Background: Hainan is one of the provinces most severely affected by malaria epidemics in China. The distribution pattern and major determinant climate factors of malaria in this region have remained obscure, making it difficult to target countermeasures for malaria surveillance and control. This study detected the spatiotemporal distribution of malaria and explored the association between malaria epidemics and climate factors in Hainan.
Methods: The cumulative and annual malaria incidences of each county were calculated and mapped from 1995 to 2008 to show the spatial distribution of malaria in Hainan. The annual and monthly cumulative malaria incidences of the province between 1995 and 2008 were calculated and plotted to observe the annual and seasonal fluctuation. The Cochran-Armitage trend test was employed to explore the temporal trends in the annual malaria incidences. Cross correlation and autocorrelation analyses were performed to detect the lagged effect of climate factors on malaria transmission and the auto correlation of malaria incidence. A multivariate time series analysis was conducted to construct a model of climate factors to explore the association between malaria epidemics and climate factors.
Results: The highest malaria incidences were mainly distributed in the central-south counties of the province. A fluctuating but distinctly declining temporal trend of annual malaria incidences was identified (Cochran-Armitage trend test Z = -25.14, P < 0.05). The peak incidence period was May to October when nearly 70% of annual malaria cases were reported. The mean temperature of the previous month, of the previous two months and the number of cases during the previous month were included in the model. The model effectively explained the association between malaria epidemics and climate factors (F = 85.06, P < 0.05, adjusted R(2) = 0.81). The autocorrelations of the fitting residuals were not significant (P > 0.05), indicating that the model extracted information sufficiently. There was no significant difference between the monthly predicted value and the actual value (t = -1.91, P = 0.08). The R(2) for predicting was 0.70, and the autocorrelations of the predictive residuals were not significant (P > 0.05), indicating that the model had a good predictive ability.
Discussion: Public health resource allocations should focus on the areas and months with the highest malaria risk in Hainan. Malaria epidemics can be accurately predicted by monitoring the fluctuations of the mean temperature of the previous month and of the previous two months in the area. Therefore, targeted countermeasures can be taken ahead of time, which will make malaria surveillance and control in Hainan more effective and simpler. This model was constructed using relatively long-term data and had a good fit and predictive validity, making the results more reliable than the previous report.
Conclusions: The spatiotemporal distribution of malaria in Hainan varied in different areas and during different years. The monthly trends in the malaria epidemics in Hainan could be predicted effectively by using the multivariate time series model. This model will make malaria surveillance simpler and the control of malaria more targeted in Hainan.