Current situation of DF epidemics in China

From 2003 to 2022, a total of 329 DF outbreaks occurred within 104 months with 98,560 DF cases in China; the number of DF outbreaks displayed an overall upward trend, while the numbers of both imported and local DF cases sharply decreased from 2020 to 2022 (Fig. 1a). The proportions of local DF cases were much greater than those of imported cases over 15 years, especially since 2012 (Fig. 1b). At the city level, the numbers of local DF cases in China were increasingly closely associated with those of imported cases from 2003 to 2012 (r = 0.23, P < 0.01), 2013 to 2022 (r = 0.44, P < 0.01), and 2003 to 2022 (r = 0.41, P < 0.01).

Fig. 1

Temporal variations in dengue fever epidemics in China during 2003–2022. a Yearly numbers of total, imported, and local DF cases; b Yearly proportions of imported and local DF cases to total cases, and the ratio of local cases to imported cases; c Yearly numbers of cities with different DF epidemics; d Yearly proportions of cities with different DF epidemics in the S-QH region. DF Dengue fever; S-QH region The southern region of the Q-H Line

Moreover, an increasing number of cities are affected by DF epidemics. Despite an acute decrease in DF between 2020 and 2022, the numbers of cities with imported or local epidemics, both types of epidemics, or only imported epidemics displayed similar temporal variations (Fig. 1c). In contrast, the number of cities with only local epidemics remained below 3, except for 8 cities in 2014 (Fig. 1c). Nevertheless, the number of cities experiencing local epidemics was significantly associated with the number of cities experiencing imported epidemics at the national level (r = 0.79, P < 0.01).

China’s DF epidemic also displayed obvious temporal variations in frequency and intensity. As illustrated in Fig. 2a, the frequency of imported DF epidemics was less than 9 months per year before 2006 and then increased rapidly to 11 months per year or more since 2007, with a decrease to 9 months in 2021–2022. In comparison, the frequency of local epidemics exhibited a relatively slow and fluctuating uptrend, varying from 0 to 8. However, in terms of intensity, local DF epidemics were more severe (from 14 to 6641 cases per month) than imported DF epidemics (from one to 496 cases per month), despite the latter displaying a steadier and quicker increase (Fig. 2b). In addition, another frequency (i.e., the times of local DF outbreaks per year) presented a clear uptrend, and its intensity (i.e., the number of local DF cases per time) has remained higher than 100 cases per time since 2012 (Fig. 2c). These analyses showed that local DF epidemics in China have become increasingly severe over time as the number of imported DF cases has increased.

Fig. 2

Temporal variations in the frequency and intensity of dengue fever epidemics in China from 2003 to 2022. a The frequencies of imported DF epidemics and local DF epidemics; b The intensities of imported DF epidemics and local DF epidemics; c Temporal variations in the frequency and intensity of local DF outbreaks. DF Dengue fever

Geographical distribution of the DF epidemic

The distributions of the cities with DF epidemics were spatially featured across China. As illustrated in Additional file 1: (Fig. S4), the vast majority (96–100%) of the cities with imported or local epidemics were distributed in the E–H region. Among them, more than 75% were located in the S-QH region and displayed a downward trend from 2012 (Fig. 1d), implying that China’s DF epidemics were expanding toward the N-QH region. Moreover, the overwhelming majority of the cities with both imported and local epidemics were located in the S-QH region (Fig. 1d). In comparison, most of the cities with only imported epidemics were located in the E–H region, while those with only local epidemics were much fewer.

Moreover, the frequency and intensity of imported and local DF epidemics presented geographical disparities at the city level. The cities with an average frequency of imported epidemics greater than one month per year were mainly distributed in the E–H region (Additional file 1: Fig. S5a). A few cities with an average frequency much greater than 1.5 months per year were sparsely located, especially in the S-QH region. In contrast, fewer cities had an average frequency of local epidemics greater than 0.5 months per year and were mainly concentrated in the S-QH region (Fig. 3a). Similarly, the cities with a relatively high intensity of imported epidemics were mainly located in the S-QH region (Additional file 1: Fig. S5b), whereas those with a relatively high intensity of local epidemics were sparsely distributed in this region (Fig. 3b). Notably, provincial capitals often had a much higher frequency and intensity of imported epidemics, and some inland cities, such as Ji’an, Yichun, Chongqing, and Hangzhou, had the highest intensity of local epidemics despite a relatively lower frequency (Fig. 3b).

