Mapping the occasionality of inevitable dengue fever prevalence in China

doi:10.21203/rs.3.rs-3810038/v1

Download PDF

Research Article

Mapping the occasionality of inevitable dengue fever prevalence in China

https://doi.org/10.21203/rs.3.rs-3810038/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

Dengue fever (DF) is an acute mosquito-borne viral infectious disease in the world, and increasing DF outbreaks in China have posed serious impacts on public health in recent years. Thus, comprehensively investigating spatiotemporal features and driving or restrictive factors of DF epidemics is critical for the improvement of intervention capacity against this disease.

Methods

Two famous dividing lines (Hu Line and Q-H Line) were applied to divide the mainland into four regions for geographically characterizing China’s DF prevalence. We defined the stages with suitable relative humidity, temperature, and precipitation as basic time windows for the mosquito vectors’ activities. The Random Forest (RF) model was employed to fit the relationships between local epidemics and included climatic and socioeconomic factors, quantify these factors’ contribution, and then map the city-level risk of local DF prevalence.

Results The situation of China’s DF epidemics was increasingly serious due to ascending intensities of local prevalence triggered by more frequently imported cases. The cities with DF cases, together with their frequencies and intensities presented clear geographical disparities on the city scale, and well matched with the time windows for either DF transmission (95.74%) or mosquito vectors’ activities (83.59%). Among these included factors, the imported cases acted as the driving factor of local epidemics in the region I and III because of not only their strongest association (r=0.43, P<0.01; r=0.46, P<0.01) but also the largest contribution (24.82% and 31.01%). Moreover, in terms of SHAP values, the imported DF cases possessed a steady promoting impact on local epidemics, while the rest 11 inputs had comprehensive promoting or inhibiting effects with different inflexion values. Besides, the RF models considering the time windows owned higher testing AUC value (0.92) while fitting the relationships between local DF epidemics and potential factors, by which we successfully identified about 96% of the cities with the highest and higher risks of local DF prevalence.

Conclusions China is being confronted with increasingly larger intensities of occasionally localized DF epidemics triggered by unavoidable higher frequencies of imported epidemics. This study would supply useful clues for the health authorities improving their intervention capacity against this disease.

Inevitability

Occasionality

Dengue fever

China

Time windows

Random forest

China’ DF epidemic was featured by ever-increasing frequency and intensity;
Spatial distribution of China’s DF epidemic was divided by Hu Line and Q-H Line;
Capability of imported DF case triggering local epidemics is rising;
Local DF prevalence is geographically restricted by the city-level time windows;
Unavoidable DF outbreak would occasionally occur and expand toward inland cities.

Dengue fever (DF) is an acute infectious disease caused by dengue virus, which is transmitted by Aedes albopictus and Aedes aegypti [1]. Its natural focuses are widely distributed in global tropical and subtropical areas (e.g., the Southeast Asia, the Western Pacific, and the South Africa) [2], causing about one-third of the global population exposed to this disease [3]. Since the globalization and climate change was much deeper and continuous in recent years, the increasing incidence and range DF epidemics have become an important public health concern that should not be underestimated.

In China, the DF prevalence is mainly induced by imported cases from the global natural focuses. No DF case was reported during 1949–1977 until a sudden outbreak occurred in the city of Foshan, Guangdong Province in 1978 [4]. From then on, China’s DF epidemic tended to be intermittent and has been included in the list of notifiable communicable diseases since 1990s [5]. Due to more and more serious DF epidemic in the world and China’s increasingly important roles in the international economy and trade affairs, local DF prevalence induced by imported cases has ascended substantially in recent years [6, 7]. Meanwhile, the DF prevalence presented obvious spatial expansion from southern China (southwest border and southeast coastal areas) toward inland northern regions, such as Jiangxi (Yichun), Chongqing, Henan (Xuchang) and Shandong (Jining) [8, 9]. Overall, the DF prevalence in China has shown shorter and shorter time intervals and clear spatial expansion.

Numerous studies have been conducted by domestic and foreign scholars on a series of factors influencing the spread and prevalence of this disease at various levels, including the dengue virus, mosquito vectors, susceptible population, environmental conditions, and socioeconomic status [10, 11]. Among these influential factors, environmental conditions, like climate, hydrology, vegetation coverage, mainly affect the activity of the dengue virus, as well as the breeding, survival surroundings, and biting power of mosquito vectors [12, 13]. Socio-economic factors, such as population density and mobility, land use, convenience level of public transportation, residents’ income level and living habits, chiefly alter the probability of human-mosquito contact [14, 15]. In general, the DF transmission and prevalence is a complex and typical geo-ecological process.

Earlier studies on the spatial and temporal variations at various scales, and their influencing factors of DF epidemics not only have greatly supported the prevention and control against this disease in the absence of effective clinical vaccines, but also have improved our understanding of the DF prevalence and spread in China [16, 17]. However, comprehensive knowledge is needed regarding the current ever-increasing growth of cities with DF outbreaks, the limitations on geographical patterns and the quantitative contributions of potential influencing factors of DF epidemics in China.

Therefore, our study was conducted to 1) characterize the frequency and intensity, geographical distribution, 2) use a random forest (RF) model to quantitatively explore the relationships between local DF prevalence and a series of potential influencing factors, and then 3) map the city-level risk of DF prevalence across China. The results of this study will provide important support for strengthening the capacities of interventions against this disease and comprehensively raising the current understandings on China’s DF epidemics.

