Background
Malaria is one of the major causes of morbidity and mortality in the world, with less than one million deaths annually reported by WHO [
1]. Malaria is a serious global public health problem, and its prevention and control is addressed in the United Nations (UN) Millennium Development Goals (MDG) [
2]. Yongcheng prefecture, Henan province is a research center of the 2010–2020 plan for national malaria elimination in China [
3], which is an important part of global malaria action plan (GMAP) [
4]. In the past, malaria in Yongcheng prefecture were severe, with incidence as high as 3.34 per 100,000 population in 1970. Malaria incidence was drastically reduced after national comprehensive interventions, which included case management and
Anopheles elimination. In 2003, malaria re-emerged in Yongcheng prefecture after a period of eleven years without a reported case [
5,
6]. Malaria incidence increased to 0.02 per 100,000 population in 2006, which accounted for 4.52% of the total malaria cases in China [
7].
In central eastern China including Yongcheng prefecture, where malaria incidence remains seasonal and unstable, there is a need for timely confirmation of significant factors and development of factors targeting malaria interventions to curtail malaria incidence. Although studies have described malaria and its interventions in detail or factor analysis in central China [
5,
6,
8‐
10], but thus far, attempts to develop predictive models of malaria epidemics with environmental variables, which are accurate on the local scale, have not met with success.
Malaria is one of the important environmental diseases. When the environmental parameters (such as temperature, humidity) permit,
Anopheles mosquitoes would transmit the pathogen,
Plasmodium spp [
11]. Therefore, to figure out how malaria varies due to seasonal or year-to-year changes in environmental variables is essential for the national malaria elimination plan in order for it to allow interventions to be adapted to the specific sites or times of year. Although previous studies proved that weather variables have exerted great influences on malaria incidence in different regions of the world [
12‐
20], controversial issues still remain. First of all, some experts argued that these correlations were questionable [
20,
21]. For example, some of them claimed that rainfall had an significant effect on the incidence of malaria [22-23], whereas others did not detect a significant relationship [24,25]. Moreover, the meteorological factors that have significant statistical correlations with malaria vary greatly between geographic areas within the world, which would complicates the decision-making process in choosing weather monitoring targets [26]. In addition to those controversies,
Anopheles density is considered as a proximate factor of malaria transmission while climatic indicators are distal risk factors, however, few researchers have treated mosquitoes abundance as an independent variable in statistical models to account for malaria incidence due to limited availability of entomological data [27]. Last but not the least, studies only conducted correlation analyses might have been insufficient [24,28] In sum, it is necessary to perform further studies to elucidate the impacts of environmental factors, such as meteorological factors,
Anopheles density on malaria incidence; this study did this based on data from Yongcheng prefecture, China as one example.
The data used in this study were collected from Yongcheng prefecture during the period between 2006 and 2010, and were analyzed using Geographic Information System (GIS) and Generalized Estimating Equations (GEE) approach. The aim was to provide not only a scientific basis for malaria monitoring and control in Yongcheng prefecture but also valuable information for malaria elimination in other areas of seasonal and unstable malaria incidence.
Ethical approval
Ethical approval of this study was obtained from the Ethical Committee of China CDC and permission was also got from the Municipal Government, the Municipal Health Bureau and CDC in Yongcheng city.
Kriging analysis for spatial interpolation of meteorological factors
Some weather values at unsampled locations were interpolated by Kriging method [31]. Monthly observed values of climatic data from eight sampled weather stations neighbouring Yongcheng prefecture was the main input variables. Kriging was performed following the procedures described by Journel et al. [32] and Burgess et al. [33], which included preliminary data analysis, structural data analysis, log kriging estimations, and image generations of spatial results. Spatially distributed values of weather factors in Yongcheng prefecture were estimated based on spherical model in this study, which was regarded as the most widely used semi-variogram model [34]. Monthly weather values at county- or prefecture-lever were calculated by using the method of Zonal statistics. The above analyses were conducted with ArcGIS 9.2 software (ESRI Inc, Redlands, California).
Mapping malaria incidence and average temperature with GIS
We conducted GIS-based analyses of the spatial distribution of yearly malaria incidence and average temperature as well as their visual correlation. The annualized average incidence of malaria per 100,000 persons and annual average temperature at each tow-lever over the five years (from 2006 to 2010) were calculated and mapped based a town-level polygon map of Yongcheng at 1:1000 000 scale with ArcGIS 9.2 software (ESRI Inc, Redlands, California). Regions with different intensities of malaria incidence and various values of average temperature were marked with different colours on the town-level, with higher incidence and temperature being indicated by darker colour.
