Title: Assessment of seasonal and climatic effects on the incidence and species composition of malaria by u
1Assessment of seasonal and climatic effects on
the incidence and species composition of malaria
by using GIS methods
- Ali-Akbar Haghdoost
- Neal Alexander (supervisor)
2Main objectives
- Assessment of the feasibility of an early warning
system based on ground climate and remote sensing
data - Assessment of the interaction between Plasmodium
spp from different points of view meta-analysis,
modelling, and extended analysis of a large
epidemiological dataset
3Climate effects on malaria
- The rate at which mosquitoes develop into adults
- Frequency of blood feeding
- Adult mosquito survival
- The incubation time of parasites in the mosquito
4Other considerations related to climate
- Deforestation
- Migration and urbanisation
- Changing human behaviour
- Natural disaster and conflict
5GIS and malaria
- Sipe (2003) reviewed the GIS and malaria
literature and divided the publications into the
five categories outlined below - Mapping malaria incidence/prevalence
- Mapping the relationships between malaria
incidence/prevalence and other potential related
variables - Using innovative methods of collecting data such
as remote sensing (e.g., GIS) - Modelling malaria risks
- General commentary and reviews of GIS used in
malaria control and research
6Modelling of malaria (1)
- Modelling of the abundance of vectors
- Modelling of the frequency of malaria
cases/infections
7Research setting (1)
8Research setting (2)
9Research setting (3) Kahnooj District
- Arid and semiarid
- Around 230,000 population in 800 villages and 5
cities - Area 32,000km2, less than 8 of area is used for
agriculture purposes
10Research setting (4) Kahnooj
11Research setting (5) Malaria In Iran
Annual number of malaria cases dropped from
around 100,000 to 15,000 between 1985 and 2002
More than 80 of cases are infected by P.vivax in
recent years
12Research setting(6) Malaria In Kahnooj
13Research setting (7) Health System
- Rural health centres
- Trained health workers
- Microscopists
- GPs
- Malaria Surveillance system
- Active follow-up of cases up to one year,
febrile people and their families - Passive case finding in all rural and urban
health centres free of charge - Private sector does not have access to malaria
drugs, it refers all cases to public sector - Reporting system weekly report to the district
centre - Supervision An external quality control scheme
is in place -
14Research setting (8) Treatment Of Malaria
- GPs Prescribe medicine
- P.falciparum chloroquine (3 days) primaquine
(with the second dose of chloroquine) - P.vivax chloroquine (3 days) primaquine
(weekly does for eight weeks, or daily dose for
two weeks) - Health works supervise that patients take drugs
completely, also take follow-up slides
15Objective
- Assessment of the feasibility of an early warning
system based on ground climate and remote sensing
data
16Data Collection (1)
- Surveillance malaria data between 1994 and 2002
- Age
- Sex
- Village
- Date of taking blood slides
- Plasmodium species
17Data Collection (2)
- The ground climate data (1975-2003) from the
synoptic centre in Kahnooj City - Daily temperature
- Relative humidity
- Rainfall
-
18Data Collection (3)
- GIS maps and RS data
- Electronic maps of Kahnooj contain the borders,
roads, villages and cities. The map scale was
150,000 in Arcview format - Landsat data with 30x30m spatial resolution in
January 2001, contained NDVI - NOAA-AVHRR data with 8x8km spatial resolution and
10 day temporal resolution from 1990 to 2001,
contained NDVI and LST - DEM images with 1x1km resolution (National
Imagery and Mapping Agency of United State of
America, http//geoengine.nima.mil/)
19Statistical methods (1)
- The risk of disease was estimated per village per
dekad (10 days) - Using mean-median smoothing method the temporal
variations were explored - Poisson method was used to model the risk of
disease - Fractional polynomial method was used to maximise
the accuracy of models - The time trend was model by using parametric
method (sine and cos)
20Statistical methods (2)
- Models predicted the risk of malaria in three
distinct spatial levels district, sub-sub-
district (SSD) and village - Using sensitivity analysis the best gap between
the predictors and malaria risk was estimated - The data were allocated into modelling (75) and
checking parts (25) - Using forward method the significant variables
were entered in the model. The significance of
variables were assessed by likelihood ratio test
and pseudo-R2
21Statistical methods (3)
- Using sensitivity analysis the best buffer zone
around each village was defined - The number of under and over-estimations and
percentages in the final model were computed - The feasibility of models were assessed by
comparing the over and under-estimations of
models with their corresponding values based on
the extrapolation from the previous month
22Results (1)
malaria risk factors
23Results (2)
Pearson correlation coefficients between the
annual risk of malaria and meteorological
variables in Kahnooj 1887-2001
24Results (3)
Temporal variations of malaria over a year the
observed numbers classified by species, based on
8-year data
25Results (4)
The seasonality and time trend of malaria
classified by species
26Results (5)
The fitted values of models based on seasonality,
time trend and meteorological variables
27Results (6)
Autocorrelations and partial autocorrelations
between the residuals of models, which estimated
risks, based on climate, seasonality and time
trend
28Results (7)
29Results (8)
30Results (9)
31Results (10)
The pseudo R2 between malaria risks and the
average NDVI around villages in 2001
1 The average NDVI around each village was
computed in circles with 15m up to 6km
radiuses 2 Fractional polynomial, degree two 3
Powers (1,2) 4 powers (-2,-0.5) 5 powers
(-2,-0.5))
32Results (11)
The observed and predicted risk maps of malaria
in 2001 in Kahnooj, the predicted maps were
computed based on NDVI around villages (in 5km
radius)
33Results (12)
The observed and predicted risk maps of malaria
in 1994-2001 in Kahnooj, the predicted maps were
computed based on the mean of altitude three
kilometres around villages by using fractional
polynomial models
Malaria was rare in villages with less than 450
or more than 1400 meter altitude. The maximum
risks were observed in villages with 700 to 900
meters altitude.
34Results (13)
The pseudo R2 of Poisson models classified by the
species based on village, SSD or whole district
data
35Results (14)
Over and under-predictions of models based on
seasonality, time trend and ground and remote
sensing data
District
SSD
Village
36Results (15)
Species-specific ROCs, they assess the
relationship between sensitivity and specificity
of the full models (with NDVI and LST) in
predicting local transmissions in all data
37Results (16)
Comparing the fitted and observed risk maps of
local transmission, the fitted values were
computed based on seasonality, time trend,
history of disease, NDVI and LST
38Summary of main findings (1)
- Ground climate data explained around 80 of P.
vivax and 75 of P. falciparum variations one
month ahead - Comparing to the extrapolation of data from
previous month, ground climate data improve the
accuracies around 10 but remote sensing data
does not improve - The ground climate data are freely available in
the filed therefore, it was concluded that the
models based on ground climate data are feasible.
39Summary of main findings (2)
- 4. Ground climate data improved predictions
around 10 one month ahead in district level - 5. NDVI and LST (with 8x8km resolution) did not
improve the prediction - 6. Elevation (with 1x1km resolution) improved
predictions around 15 - 7. NDVI (with 30x30m resolution) did not improve
the predictions
40Summary of main findings (3)
- 8. Elevation (with 1x1km resolution) improved
predictions around 15 - 9. NDVI (with 30x30m resolution) did not improve
the predictions - 10. P. falciparum and P. vivax models had
different parameters. - 11. The accuracy of temporal P. vivax variations
was less than that in P. falciparum
41conclusion
- Ground climate data (which are available free of
charge) improved the model accuracies around 10
and it seems that early warning system based on
these models is feasible
42Time for your comments
- Thanks for you kind attention