Fig. 3

The distribution of the frequency and intensity of local dengue fever epidemics at the city level during 2003–2022. a The frequency; b The intensity. Hu Line The Huhuanyong Line; Q-H Line The Qin Mountain-Huai River Line. Map approval No.: GS (2024)2127

Time windows for local DF epidemics

The city-level time windows for local DF transmission exhibited obvious spatial differences on either side of the Hu Line and Q-H Line (Additional file 1: Fig. S7). No time windows were observed in most of the cities in the W–H region throughout the year, whereas the cities in the E–H region presented various durations of time windows for local DF transmission. Furthermore, the city-level time windows in the E–H region exhibited significant geographical variation when compared to the S-QH and N-QH regions because of differences in their beginning and ending months as well as durations (Fig. 4a). In the N-QH region, the time windows often opened relatively later (beginning in June–July) and closed earlier (ending in September–October), and a few cities occasionally closed their time windows in some years (e.g., 2003–2004, 2006–2007, 2009–2011, and 2014–2016). In contrast, the time windows in the S-QH region began earlier (April–May) and ended later (October–November). As a result, these two regions exhibited long (seven or eight months in the S-QH region) or short (three or four months in the N-QH region) durations of time windows. Similarly, the city-level time windows for mosquito vector activity were also obviously spatially differentiated between the S-QH and N-QH regions due to the variations in their durations (Fig. 4b). Regarding the match degree, the time windows for local DF transmission were much greater (95.74%) than those for mosquito vector activity (83.59%), although the time windows for mosquito vector activity displayed similar spatial differences at the city level (Additional file 1: Fig. S6).

Fig. 4

The durations from the beginning to the ending months of time windows in the S-QH and N-QH regions during 2003–2022. a Average beginning and ending months of time windows for local DF transmission; (b) Average beginning and ending months of time windows for mosquito vector activity. DF Dengue fever; S-QH region The southern region of the Q-H Line; N-QH region The northern region of the Q-H Line; Hu Line The Huhuanyong Line; Q-H Line The Qin Mountain-Huai River Line

Analysis of the relationships among the time windows

Within the time windows for local DF transmission in the E–H region, the occurrence of local DF epidemics showed a significant positive correlation with imported cases (r = 0.42, P < 0.01) and most of the natural and socioeconomic factors (Table 3), except for Cropland (r = -0.14, P < 0.01). In contrast, the relationships observed in the S-QH and N-QH regions differed. Local DF epidemics were significantly negatively associated with Forest (r = -0.09) in the S-Q region and positively associated with only three factors in the N-Q region, namely, imported cases (r = 0.12), GDP (r = 0.08), and Pop (r = 0.08), at a significance level of 0.01.

Table 3 Correlation coefficients between the occurrence of local dengue fever epidemics and potential influencing factors

Furthermore, RF models were utilized to explore the relationships between the occurrence of local DF epidemics and influencing factors at the city level because of their superior ability beyond that of the GBM and SVM models (Additional file 1: Table S6). According to the higher AUC values in Table 4, the RF models in the E–H region (Models 1, 2, and 3) performed considerably better than the other models, although Model 4 also showed good performance in the S-QH region (AUC = 0.85). Moreover, the AUC slightly increased from 0.90 (Model 3) to 0.92 (Models 1 and 2) when considering the time windows for either local DF transmission or mosquito vector activity in the E–H region. These results showed that RF models considering time windows could be rationally applied to explore the contributions of factors and predict the city-level risk of local DF epidemics.