Temporal analysis with GEE technology
The monthly malaria incidence, An. sinensi density, and weather variables were calculated and plotted to observe their seasonal fluctuations and correlations from 2006 to 2010.
Time varying influencing factors were treated with different time lags, from 0- to 3- month lags, to account for delays in their effects on malaria incidence. The lag size was determined by comparing quasi-likelihood under the independence model criterion (QIC) values in models with various lag sizes [35].
Univariate analyses were made by regressing single factors of interest against monthly malaria incidence to estimate crude associations between malaria incidences and influencing factors. Multivariable models were built to examine the effects of combinations of influencing factors on malaria incidence. Candidate factors selected for the multivariable model were determined through statistical performance of factors in the univariate analysis, and hypothesized relationships. Candidate influencing factors were inputted into the model in their presumed order of importance, and then non-statistically significant factors were removed in their presumed inverse order of importance unless the remaining factors were deemed important for theoretical reasons at α = 0.05 level.
Two kinds of multivariable models were built up in this study using GEE approach. The difference between two models lay in whether they included malaria incidence of the previous month into the model. Model 1 was constructed to estimate the relationships between An. sinensis density, weather variables and malaria incidence. Model 2 was developed to examine the effects of An. sinensis density, weather variables and the malaria incidence of the previous month on malaria incidence.
The goodness of fit of the GEE model was measured by “marginal R-square”, which was interpreted as the amount of variance in the response variables that were explained by the fitted model[36], and QIC value, which was useful in selecting an appropriate correlation structure [8,37,38]. The model with a lowest QIC score and a highest R-square was preferred. GEE analysis was implemented by STATA software 11.0 (Stata Corp. College Station, Texas).
Discussion
This is the first study spatially and temporally exploring the effect of weather variables and vector parameters on malaria incidence in China. The data involved in this study were relevant because all of them were obtained from national monitoring data. In this study, clear spatial heterogeneity and temporal clustering of malaria incidence could be found, with higher incidence distributed in the Southern Yongcheng spatially and in July to November temporally. The finding indicated that areas and months with higher malaria transmission risk should be focused on more public health attention and resources. Spatial heterogeneity of malaria incidence (higher incidence in the south) could be explained by the spatial variability of temperature and the distribution of malaria imported cases. First of all, Southern Yongcheng prefecture has been confronted with higher pressure from imported cases, as its was adjacent to Suixi, Guoyang, Xiao, and Huaibei county in Anhui province, which were identified as high-endemic areas of malaria (incidence > 30/100,000) [
10]. Furthermore, its spatial decreasing pattern from the north to south was somehow in accordance of temperature change, particularly in 2006. Only temperature was selected to assess the influencing factor associated with spatial heterogeneity owing to the founding from previous studies that temperature might be a major determinant of malaria incidence in China [26]. Therefore, public resource should be allocated proportionally in different areas based on the risk predicted by imported possibility and average temperature, and close collaboration should be established between Henan and Anhui to effectively control and prevent malaria together. From the perspective of time, annual malaria incidence in Yongcheng prefecture showed an obvious decrease from 2006 to 2010. However, prevention and control of malaria should not be taken slightly, because Yongcheng prefecture is still at risk for malaria due to the existing of
An. sinensis, suitable weather condition for the growing of
An. sinensis and
Plasmodium, and the risk has been possibly varying with weather changes and population movement [39,40]. Determining the principal influencing factors of malaria incidence would be beneficial for malaria risk assessment and thus providing a basis for the policy making for malaria control technologies.
Malaria is a vector-borne infectious disease, and as such, is sensitive to environmental change [40-42]. Anopheles density, as a proximate environmental factor of malaria transmission, play an important role in estimating and predicting malaria risk [27]. Climatic variables have also been established as important environmental drivers of malaria transmission [43], because of their impacts on the growth and reproduction rates of mosquitoes, the temporal activity pattern of the population as well as the life cycle of Plasmodium [44-47].
