohort for falciparum malaria
(
Guerra et al., 2007
;
Gething et al., 2011a
), because all age ranges are typi-
cally sampled for
P. vivax
, which is found at relatively lower prevalence rates.
To determine the endemicity of
P. vivax
, it was also important to
incorporate Duffy negativity into the modelling framework because of
the refractory nature of the phenotype to the parasite. The map of Duffy