Towards malaria risk prediction in Afghanistan using remote sensing.

BACKGROUND: Malaria is a significant public health concern in Afghanistan. Currently, approximately 60% of the population, or nearly 14 million people, live in a malaria-endemic area. Afghanistan's diverse landscape and terrain contributes to the heterogeneous malaria prevalence across the country. Understanding the role of environmental variables on malaria transmission can further the effort for malaria control programme. METHODS: Provincial malaria epidemiological data (2004-2007) collected by the health posts in 23 provinces were used in conjunction with space-borne observations from NASA satellites. Specifically, the environmental variables, including precipitation, temperature and vegetation index measured by the Tropical Rainfall Measuring Mission and the Moderate Resolution Imaging Spectoradiometer, were used. Regression techniques were employed to model malaria cases as a function of environmental predictors. The resulting model was used for predicting malaria risks in Afghanistan. The entire time series except the last 6 months is used for training, and the last 6-month data is used for prediction and validation. RESULTS: Vegetation index, in general, is the strongest predictor, reflecting the fact that irrigation is the main factor that promotes malaria transmission in Afghanistan. Surface temperature is the second strongest predictor. Precipitation is not shown as a significant predictor, as it may not directly lead to higher larval population. Autoregressiveness of the malaria epidemiological data is apparent from the analysis. The malaria time series are modelled well, with provincial average R2 of 0.845. Although the R2 for prediction has larger variation, the total 6-month cases prediction is only 8.9% higher than the actual cases. CONCLUSIONS: The provincial monthly malaria cases can be modelled and predicted using satellite-measured environmental parameters with reasonable accuracy. The Third Strategic Approach of the WHO EMRO Malaria Control and Elimination Plan is aimed to develop a cost-effective surveillance system that includes forecasting, early warning and detection. The predictive and early warning capabilities shown in this paper support this strategy.

Modeling and predicting seasonal influenza transmission in warm regions using climatological parameters.

BACKGROUND: Influenza transmission is often associated with climatic factors. As the epidemic pattern varies geographically, the roles of climatic factors may not be unique. Previous in vivo studies revealed the direct effect of winter-like humidity on air-borne influenza transmission that dominates in regions with temperate climate, while influenza in the tropics is more effectively transmitted through direct contact. METHODOLOGY/PRINCIPAL FINDINGS: Using time series model, we analyzed the role of climatic factors on the epidemiology of influenza transmission in two regions characterized by warm climate: Hong Kong (China) and Maricopa County (Arizona, USA). These two regions have comparable temperature but distinctly different rainfall. Specifically we employed Autoregressive Integrated Moving Average (ARIMA) model along with climatic parameters as measured from ground stations and NASA satellites. Our studies showed that including the climatic variables as input series result in models with better performance than the univariate model where the influenza cases depend only on its past values and error signal. The best model for Hong Kong influenza was obtained when Land Surface Temperature (LST), rainfall and relative humidity were included as input series. Meanwhile for Maricopa County we found that including either maximum atmospheric pressure or mean air temperature gave the most improvement in the model performances. CONCLUSIONS/SIGNIFICANCE: Our results showed that including the environmental variables generally increases the prediction capability. Therefore, for countries without advanced influenza surveillance systems, environmental variables can be used for estimating influenza transmission at present and in the near future.

Meteorological, environmental remote sensing and neural network analysis of the epidemiology of malaria transmission in Thailand.

In many malarious regions malaria transmission roughly coincides with rainy seasons, which provide for more abundant larval habitats. In addition to precipitation, other meteorological and environmental factors may also influence malaria transmission. These factors can be remotely sensed using earth observing environmental satellites and estimated with seasonal climate forecasts. The use of remote sensing usage as an early warning tool for malaria epidemics have been broadly studied in recent years, especially for Africa, where the majority of the world's malaria occurs. Although the Greater Mekong Subregion (GMS), which includes Thailand and the surrounding countries, is an epicenter of multidrug resistant falciparum malaria, the meteorological and environmental factors affecting malaria transmissions in the GMS have not been examined in detail. In this study, the parasitological data used consisted of the monthly malaria epidemiology data at the provincial level compiled by the Thai Ministry of Public Health. Precipitation, temperature, relative humidity, and vegetation index obtained from both climate time series and satellite measurements were used as independent variables to model malaria. We used neural network methods, an artificial-intelligence technique, to model the dependency of malaria transmission on these variables. The average training accuracy of the neural network analysis for three provinces (Kanchanaburi, Mae Hong Son, and Tak) which are among the provinces most endemic for malaria, is 72.8% and the average testing accuracy is 62.9% based on the 1994-1999 data. A more complex neural network architecture resulted in higher training accuracy but also lower testing accuracy. Taking into account of the uncertainty regarding reported malaria cases, we divided the malaria cases into bands (classes) to compute training accuracy. Using the same neural network architecture on the 19 most endemic provinces for years 1994 to 2000, the mean training accuracy weighted by provincial malaria cases was 73%. Prediction of malaria cases for 2001 using neural networks trained for 1994-2000 gave a weighted accuracy of 53%. Because there was a significant decrease (31%) in the number of malaria cases in the 19 provinces from 2000 to 2001, the networks overestimated malaria transmissions. The decrease in transmission was not due to climatic or environmental changes. Thailand is a country with long borders. Migrant populations from the neighboring countries enlarge the human malaria reservoir because these populations have more limited access to health care. This issue also confounds the complexity of modeling malaria based on meteorological and environmental variables alone. In spite of the relatively low resolution of the data and the impact of migrant populations, we have uncovered a reasonably clear dependency of malaria on meteorological and environmental remote sensing variables. When other contextual determinants do not vary significantly, using neural network analysis along with remote sensing variables to predict malaria endemicity should be feasible.