The best-fit model (model 2) derived from the study was reliable and had a good fit and predictive validity (QIC = 16.934, P<0.001, R2 = 0.818), which provided insights into the most important drivers of P. vivax malaria, including maximum temperature, average humidity and incidence of previous month that influenced seasonal fluctuation of P. vivax malaria incidence. The result that temperature rise would contribute to malaria transmission was in agreement with some researchers [46-52], although there were still some other researchers who argued that this relationship was not significant [19,53], or that it was uncertain [54]. It has been demonstrated that temperature increase would improve the survival chances of Anopheles and thus contribute to the malaria transmission [55,56] Moreover, relative humidity exerted an influence on the survival of mosquito eggs and adults and the moderate increase in malaria risk associated with average humidity observed in this study was consistent with previous findings [57]. Conversely, some literature found a correlation between rainfall and malaria [49], while other studies found no correlation [51]. In this study, rainfall failed to enter the best-fit model as a predictor of malaria epidemics. The phenomenon could be explained by the complex nonlinear association between rainfall and malaria incidence. Rainfall is beneficial to the growth and reproduction of mosquito if it is moderate, because it often leads to puddles and increased local humidity; however, excessive rain can also wash away eggs and completely destroy breeding sites [58]. This result indicated that it was not necessary to consider rainfall as a predictor in Yongcheng, which made malaria surveillance simpler in this area.
In addition to factors mentioned above, although
An. sinensis densities failed to enter into best-fit model (model 2), D
bait was included in model 1. The reasons why D
bait rather than D
net were included in model 1 probably lay in two aspects. Firstly,
An. sinensis is slightly exophagic (biting outdoors) [59], and thus D
bait is more representative of malaria transmission in Yongcheng. Secondly, most of the malaria cases in Yongcheng were farmers, and they would like to sleep and work outdoors in the summer without effective protections, thus having more opportunities to be infected [
5,
9]. Therefore, D
bait would contribute more to the prediction of malaria incidence. Moreover, comparing the time series plots of D
bait (Figure
5), expected values predicted by model 1(Figure
6) and actual values (Figure
6), we found out expected values were over-estimated in 2009 and 2010 when actual malaria incidence was relatively low, which mainly due to the existence of predictor D
bait reasoned by their similar rising tendency. It could be concluded that
An. sinensis density probably had its shortcomings as a routine monitoring and predicting index.
An. sinensis density may be a better predictor of malaria incidence when transmission is relatively high, as many of the female mosquitoes may have been infected by
Plasmodium, and increase in
An. sinensis density would lead to a direct rise in malaria incidence. However, when the incidence is at a low level, most of the female mosquitoes are possibly free of
Plasmodium. In this circumstance, when a healthy human being is bitten by a female
An. sinensis the probability of infection by
Plasmodium is low. Therefore, we concluded that rise in
An. sinensis density would possibly contribute little to the increase of malaria incidence in low transmission areas, which agreed with the result of another study that malaria transmission potential would be very low in spite of a high human biting rate in unstable malaria areas [60]. Therefore, control target should vary with the severity of malaria epidemics. When malaria incidence is high, public resource allocation should be focused on mosquito control and elimination; however, when malaria incidence is low, the key control point should lie in controlling sources of infections. Furthermore, it is necessary to find a substitute, such as entomological infection rate (EIR), which can overcome the weakness of
An. sinensis density as an indicator for malaria surveillance and prediction, although it can be used as a good index for predicting malaria potential risk as a previous study showed [40].
As far as the lag effect was concerned, this study found significant one month lag effects of entomological and meteorological variables on malaria incidence, and this finding was supported by several earlier studies [17,19,35,49]. This phenomenon could be explained by the approximately one month duration of malaria infection cycle. The time incorporates several processes above. An adult mosquito first bites an infected human, and then the parasite develops in the adult mosquito (Extrinsic Incubation Period). Ten days later when the P. vivax sporozoites move the salivary glands, the mosquito transmits malaria to a human when it takes another blood meal. Once the person is infected, time to development of malaria symptoms and infectivity (Intrinsic Incubation Period) takes about another 1–2 weeks [40,61]. Knowing the approximate lag size of effects on malaria incidence would benefit us to get prepared for quick and effective response on malaria epidemic easily at least one month in advance.
The transmission of malaria is complicated, and we still need further research to figure it out. For example, the temporal variation in malaria incidence could be also partially explained by continuous and effective control efforts Chinese local and national health agencies made, such as treatment in the rest period of malaria(conducted from 2004) [62] and comprehensive vector control action characterized by biological larviciding and residual spray(conducted from 2007) [
8]. Human interventions are failed to be inputted into the predictive model in this study because it is difficult to measured and quantify.
Competing interest
The authors declare that no competing interests exist.
Authors’ contributions
YZ conceived this study and was involved in manuscript drafting. QYL participated in the design of the study, and helped in drafting the manuscript. RSL was involved in the study design and manuscript modification. XBL, GCZ, JYJ, HSL, and ZFL took part in the data collection and entry. All authors have read and approved the submitted version of the manuscript.