Spatio-temporal Distribution, The Construction And Evaluation Of Prediction Model On Malaria Epidemics In Hainan, China

ObjectiveMalaria is a parasitic disease that caused by protozoan parasites and transmitted by Anopheles. It seriously threatens human health and is a crucial public health problem in China. Hainan is one of the two endemic provinces of Plasmodium falciparum malaria in China, which ranked the top two provinces of malaria incidence during recent ten years. Our group explored the distribution of malaria and the association between its epidemic and enviroment in Hainan preliminarily. We also contructed the ARIMA model in Wanning City using data from 1995 to 2001 and predicted malaria epidemic in this area in 2002. However, there was a fluctuation of malaria incidence in Hainan after 2000, while little is known about malaria epidemics since 2000 except for the reported malaria incidence, which makes malaria control and prevention more difficult in Hainan. This study explored the spatio-temporal distribution patterns of Plasmodium falciparum and Plasmodium vivax malaria from time and space dimensions in Hainan. Then a prediction model was constructed and evaluated by climate factors using multivariable time series analysis. It is the further analysis based on our previous study. The aim of this study is to provide some scientific basic evidences for malaria epidemic prediction, public health resource allocation and malaria surveillance and control in Hainan.Methods1. The Plasmodium falciparum and Plasmodium vivax malaria incidences from 1995 to 2008 were calculated and plotted by year and by month to show annual and seasonal fluctuations in Hainan visually. The Cochran-Armitage trend test was employed to examine temporal trends in the annual incidence of Plasmodium falciparum and Plasmodium vivax malaria. The seasonal index was calculated and plotted from January to December to observe the seasonal fluctuations of these two types of malaria epidemic.2. Morlet wavelet was employed as basis functions for wavelet transform to detect the cyclical fluctuations of Plasmodium falciparum and Plasmodium vivax malaria epidemic. The local wavelet power spectrum (LWPS) was obtained and color coded with power increasing from blue to red. The global wavelet power spectrum (GWPS) is estimated by averaging the LWPS across time.3. The cumulative and annual malaria incidences of Plasmodium falciparum and Plasmodium vivax malaria in each county were calculated and mapped from 1995 to 2008 respectively. Each region was marked with different color on the county-level digital map.4. Temporal cluster analysis of Plasmodium falciparum and Plasmodium vivax malaria data from 1995 to 2008 were performed respectively to divide this period into several stages. Spatial cluster analysis for each stage was conducted to detect where malaria cases were clustered and its changes with time.5. Ljung-Box Q test was performed to explore the autocorrelation of monthly malaria incidence. Cross correlation coefficient and its standard error were calculated to detect the association between malaria incidence and climate factors. Co-linearity statistics including variance inflation factors and tolerances were calculated to examine the co-linearity between variables that added to the model for variable selection. The relationship between malaria incidence and climate factors was evaluated using scatter plots.6. The prediction model was constructed by climate factors and malaria cases of several months before. The stepwise least squares method was employed to fit the model.7. The adjusted R2, residuals of fitting and predicting were calculated. Ljung-Box Q test was conducted to verify whether the residual were white noise sequence.Results1. Both of the annual incidences of Plasmodium falciparum and Plasmodium vivax malaria showed a clear year-to-year variation from 1995 to 2008 in Hainan. A fluctuating but distinctly declining temporal trend of annual malaria incidence was identified (Cochran-Armitage trend test, Plasmodium falciparum malaria: Z = -54.62, P < 0.01; Plasmodium vivax malaria : Z = -43.12, P < 0.01 ).2. The period with the relatively high Plasmodium falciparum and Plasmodium vivax malaria incidence was from May to October. The cumulative monthly incidence of these two types of malaria reached to their peak in August, with 16.20 cases per 100,000 of Plasmodium falciparum malaria, and 48.78 cases per 100,000 of Plasmodium vivax malaria. Both of these two types of malaria existed clear seasonal fluctuation over time, which were weakened gradually.3. The epidemic peak occured once a year between 1995 and 2003 of Plasmodium falciparum malaria and between 1995 and 2005 of Plasmodium vivax malaria, which was then disappeared. There were a three year cyclical fluctuation of Plasmodium falciparum malaria and Plasmodium vivax malaria between 1998 and 2006.4. Plasmodium falciparum and Plasmodium vivax malaria were reported in all counties in Hainan during 1995 to 2008. The highest incidences of these two types of malaria were mainly distributed in the south counties of the Province. The highest incidence of Plasmodium falciparum malaria was 193.42 cases per 100,000 in Baoting in 1995, and that of Plasmodium vivax malaria was 415.82 cases per 100,000 in Baoting in 1998.5. The spatial cluster analysis of Plasmodium falciparum malaria data showed that seven counties in south-central Hainan constituted the most likely cluster during 1995¹998. Eight counties in south-central and southwest Hainan made up the most likely cluster and two counties in southeast Hainan composed the secondary one during 1999²008. The spatial cluster analysis of Plasmodium vivax malaria data showed that six counties in south-central Hainan constituted the most likely cluster during 1995²001. Five counties in south-central Hainan made up the most likely cluster and two counties in southeast Hainan composed the secondary one during 2002²004. Four counties in south-central Hainan made up the most likely cluster and two counties in southeast Hainan composed the secondary one during 2005²008.6. The prediction model of malaria epidemic in Hainan is N=0.0034×MT₁^3.3559+0.0269×MT₂^2.8131+0.8165×N₁ Where MT1, MT2, and N1 were the mean temperature of the previous month, the mean temperature of the previous two months and malaria cases of the previous month, respectively. The adjusted R² of fitting and predicting were 0.81 and 0.70. The fitting and predicting residuals were white noise sequence.Conclusions1. There were a distinctly declining temporal trend of annual Plasmodium falciparum and Plasmodium vivax malaria incidence. The epidemic peak in each year of Plasmodium falciparum malaria was disappeared after 2003 and that of Plasmodium vivax malaria was disappeared after 2005. There were a three year cyclical fluctuation of Plasmodium falciparum malaria and Plasmodium vivax malaria between 1998 and 2006.2. The area with the highest risk of Plasmodium falciparum malaria expanded gradually from central-south Hainan to southeastern and southwestern Hainan and the area with the highest risk of Plasmodium vivax malaria expanded gradually from central-south Hainan to southeastern Hainan.3. Good predictive ability of malaria epidemic can be achieved in Hainan using power function by monitoring the fluctuations of the mean temperature of the previous month, the mean temperature of the previous two months and malaria cases of the previous month.