Data collection

Dengue cases The dengue cases were sourced from the National Infectious Disease Surveillance System, covering the period from 2003 to 2022 and encompassing confirmed dengue cases in China. The dataset included information on the date of onset, residential address, and reporting address of these cases. Based on the origin source, dengue cases were classified into imported and local cases. Imported cases referred to those who have traveled to a dengue-endemic country or region within 14 days before the onset of the disease. Local cases were those who have not left the city within 14 days before the onset of the disease or those who have left the city within 14 days before the onset of the disease and visited other domestic dengue epidemic cities [18]. An epidemic outbreak was defined as the occurrence of three or more local cases within the maximum incubation period of 14 days in a city. If no local cases were reported within 14 days after the outbreak, it was considered that the outbreak had ended. The duration from the beginning of the outbreak to its end was referred to the outbreak period for the local DF epidemic. After filtering out isolated local cases occurring outside the outbreak period, imported and local cases were consolidated at the city level based on their reporting address codes.

Potential influencing factors According to previous research [10–15], we collected 11 potential influencing factors, which were categorized into two dimensions: socioeconomic factors and natural factors (Table 1). Among 6 socioeconomic factors, factors related to the economy and population included 1 km × 1 km gridded gross domestic product (GDP) and population density (Popu). Urbanization-related land-use factors included the percentage of four land use types: cropland (Cropland), forest (Forest), water (Water), and impervious surface (Impervious), which were collected from yearly 30m × 30m gridded data. In addition to the above socioeconomic factors, 5 natural factors including monthly maximum temperature (Tmax), monthly average temperature (Tmean), monthly minimum temperature (Tmin), monthly average relative humidity (RH), and monthly precipitation (Prec) were also considered as potential influencing factors for the spread of the epidemic, which were collected from 1 km × 1 km gridded data at a monthly scale.

To facilitate statistical and spatial analysis, the variables of each factor at the city level were extracted using zoning statistics and spatial connectivity tools in ArcGIS 10.6 software. Furthermore, for missing data in certain years, adjacent year data was used as a replacement.

Table 1

Data collection and sources in this study
Data group	Selected variables from previous studies	Data unit	Time scale	Resolution	Source
Socioeconomic factors	Gross domestic product per capita (GDP) [10, 14]	Million dollars/km²	Annual	1 km	The Scientific Data (https://doi.org/10.6084/m9.figshare.17004523.v1) WorldPop(https://hub.worldpop.org/) Earth System Science Data (https://doi.org/10.5281/zenodo.5816591)
	Population density (Popu) [10, 14, 15]	Person/km²	Annual	1 km
	Annual average percentage of cropland (Cropland) [15]	%	Annual	30 m
	Annual average percentage of forest (Forest) [15]	%	Annual	30 m
	Annual average percentage of water (Water) [15]	%	Annual	30 m
	Annual average percentage of impervious (Impervious) [15]	%	Annual	30 m
Natural factors	monthly maximum air temperature (Tmax) [11, 12]	degree Celsius	Monthly	1 km	Earth System Science Data (https://doi.org/10.5281/zenodo.5112232) National Earth System Science Data Center (http://www.geodata.cn) National Tibetan Plateau/Third Pole Environment Data Center (https://data.tpdc.ac.cn/)
	monthly mean air temperature (Tmean) [11–13]	degree Celsius	Monthly	1 km
	monthly minimum air temperature (Tmin) [11, 12]	degree Celsius	Monthly	1 km
	monthly mean relative humidity (RH) [13]	%	Monthly	1 km
	monthly precipitation (Prec) [11–13]	0.1mm	Monthly	1 km

Descriptive analysis on frequency and intensity of DF outbreaks in China

The yearly counts of local and imported cases were separately aggregated from 2003 to 2022. According to the occurrences of either imported or local epidemics, these cities were classified into four groups including the cities with either imported or local epidemics, the cities with both of them, the cities with imported epidemics only, and the cities with local epidemics only. Furthermore, we analyzed the spatial distribution characteristics of these four groups of cities and determined the correlations between imported and local epidemics by using the Spearman. To further characterize DF epidemics in China in the past two decades, we established statistical indicators of DF epidemics from the perspective of prevalence frequency and intensity (Table 2). Subsequently, we analyzed the variations in the frequency and intensity of epidemic prevalence and characterized the spatial distribution of overall frequency and intensity.

Table 2

Statistical indicators of DF epidemics
Category	Definition	Formula
Frequency	Number of months with epidemics within the specified time	Number of months with epidemics / the specified time
	Times of outbreaks within the specified time	Times of outbreaks / the specified time
Intensity	Monthly average number of cases in the months with epidemics	Total cases / number of the months with epidemics
	Average number of cases per outbreak	Total cases / times of outbreaks

The Retrieval of the time windows for local DF epidemics

The suitability of natural conditions such as temperature, humidity and precipitation influenced mosquito breeding and activities, which directly affected the spread of local epidemics. Previous studies have found that it was suitable for the breeding of Aedes mosquito when the monthly minimum temperature was higher than 10 ℃ [19–21], the monthly average temperature was between 15–32 ℃ [22, 23], the monthly maximum temperature was lower than 38 ℃ [20, 21], the monthly average relative humidity was between 60% and 90% [24, 25] and the monthly precipitation was between 60 mm and 650 mm [26, 27]. Moreover, we defined the periods meeting all these conditions as the time windows for the mosquito vectors’ activities. Considering the incubation periods of dengue viruses [28], we extended the time windows of mosquito vectors by one month and defined it as the time windows of local DF prevalence. To further evaluate the effectiveness of the time windows, we defined the match degree as the proportion of the number of outbreaks where the outbreak period overlapped with the time windows to the total number of outbreaks. The higher the match degree, the higher the proportion of outbreaks occurring within the time windows.

