Predicting Malaria Epidemics in Ethiopia Eskindir Loha, Hawassa University Torleif Markussen Lunde, UoB 5th Conference on Global Health and Vaccination Research: Environmental Change and Global Health Tromsø 6-8 June, 2010 Addis Ababa University School of Public Health Aklilu Lemma Institute of Pathobiology National Meteorological Agency of Ethiopia University of Bergen Centre for International Health Geophysical Institute Bjerknes Centre for Climate Research Arba Minch University
Motivation 1. Malaria is a major public health problem in Ethiopia 2. The spread of malaria is sensitive to weather 3. Weather is predictable, so is malaria predictable? 4. Malaria prediction could be useful in making efficient use of the limited resources for malaria control.
Malaria forecasting? Long-term epidemic forecasting Malaria epidemic early warning: based on surveying transmission risks related to abnormal rainfall or temperature Epidemic early detection
Overview System components Mosquito module Epidemiological module Hydrological module Meteorological module Malaria model Preliminary results Long-term forecasting Epidemic early detection
Mosquito module The main malaria vector in Ethiopia is Anopheles arabiensis. Early development (egg, larva, pupa) is dependent on water temperature Adult mortality is dependent on temperature and humidity Egg laying behavior is dependent on eg. the availability of small water pools
Mosquito module Collection of eggs, larva and pupa Typical breeding site for Anopheles arabiensis
Mosquito module - Monitoring mosquito densities in villages - Monitoring breeding sites (egg, larva, pupa) Species composition Survival from egg to pupa Resting behavior Feeding behavior
Epidemiological module Same locations as mosquito collection Find out the prevalence and incidence of malaria in the study areas Examine the relative importance of malaria as a cause of morbidity and mortality To assess the potential epidemiological risk reasons for malaria epidemics. To examine the health seeking behavior of the people on malaria
Epidemiological module Each household is followed for two years Chano - 1212 households (weekly basis) Butajira (survey every 3 months) - 200 households from semi-low-altitude - 250 from mid-altitude - 300 households at high-altitudes Evaluation of clinical and parasitological response to Artemether-lumefantrine (children)
Hydrological module
Meteorological module Two main projects: 1. Where does the water come from? 2. Dynamical downscaling - seasonal weather forecasts.
Where does the water come from?
Where does the water come from? Previous "truth" moisture from west Knowledge helpful in seasonal weather forecasts
Seasonal forecasts Limitations: 1. The predictability of where rainfall will fall down is low 2. Model suffers from this, and main goal is to reproduce whether a season is wetter or dryer than the normal (compared to stations) 3. Limited computational power and storage means clouds can not be fully resolved at the moment. 1979-2008, based on monthly sums. In the rainy season, a lot of rain one place does not mean a lot of rain in the neighboring station.
Seasonal forecasts Produce high space-time resolution data on: Temperature Rainfall Soil moisture Soil temperature Accumulated rainfall Relative humidity etc... Which can be used as input in malaria model Temperature
Malaria-Meteorology models Difficult to model climate-malaria relationship Mainly because non-climatic factors (drugs, socioeconomic conditions,...) can overrule the influence of climate over long time. Malaria-weather models make more sense since weather can be a driving factor when other factors are kept constant.
Dynamical models - hind-cast
Dynamical models - forecast Use predicted sea surface temp. to predict seasonal weather
Reproduction of rainfall Model seems to capture dry and wet years very well (averaged anomalies all stations). The example shows model data from 2000-2006, and corresponding station data.
Better than existing products? Does a better job in catching dry and wet years than the Tropical Rainfall Measuring Mission (TRMM)
Dynamical model - forced by model weather Next slide shows results from one dynamical model (we are currently working on many). The human component has been excluded, and only number of mosquitoes is modeled This means that biting behavior is not taken into account, and human immunity, bed net usage, new drugs etc. are not allowed to vary. The model is run for each location, but merged afterwards since the modeled weather not necessarily falls down the correct place. The model is evaluated every 6 hours.
Dynamical model - forced by model weather
Malaria-Meteorology models South Ethiopia
Malaria-Meteorology models Lessons o Complexity of malaria-meteorology link/models Different stat. or mathematical form Different variable and lag combination Varying data transformation forms
Malaria-Meteorology models Lessons o Biologic approach: less prediction power o Crudeness of models (malaria-meteorology link) questioned and local variations emphasized o Non-climatic factors: vector composition, land use, population movement, health services, immunity
Malaria-Meteorology models Objective Using area-specific Plasmodium falciparum malaria incidence and meteorological data, to unveil local variations of the malaria-meteorology link
Malaria-Meteorology models Method 35 locations (1998-2007) Total cases (microscopically confirmed PF)= 210 659 with in ave. 6.7 years, mean serial length 80 (range 51-118) months
Malaria-Meteorology models Rainfall (35 locations), temperature (17 locations) and RH (3 locations) The model: Time series (SPSS 17) o Transfer function (TF) or Univariate ARIMA
Locations with 1998-2007 Plasmodium falciparum incidence and meteorological data for which the models were developed Administrative regions of Ethiopia Geographic coordinates of the locations
Malaria-Meteorology models Result 5 models -dropped (diagnostic stat) so 30 Past Plasmodium falciparum malaria incidence o 17 locations (alone) and 4 locations (with meteorological predictors) o Seasonal ARIMA orders: altitude 1742 and above
Malaria-Meteorology models Meteorological data o Rainfall- 4 locations o Minimum temp.- 5 locations o Maximum o RH- temp.- 2 locations no significant effect
Malaria-Meteorology models Goodness of fit of models (R2 or stationary R2) o o o o o One model had negative value The rest ranged 16%-97%. Two third had above 50% Seasonal ARIMA above 60% (altitude 1742 and above) No significant correlation: R squared versus serial length or average monthly case numbers
Malaria-Meteorology models Model variations/similarities o o o o o o 7 locations only dependent on past incidence but no significant weather variable The rest included weather variable or took different data transformation forms. Some did not contain incidence at any AR or MA orders No apparent reiteration in line with altitude Two did not have any significant predictor 5 models failed due to Ljung-Box Q Models are different
Mean condition of 23 locations Locations Available data Model structure (ARIMA) 23 Incidence Rainfall (1,0,0)(0,1,0) Stationary Sig. variable Serial Ave. R2 and model length Incid./ description month 0.67 62 (Jul 02Aug 07) Incidence at AR lag 1 and first order seasonal differencing Rainfall at delay 4 and num. TF order of 0 and first order seasonal differencing 2579
Malaria-Meteorology models Conclusion Local predictions Malaria-Meteorology link varied from place to place, so averaging- not logical or precise Past Plasmodium falciparum malaria incidence is worth considering (outweighed the impact of weather variables) A need to include non-meteorological factors Regional predictions Could give a longer lead to prepare for epidemics with limitations of not knowing the exact location
EMaPS Supervisors:. Ahmed Ali, Teshome Gebre-Micheal, Meshesha Balkew, Semu Ayalew Moges, Asgeir Sorteberg, Wakgari Deressa (coordinator), and Bernt Lindtjørn (coordinator) PhDs: Abebe Animut describes the malaria mosquitoes in the Butajira area. Adugna Woyessa: Describe human malaria in a Butajira area in rural south central Ethiopia. Korecha Diriba: Studies the link between large-scale climate and weather with the local weather experienced in Ethiopia. Dereje Tesfahun: Considers how ground conditions influence surface water distribution. Ellen Viste: Looks at the moisture transport and other driving forces of Ethiopian rainfall. Fekadu Masebo: describes the malaria mosquitoes in the Chano area. http://emaps.uib.no/