Identification of the dengue driving forces at the city level

We aggregated the number of imported cases within the time windows from 2003 to 2019 as an independent variable at the city level. Meanwhile, we counted the number of local cases in the year and transformed it into a binary variable as the dependent variable according to the occurrence of local epidemics. Additionally, six socioeconomic variables and five environmental variables (Table 1) of the average value in the months of the time windows were considered. And preliminary correlation analyses between the number of local cases and input variables were conducted using the Spearman. The Random Forest (RF), the Gradient Boosting Machine (GBM) and the Support Vector Machines (SVM) were powerful machine learning methods for classification and regression, which were typically employed for the analysis of the dengue driving forces [29–31]. We applied these three methods to fit the relation between local epidemic outbreaks and the influencing factors. The data from 2003 to 2018 were divided into the training set (70%) and the test set (30%). In the RF model, the number of trees to grow was the main parameter, and a range of tree numbers (from 100 to 2000 with an interval of 100) were set to select the optimal parameter based on the performances of the model. In the GBM model, the number of trees to grow and the learning rate were the main parameters, and the ranges of tree numbers and learning rates are 100–2000 (step = 100) and 0.01–0.2 (step = 0.01). In the SVM model, we applied ‘rbf’ as the kernel function, and the ranges of regularization parameter and kernel coefficient were 0.1-3 (step = 0.1) and 0.01–0.5 (step = 0.01). In these models, the five-fold cross-validation was performed to boost the stability of modeling. The Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve was used as a measured value of the model predictive ability and a larger AUC indicated better predictive capability of the model [32]. The model accuracy characterized by the AUC, was as follows: 0.50–0.70, indicating a poor model; 0.70–0.80, suggesting an average model; and 0.80-1.00, reflecting a good model [33, 34]. According to the AUC of above three models, we looked for the optimal model.

The SHapley Additive exPlanations (SHAP) method, widely applied for interpreting machine learning models [35], was introduced to quantify the contributions and driving forces of each independent variable to local epidemic outbreak. Globel SHAP values were calculated for each variable in the RF model, with larger absolute values indicating greater contributions to the model. Local SHAP values for each instance reflected the driving force for the epidemic outbreak. Positive SHAP values indicated a positive driving force, with larger values indicating a stronger positive influence. Conversely, negative SHAP values indicated a negative driving force, with smaller values signifying a stronger negative influence [36, 37].

To analyze the impact of the time windows on the model, the model without considering the time windows was built. And the model considered the imported cases and the mean values of natural environmental variables in the entire year, while keeping other variables constant. The specific modeling process and evaluation methods were consistent with those described above. By comparing the testing accuracies of different models, we selected the optimal model for mapping and assessing the risk of local DF prevalence in China from 2019 to 2022. In this study, we utilized the sklearn package and shap package in Python for building and interpreting the RF model.

Current situation of the DF prevalence in China

There was total 98,560 DF cases displaying an overall upward trend over the past two decades although the amounts of both imported and local DF cases sharply decreased in 2020–2022, causing total 329 DF outbreaks within 104 months (Fig. 1a). The proportions of local DF cases in the total were much higher than those of imported cases in 15 years, especially holding a steady situation since 2012 (Fig. 1b). At the same time, the counts of local DF cases were significantly correlated with those of imported cases on the national scale in 2003–2012 (r = 0.23, P < 0.01), 2013–2022 (r = 0.44, P < 0.01), and 2003–2022 (r = 0.41, P < 0.01). These results indicated that China’s DF epidemic was dominated by the local prevalence and was increasingly closely associated with imported DF epidemics.

Meanwhile, there were more and more cities being affected by this disease. Despite of an acute drop in 2020–2022, the amounts of cities with either imported or local epidemics (Fig. 1c), both of them (Fig. 1d), and only imported epidemics (Fig. 1e) displayed similar temporal variations. By contrast, the counts of cities with local epidemics only persisted below 3 except for 2014 with 8 cities (Fig. 1f), which seemed to be not serious. Even though, the amounts of cities with local epidemics were significantly associated with those of cities with imported epidemics on the nation scale (r = 0.79, P < 0.01), which further indicated that the local DF prevalence was obviously associated with imported epidemics on the city scale across China.

The frequencies of imported DF epidemics were less than 9 months per year before 2006, and then quickly ascended to 11 months per year or higher since 2007 despite of a drop to 9 in 2021–2022 (Fig. 2a). In comparison, the frequencies of local epidemics possessed a relatively slower and fluctuating uptrend varying from 0 to 8. But, in terms of the intensity, local DF epidemics were more serious than imported epidemics although the latter displayed a steadier and quickly uptrend (Fig. 2b). Besides, another frequency (i.e., the times of local DF outbreaks per year) also presented clear uptrend and their intensities persisted at relatively higher levels than 100 cases per outbreak since 2012 (Additional file 1). These results indicated that the local DF epidemic was increasingly serious in China over the past two decades.

Geographical distribution of the DF epidemic

The city-level DF epidemics were spatially featured across China. As illustrated in additional file 1, the vast majorities (96% − 100%) of the cities with either imported or local epidemics were distributed on the right side of Huhuanyong line (Hu Line). Among them, more than 75% was located on the south side of Qin Mountain – Huai River line (Q-H Line) and displayed a downtrend from 2012 (Fig. 1c), which implied that the DF prevalence expanded towards the north side of Q-H Line. Meanwhile, the overwhelming proportion of the cities with both imported and local epidemics was also distributed on the south side of Q-H Line. In comparison, the cities with only imported epidemics were mainly located on the right side of Hu Line, while fewer cities were infected by only local epidemics. As Hu Line and Q-H Line acted as important geographical dividing lines for China’s DF epidemics, three regions were divided for China’s DF epidemics including region I (the cities on the right side of Hu Line), region II (the cities on the north side of Q-H Line), and region III (the cities on the south side of Q-H Line). These analyses implied that the local DF epidemics tended to be geographically differentiated on the city scale across China.

On the city scale, China’s DF epidemics termed by the frequency and intensity also presented geographical disparities. The cities with higher average frequencies of imported epidemics than 1 month per year were mainly located in the region III (Fig. 3a). Among them, a few inland provincial capitals and several prefecture-level cities in the southwestern border and southeastern coastal regions owned much higher average frequencies than 1.5 months per year. By contrast, fewer cities possessed higher average frequencies of local epidemics than 0.5 months per year and mainly concentrated in the region III (Fig. 3b). Similarly, the cities with relatively higher intensities of imported epidemics were mainly located in the region III (Fig. 3c), while those of local epidemics were sparsely distributed here (Fig. 3d). It should be mentioned that the provincial capitals often possessed much higher frequencies and intensities of imported epidemics, and that some inland cities, like Ji’ an, Yichun, Chongqing, and Hangzhou, owned the highest intensities of local epidemics despite of relatively lower frequencies. These results showed that the region III tended to be easily confronted by the imported epidemics and to be much more heavily affected by local epidemics.

Time windows for local DF prevalence

In terms of match degrees, the time windows for local DF prevalence performed much better (95.74%) than those of the mosquito vectors’ activities (83.59%). Due to different average beginning or duration on the city scale, the time windows showed obvious spatial differences in the region I (Fig. 4), among which the region II possessed relatively later beginnings (June – July) and shorter durations (three - four months) while the region III displayed earlier beginnings (April – May) and longer durations (seven - eight months, Fig. 4u). As a result, some cities in the region II occasionally closed their time windows or opened for a short period as those in the region III held it for a longer time.

Relationships analysis within the time windows

As far as the relationships between the local DF epidemics and 12 potential influencing factors within the time windows were concerned, the local DF epidemics termed by city-level case numbers were significantly positively associated with most natural and socioeconomic factors, except for Cropland (r=-0.14, P < 0.01) (Table 3) in the region I, which was similar in the region III. In contrast, the local DF epidemics showed significant correlations only with several factors (Imported Cases, GDP, Popu and Impervious) in region II. Among these inputs, the counts of imported cases consistently possessed the strongest association.

Table 3

Correlation coefficients between local DF epidemic and input variables.
Variable type	Variable name	r
		Region I	Region II	Region III
	Imported Cases	0.43**	0.12**	0.46**
Socioeconomic variables	GDP	0.17**	0.08**	0.25**
	Popu	0.14**	0.08**	0.18**
	Cropland	-0.14**	0.01	− .09**
	Forest	0.10**	-0.01	0.01
	Water	0.13**	0.02	0.11**
	Impervious	0.06**	0.06**	0.18**
Environmental variables	Tmax	0.09**	0.02	0.11**
	Tmean	0.13**	0.02	0.12**
	Tmin	0.16**	0.02	0.12**
	RH	0.15**	-0.03	0.08**
	Prec	0.21**	-0.03	0.19**
Notes: r is the Pearson correlation coefficient. ** indicates this value is significant at the level of 0.01. ImportedCases: The imported DF cases; GDP: Gross domestic product; Popu: Population density; Cropland: The percentages of cropland; Forest: The percentages of forest; Water: The percentages of water; Impervious: The percentages of impervious; Tmax: The average values of maximum temperature; Tmean: Monthly average values of mean temperature; Tmin: Monthly average values of minimum temperature; RH: Monthly average values of relative humidity; Prec: Monthly average values of precipitation.

The accuracy of RF, GBM, SVM models was showed in the additional file 1, which showed that the RF model was superior compared to other two models. In terms of AUC values (Table 4), the RF models possessed good performances of fitting the relationships between local epidemics and influencing factors in the region I (0.92) and III (0.85). In particular, the RF models considering the time windows of local DF epidemics in the region I tended to possess slightly higher AUC value (0.92) than those without regarding to the time windows (0.90). Accordingly, the RF model considering the time windows could be further used for exploring the factors’ contributions, as well as mapping the city-level risk of local DF prevalence.

Table 4

The AUC values derived from the RF modelling
	Region I			Region III
	Training	Testing	Prediction	Training	Testing
Model 1	0.87	0.92	0.91	0.88	0.85
Model 2	0.87	0.92	0.88
Model 3	0.89	0.90
Note: Model 1 and Model 2 respectively represented the models with regard for the time windows of local DF prevalence and mosquito vectors. Model 3 represented the model without regard for the time windows. Region I and III were assigned for the regions on the right side of Hu Line and on the south side of Q-H Line.

Dominant influencing factors

Regarding the contributions to local DF epidemics, the imported cases ranked the first (24.82%) and greatly exceeded the second (Tmin, 16.88%) and the third (Forest, 8.58%) in the region I, while the rest factors possessed smaller contributions (Fig. 5a). In the region III, it also ranked the first with the highest contribution (31.01%, Fig. 5b). Besides, top five inputs with relative higher contributions included Tmin, Forest, Popu, Prec and Tmean in the region I, which differed from those in the region III (GDP, Popu, RH, Cropland, and Forest). For this point, the influencing factors’ groups were differentiated among these two regions. In the region I, the natural factors made larger contributions (41.04%) than those of socioeconomic conditions (34.15%) to the local DF epidemics, while the socioeconomic group possessed much higher contributions (44.26%) than those of natural group (24.74%) in the region III. These results implied that the dominant influencing factors of local DF epidemics were geographically differentiated although the local DF prevalence was absolutely induced by the imported cases.

According to the SHAP values, these 12 inputs possessed protective or risk effects on the city-level local DF epidemics in the region I. Along with the count of imported cases ascending, its promoting effect on the city-level local epidemics were increasingly powerful before 50 imported cases and then stopped to persist at a high promotion (Fig. 6a). In comparison with its single promoting effect, the rest factors generally possessed composite promoting and inhibiting impacts, resulting in two groups including inhibiting before promoting (Group 1) and promoting before inhibiting (Group 2). Among top five factors (Fig. 5a), Tmin, Forest, Popu, and Prec belonged to Group 1 with respective inflexion values of 20℃ (Fig. 6b), 60% (Fig. 6c), 1,000 persons per square kilometer (Fig. 6d), and 185 millimeters (Fig. 6e) while other one variable (Tmean) in Group 2 turned their promotive effects to inhibitive roles at the value of 21 ℃ (Fig. 6f). The rest six factors could be attributed into Group 1 (GDP, Impervious and Water) and Group 2 (RH, Cropland and Tmax), respectively. In comparison, included 12 input variables also possessed comprehensively promoting or inhibiting effects on the local epidemics in the region III (Additional file 1), although their inflexion values were slightly different from those in the region I. Above analyses validated that the imported DF cases acted as the trigger of local epidemics, and showed that the local epidemics were comprehensively promoted or inhibited by natural and socioeconomic factors on the city scale across China.

Mapping the city-level risk of local DF prevalence

Among total 295 cities with time windows opened for the local epidemic in the region I, there were 20, 25, 23, 34, and 193 cities categorized into five levels for their various probabilities of local epidemics in 2019. It was satisfied that 20 cities with the highest risk were exactly identified because the local DF prevalence occurred in these cities indeed, showing a 100% hitting rate. By contrast, among the cities beyond the risk of higher (45 cities), middle (68 cities), lower (102 cities) levels, and the lowest (295 cities), the hitting rates decreased gradually to 95.56%, 79.41%, 67.65%, and 29.15% even though the accumulative counts of identified cities increased to 43, 54, 69, and 86. Although the RF models’ mapping abilities displayed gradual declines in 2020–2022, the hitting rates in 2020 were still acceptable because one city (i.e., Guangzhou) was not only identified for its highest risk but also infected by this disease indeed.

Table 5

Assessments on the mapping of the city-level risk of local DF prevalence in the region I.
Years	Statistics of the cities	Probability of local DF epidemics
Years	Statistics of the cities	0.8 ~ 1	0.6 ~ 1	0.4 ~ 1	0.2 ~ 1	0.0 ~ 1
2019	Number of cities identified (NCI)	20	45	68	102	295
	Number of cities with actual epidemics (NCA)	20	43	54	69	86
	Percentages of NCA in NCI	100.00%	95.56%	79.41%	67.65%	29.15%
2020	Number of cities identified (NCI)	1	2	5	23	295
	Number of cities with actual epidemics (NCA)	1	1	2	2	4
	Percentages of NCA in NCI	100.00%	50.00%	20.00%	8.70%	1.36%
2021	Number of cities identified (NCI)	0	1	5	17	295
	Number of cities with actual epidemics (NCA)	0	0	0	0	1
	Percentages of NCA in NCI	/	0.00%	0.00%	0.00%	0.34%
2022	Number of cities identified (NCI)	0	1	4	14	295
	Number of cities with actual epidemics (NCA)	0	0	1	1	2
	Percentages of NCA in NCI	/	0.00%	25.00%	7.14%	0.68%
Notes: The probabilities, like 0.80–1, 0.60–0.80, 0.40–0.60, 0.20–0.40, and 0–0.20, were respectively categorized as the highest, higher, middle, lower, and the lowest risk levels.

According to local DF risk in the region I in 2019 (Fig. 7), all of 20 cities with the highest risk identified by the RF models were mainly located in the southwestern border (two cities) and southeastern coastal areas (16 cities). Meanwhile, 25 cities in the southeastern areas and central regions, as well as 23 cities scattered in the region I were also recognized as the areas with higher and middle risks. As a result, above 68 cities were mostly distributed in the region III. Besides, the cities recognized as the areas at various risk levels in 2020–2022 were also mainly located in this region. These results indicated that it was feasible to employ the RF models in mapping the city-level risk of local DF prevalence in China.

Since DF epidemics in China were ever-increasingly serious in recent years, it is urgent to reveal the comprehensive features of DF outbreaks for appropriately mapping its risk. Our study analyzed its frequencies and intensities, and then identified their potential influencing factors for mapping the probability of city-level local DF outbreaks through the RF models. Several notable findings were achieved and would provide some useful clues for making targeted interventions on this disease.

Previous studies have pointed out that China experienced an ever-increasingly serious threat enforced by DF epidemics in terms of the incidence rates or other indices on various spatial scales [7, 9, 16, 38], and that local epidemics were closely correlated with imported epidemics in the past years [39, 40]. Similarly, our study found that the DF epidemics in China presented continuous uptrends of DF case amounts, as well as obvious spatial expansions towards many more inland cities with increasing frequencies and intensities of city-level DF epidemics. Moreover, the capability of imported epidemics initiating local DF prevalence was rising due to their increasingly stronger association. As a result, many inland cities like some regional hub cities (e.g., the provincial capitals) tended to possess relatively lower frequencies but higher intensity of local epidemics, as well as the inverse appearance of high frequencies and low intensity of imported epidemics. One reasonable explanation is that there have been increasingly larger counts of imported DF cases from some global endemic countries or territories (e.g., the Southeast of Asia, the Central America, and so on) because of closer and stronger connection between China and these countries/territories in recent years [39, 41]. Apart from traditional regions (i.e., the Southeast and Southwest of China) often infected by imported epidemics [5, 6], the inland regional hub cities featured by bigger airports and larger export-oriented economy were constantly confronted by ascending numbers of inbound or outbound tourists [42, 43]. Another potential reason is the spillover effects of DF cases from either the traditional regions or hub cities to their surrounding inland cities with sustainable environment conditions [9, 44]. Hence, it can be concluded that China’s DF epidemics were featured by its ascending frequencies and intensities within much more inland cities. Accordingly, we cautiously suggest that additional attention should be emphatically paid to inland regional hub cities and their surrounding cities, especially in case of ever-growing imported epidemics and their increasing initiating ability.

China’s DF epidemics were geographically differentiated around two famous dividing lines (Hu Line and Q-H Line) [45, 46]. The local DF prevalence was still geographically restricted within the region I, especially in the region III by far, for which the distribution of city-level time windows for mosquito vectors’ activities (Additional file 1) may be a rational explanation since this disease is transmitted by Aedes species (Aedes albopictus and Aedes aegypti) in some specific phases with suitable environmental conditions (i.e., climatic elements) [11, 12]. Moreover, the city-level match degrees among the actual occurrences and the time windows for either the mosquito vectors’ activities or local DF transmission were satisfying in the region I. Under this circumstance, it was reasonable that local DF prevalence was occasionally reported in some cities in the region II with late opened and shortly held time windows. That is to say that the time windows were very crucial and non-negligible for local DF prevalence across China. Accordingly, we believe that the time windows would provide helpful information for relative departments implementing timely interventions on this disease.

Apart from the spatial differences as mentioned above, the geographical disparities of local DF prevalence were also featured by its dominant influencing factors within the region I and III differing from each other, which may be partially attributed to the different coefficients of variances (CV) of these influencing factors (Additional file 1). This finding was similar to our earlier investigation on the comparison of dominant influencing factors on the DF epidemics in two traditional hotspot regions (the Pearl River Delta and the Border of Yunan and Myanmar) [17]. Thus, it can be concluded that the local DF prevalence was not only geographically restricted by the time windows but also spatially characterized by regionally differentiated influencing factors across China. Therefore, we suggest that the health authorities of each city should take both the status of time windows and regional attributes into account for either making targeted measures before local epidemic occurring or implementing efficient interventions once imported cases initiating local prevalence.

Furthermore, the knowledge on China’s DF epidemics was comprehensively improved. First, a large and increasing number of ceaseless imported DF cases would be undoubtedly foreseen in the future since China is playing more and more important roles in the international economy and trade affairs [47, 48], which means that China would be confronted by the inevitable DF prevalence (i.e., either imported or local epidemics). Meanwhile, the domestic loops of socioeconomic development are being accelerated in China so that the spillover of DF cases would be constantly observed among domestic regions [49], especially within the cities with opening time windows for DF transmission. However, the inland cities hit by local DF prevalence remained uncertain along with imported epidemics expanding from traditional southeastern coastal regions or southwest border areas northwards many inland regional hub cities (i.e., the provincial capitals). In other words, there is somewhat occasionality of the cities hit by local epidemics initiated by imported epidemics in the inland regions with opening time windows. Here, we cautiously advise three proper solutions to the challenges imposed by the inevitability and occasionality of DF prevalence in China. First of all, we need to keep our sensitive eyes on the overseas DF epidemics, by which the inbound tourists from these endemic areas could be timely acquired for judging the situation of imported epidemics in China. Secondly, sufficient surveillance of climatic elements on the city scale should be efficiently utilized for determining the status of time windows across China, especially in the inland regional hub cities and their surrounding areas. The final and the key point is properly and scientifically building a robust and reliable prediction model, like the RF models constructed in this study, by which the health authorities could make targeted measures for preventing and controlling this disease. Of course, there were two prerequisites for our investigation that the worldwide natural focuses of this disease cannot be eliminated in a short term, and that China remains as an unnatural focus of this disease.

Several limitations are worth noting. First, the acquisition of meteorological conditions data with a higher temporal resolution (e.g., weekly, ten-days) would be helpful for more finely characterizing the time windows on the city scale, by which the current match degrees between the city-level time windows and actual stages of local DF prevalence may be well increased and then effective interventions could be more precisely and timely implemented. Second, the effectiveness of city-level time windows could be further validated through obtaining synchronous surveillance data of mosquito vectors, by which the capabilities of RF models fitting comprehensive relationships between local epidemics and potential factors within these time windows might be improved for subsequent mapping of the city-level risk for local DF prevalence across China. Finally, as the spillover of DF cases among domestic regions played important role in the DF prevalence across China, an efficient solution should be scientifically proposed to characterize the network of relationships among domestic cities or regions in terms of population flows, economic exchanges, space-time distance, and so on.

In summary, China is to be confronted by unavoidable DF prevalence in the appearance of ever-increasing frequencies of imported epidemics and stronger intensities of occasional local transmission in the inland cities. It is feasible of the RF model to map the occasionality of local DF prevalence for the cities geographically restricted by the status and duration of their time windows. This study has improved our understanding of the severity of DF prevalence and its influencing factors across China, which would supply useful clues for the health authorities improving their intervention capacity on such disease.

Dengue fever

Random Forest

GBM

Gradient Boosting Machine

SVM

Support Vector Machine

ROC

Receiver Operating Characteristic

AUC

Area Under the Curve

SHAP

SHapley Additive exPlanations

Hu Line

Huhuanyong line

Q-H Line

Qin Mountain – Huai River line

Region I

the cities on the right side of Hu Line

Region II

the cities on the north side of Q-H Line

Region III

the cities on the south side of Q-H Line

NCI

Number of cities identified

NCA

Number of cities with actual epidemics

The coefficients of variances

Availability of data and materials

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Acknowledgements

Not applicable.

Funding

HY R received the financial support from the National Natural Science Foundation of China (Grant NO.42071136) and the National Key Research and Development Program of China (2023YFC2307502).

Author information

Authors and Affiliations

State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China

Hongyan Ren & Nankang Xu

College of Resources and Environment, University of Chinese Academy of Sciences, Beijing 100049, China

Nankang Xu

National Key Laboratory of Intelligent Tracking and Forecasting for Infectious Diseases, National Institute for Communicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention, Beijing 102206, China

Liang Lu

Contributions

HR and NX conceived and designed the experimental concept. LL helped collect data. HR and NX analyzed the data and drafted the paper. HR, NX and LL revised the manuscript. All authors read and approved the final manuscript.

Corresponding authors

Correspondence to Hongyan Ren ([email protected]).

Ethics declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Bhatt S, Gething PW, Brady OJ, Messina JP, Farlow AW, Moyes CL, et al. The global distribution and burden of dengue. Nature. 2013;496(7446):504–7.
Guzman MG, Gubler DJ, Izquierdo A, Martinez E, Halstead SB. Dengue infection. Nat Reviews Disease Primers. 2016;2:16055.
Barmak DH, Dorso CO, Otero M, Solari HG. Dengue epidemics and human mobility. Phys Rev E. 2011;84(1):011901.
Xiong Yiquan, Qing C. Epidemiology of dengue fever in China since 1978. J South Med Univ. 2014;34(12):1822–5. (In Chinese).
Lai S, Huang Z, Zhou H, Anders KL, Perkins TA, Yin W, et al. The changing epidemiology of dengue in China, 1990–2014: a descriptive analysis of 25 years of nationwide surveillance data. BMC Med. 2015;13:100.
Wang Y, Wang X, Liu X, Ren R, Zhou L, Li C, et al. Epidemiology of Imported Infectious Diseases, China, 2005–2016. Emerg Infect Dis. 2019;25(1):33–41.
Ren H, Ning W, Lu L, Zhuang D, Liu Q. Characterization of dengue epidemics in mainland China over the past decade. J Infect Developing Ctries. 2015;9(9):970–6.
Huang Xue Y, Ma Hong X, Wang Hai F, Du Yan H, Su J, Li Xing L, et al. Outbreak of Dengue Fever in Central China, 2013. Biomed Environ Sci. 2014;27(11):894–7.
Yue Y, Liu X, Xu M, Ren D, Liu Q. Epidemiological dynamics of dengue fever in mainland China, 2014–2018. Int J Infect Dis. 2019;86:82–93.
Watts MJ, Kotsila P, Mortyn PG, Monteys VSI, Brancati CU. Influence of socio-economic, demographic and climate factors on the regional distribution of dengue in the United States and Mexico. Int J Health Geogr. 2020;19(1):44.
Liu H, Huang X, Guo X, Cheng P, Wang H, Liu L, et al. Climate change and Aedes albopictus risks in China: current impact and future projection. Infect Dis Poverty. 2023;12(1):26.
Zheng X, Zhong D, He Y, Zhou G. Seasonality modeling of the distribution of Aedes albopictus in China based on climatic and environmental suitability. Infect Dis Poverty. 2019;8(1):98.
Pham HV, Doan HTM, Phan TTT, Minh NNT. Ecological factors associated with dengue fever in a central highlands Province, Vietnam. BMC Infect Dis. 2011;11:172.
Qi X, Wang Y, Li Y, Meng Y, Chen Q, Ma J, et al. The Effects of Socioeconomic and Environmental Factors on the Incidence of Dengue Fever in the Pearl River Delta, China, 2013. PLoS Negl Trop Dis. 2015;9(10):e0004159.
Akter R, Naish S, Hu W, Tong S. Socio-demographic, ecological factors and dengue infection trends in Australia. PLoS ONE. 2017;12(10):e0185551.
Yue Y, Liu X, Ren D, Wu H, Liu Q. Spatial Dynamics of Dengue Fever in Mainland China, 2019. Int J Environ Res Public Health. 2021;18(6):2855.
Zheng L, Ren H-Y, Shi R-H, Lu L. Spatiotemporal characteristics and primary influencing factors of typical dengue fever epidemics in China. Infect Dis Poverty. 2019;8:24.
Zhang FC, He JF, Peng J, Tang XP, Qin CF, Lu HZ, et al. Guidelines for diagnosis and treatment of dengue in China. Chin J Intern Med. 2018;57(9):642–8.
Vezzani D, Velázquez SM, Schweigmann N. Seasonal pattern of abundance of Aedes aegypti (Diptera: Culicidae) in Buenos Aires city, Argentina. Mem Inst Oswaldo Cruz. 2004;99(4):351–6.
Githeko AK, Lindsay SW, Confalonieri UE, Patz JA. Climate change and vector-borne diseases: a regional analysis. Bull World Health Organ. 2000;78(9):1136–47.
Laporta GZ, Potter AM, Oliveira JFA, Bourke BP, Pecor DB, Linton Y-M. Global Distribution of Aedes aegypt and Aedes albopictus in a Climate Change Scenario of Regional Rivalry. Insects. 2023;14(1):49.
Tun-Lin W, Burkot TR, Kay BH. Effects of temperature and larval diet on development rates and survival of the dengue vector Aedes aegypti in north Queensland, Australia. Med Vet Entomol. 2000;14(1):31–7.
Yang HM, Macoris MLG, Galvani KC, Andrighetti MTM, Wanderley DMV. Assessing the effects of temperature on dengue transmission. Epidemiol Infect. 2009;137(8):1179–87.
Monintja TCN, Arsin AA, Amiruddin R, Syafar M. Analysis of temperature and humidity on dengue hemorrhagic fever in Manado Municipality. Gac Sanit. 2021;35:330–S3.
Chen L, Hu Y, Pang Y, Huang Y, Liu T, Xu J. Temporal and Spatial Changes of Aedes albopictus Habitats in Zhejiang,Fujian and Guangdong Based on Climatic Factors. J Hangzhou Normal University(Natural Sci Edition). 2019;18(06):642–50. (In Chinese).
Xiao J, Liu T, Lin H, Zhu G, Zeng W, Li X, et al. Weather variables and the El Nino Southern Oscillation may drive the epidemics of dengue in Guangdong Province, China. Sci Total Environ. 2018;624:926–34.
Servadio JL, Rosenthal SR, Carlson L, Bauer C. Climate patterns and mosquito-borne disease outbreaks in South and Southeast Asia. J Infect Public Health. 2018;11(4):566–71.
Chan M, Johansson MA. The Incubation Periods of Dengue Viruses. PLoS ONE. 2012;7(11):e50972.
Mohammady M, Pourghasemi HR, Amiri M. Land subsidence susceptibility assessment using random forest machine learning algorithm. Environ Earth Sci. 2019;78(16):503.
Salim NAM, Wah YB, Reeves C, Smith M, Yaacob WFW, Mudin RN, et al. Prediction of dengue outbreak in Selangor Malaysia using machine learning techniques. Sci Rep. 2021;11(1):939.
Ding F, Fu J, Jiang D, Hao M, Lin G. Mapping the spatial distribution of Aedes aegypti and Aedes albopictus. Acta Trop. 2018;178:155–62.
Yunsheng W, Bingyan X, Fanghao W, Qiming X, Liangying D. Application of ROC curve analysis in evaluating the performance of alien species' potential distribution models. Biodivers Sci. 2007;15(4):365–72.
Vanagas G. Receiver operating characteristic curves and comparison of cardiac surgery risk stratification systems. Interact Cardiovasc Thorac Surg. 2004;3(2):319–22.
Greiner M, Pfeiffer D, Smith RD. Principles and practical application of the receiver-operating characteristic analysis for diagnostic tests. Prev Vet Med. 2000;45(1–2):23–41.
Lundberg SM, Lee S-I, editors. A Unified Approach to Interpreting Model Predictions. 31st Annual Conference on Neural Information Processing Systems (NIPS); 2017 Dec 04–09; Long Beach, CA2017.
Urista DV, Carrue DB, Otero I, Arrasate S, Quevedo-Tumailli VF, Gestal M, et al. Prediction of Antimalarial Drug-Decorated Nanoparticle Delivery Systems with Random Forest Models. Biology-Basel. 2020;9(8):198.
Kim Y, Kim Y. Explainable heat-related mortality with random forest and SHapley Additive exPlanations (SHAP) models. Sustainable Cities and Society. 2022;79:103677.
Cao Z, Liu T, Li X, Wang J, Lin H, Chen L, et al. Individual and Interactive Effects of Socio-Ecological Factors on Dengue Fever at Fine Spatial Scale: A Geographical Detector-Based Analysis. Int J Environ Res Public Health. 2017;14(7):795.
Lun X, Wang Y, Zhao C, Wu H, Zhu C, Ma D, et al. Epidemiological characteristics and temporal-spatial analysis of overseas imported dengue fever cases in outbreak provinces of China, 2005–2019. Infect Dis Poverty. 2022;11(1):12.
Sang S, Yin W, Bi P, Zhang H, Wang C, Liu X, et al. Predicting Local Dengue Transmission in Guangzhou, China, through the Influence of Imported Cases, Mosquito Density and Climate Variability. PLoS ONE. 2014;9(7):e102755.
Ma T, Heywood A, MacIntyre CR. Chinese travellers visiting friends and relatives - A review of infectious risks. Travel Med Infect Dis. 2015;13(4):285–94.
Alirol E, Getaz L, Stoll B, Chappuis F, Loutan L. Urbanisation and infectious diseases in a globalised world. Lancet Infect Dis. 2011;11(2):131–41.
Lai S, Johansson MA, Yin W, Wardrop NA, van Panhuis WG, Wesolowski A, et al. Seasonal and interannual risks of dengue introduction from South-East Asia into China, 2005–2015. PLoS Negl Trop Dis. 2018;12(11):e0006743.
Yue Y, Liu Q. Exploring Epidemiological Characteristics of Domestic Imported Dengue Fever in Mainland China, 2014–2018. Int J Environ Res Public Health. 2019;16(20):3901.
Ding J, Cheng C, Zhang W, Tian Y. The ideological origins and geographical demarcation significance of Hu Huanyong Line. Acta Geogr Sin. 2021;76(6):1317–33.
Zhang X, Yan B. Climate change and city size: the role of temperature difference in the spatial distribution of China's population. Environ Sci Pollut Res. 2022;29(54):82232–42.
Liu Q. Epidemic profile of vector-borne diseases and vector control strategies in the new era. Chin J Vector Biology Control. 2019;30(01):1–6. (In Chinese).
Du J, Zhang Y. Does One Belt One Road initiative promote Chinese overseas direct investment? China Econ Rev. 2018;47:189–205.
Zhu G, Liu T, Xiao J, Zhang B, Song T, Zhang Y, et al. Effects of human mobility, temperature and mosquito control on the spatiotemporal transmission of dengue. Sci Total Environ. 2019;651:969–78.

floatimage1.jpeg
Graphical abstract
additionalfile1.docx
Additional file 1: Fig.S1 The distributions of China at provincial-level administrative divisions. Table S1 The list of the potential driving factors. Fig.S2 Spatial distribution of the ratio of local cases to imported cases in the past two decades. Fig.S3 Distribution of the cities with different occurrences of DF epidemics. Fig.S4 The frequencies and intensities of local DF outbreaks in 2003 – 2022. Fig.S5 Time windows for the city-level mosquito vectors’ activities across China. Table S2 The AUC values derived from RF, GBM, SVM models. Fig.S6 The contributions of input variables in the model without regard for the time windows in the region I. S1.1 Analyses of relationships between inputs and local DF epidemics in the region I. Fig.S7 Relationships between 12 inputs and local DF epidemics by means of SHAP values on the city scale in the region III. S1.2Analyses of relationships between inputs and local DF epidemics in the region III. Table S1 The coefficients of variances (CV) of 12 factors.

Download PDF

Editorial decision: Major revision
05 Feb, 2024
Reviewers agreed at journal
05 Jan, 2024
Reviewers invited by journal
04 Jan, 2024
Editor assigned by journal
03 Jan, 2024
First submitted to journal
26 Dec, 2023

You are reading this latest preprint version

Mapping the occasionality of inevitable dengue fever prevalence in China

Status:

Version 1

Abstract

Figures

Highlights

Introduction

Methods

Data collection

Descriptive analysis on frequency and intensity of DF outbreaks in China

The Retrieval of the time windows for local DF epidemics

Identification of the dengue driving forces at the city level

Results

Current situation of the DF prevalence in China

Geographical distribution of the DF epidemic

Time windows for local DF prevalence

Relationships analysis within the time windows

Dominant influencing factors

Mapping the city-level risk of local DF prevalence

Discussion

Conclusion

Abbreviations

Declarations

References

Supplementary Files

Status:

Version 1