Table 4 The AUC values derived from the random forest modelsDominant influencing factors of local DF epidemics

Among the influencing factors, imported cases contributed the most to local DF epidemics in the E–H (24.82%, Fig. 5a) and S-QH (31.01%, Fig. 5b) regions. Moreover, five variables associated with much greater contributions included Tmin (16.88%), Forest (8.58%), Pop (7.36%), Prec (7.28%), and Tmean (7.14%) in the E–H region, which differed from those in the S-QH region (GDP, Pop, RH, Cropland, and Forest). In addition, natural factors (i.e., five meteorological elements) were associated with greater contributions (41.04%) than socioeconomic factors (34.15%) in the E–H region. In comparison, socioeconomic factors were associated with greater contributions (44.26%) than natural conditions (24.74%) in the S-QH region.

Fig. 5

Contributions of the input variables based on the global SHAP values in the model considering the time windows for local dengue fever transmission. a The contributions of input variables in the E–H region; b The contributions of input variables in the S-QH region. Imported cases Number of imported cases; Tmin Average monthly minimum air temperature; Forest Annual average percentage of forest; Pop Population density; Prec Average monthly precipitation; Tmean Average monthly mean air temperature; GDP Gross domestic product; RH Average monthly mean relative humidity; Cropland Annual average percentage of cropland; Tmax Average monthly maximum air temperature; Impervious Annual average percentage of impervious; Water Annual average percentage of water; E–H region The eastern region of the Hu Line; S-QH region: The southern region of the Q-H Line; SHAP: The Shapley Additive exPlanations

According to the SHAP values derived from Model 1 (Fig. 6), the imported cases and the other 11 factors had complex impacts (protective or risk effects) on the occurrence of local DF epidemics at the city level. In the E–H region, the promoting effect of imported cases had a strengthening trend as the number of imported cases increased (Fig. 6a). In contrast, the other 11 factors generally presented composite promoting and inhibiting effects. These factors were categorized into two groups: Group 1 (i.e., Tmin, Forest, Pop, Prec, GDP, Impervious, and Water), which exhibited inhibiting effects before promoting effects, and Group 2 (i.e., Tmean, RH, Cropland, and Tmax), which exhibited promoting effects before inhibiting effects. Among the top five factors (Fig. 5a) derived from the contribution analyses, Tmin, Forest, Pop, and Prec belonged to Group 1, with respective inflection values of 20 °C (Fig. 6b), 60% (Fig. 6c), 1000 persons per square kilometer (Fig. 6d), and 185 millimeters (Fig. 6e). In contrast, Tmean belonged to Group 2, and its promoting effects became inhibitory at 21 °C (Fig. 6f). Similarly, the other six factors were categorized into Group 1 (GDP, Impervious, and Water) and Group 2 (RH, Cropland, and Tmax). In comparison, the factors in the S-QH region had similar composite impacts on the occurrence of local DF epidemics in this region (Additional file 1: Fig. S8), although their inflection values differed slightly from those in the E–H region. These analyses showed that the SHAP values were more reasonable than the Spearman correlation coefficients for interpreting the influences of potential factors on local epidemics. In addition, the fact that imported cases triggered local epidemics was further validated.

Fig. 6

Relationships between 12 inputs and local dengue fever occurrence according to local SHAP values at the city level in the E–H region. a Number of imported cases (Imported cases); b Average monthly minimum air temperature (Tmin); c Annual average percentage of forest (Forest); d Population density (Pop); e Average monthly precipitation (Prec); f Average monthly mean air temperature (Tmean); g Gross domestic product (GDP); h Average monthly mean relative humidity (RH); i Annual average percentage of cropland (Cropland); j Average monthly maximum air temperature (Tmax); k Annual average percentage of impervious (Impervious); l Annual average percentage of water (Water). E–H region The eastern region of the Hu Line; SHAP The Shapley Additive exPlanations

Predicting the risk of local DF epidemics at the city level

Due to its good performance (AUC = 0.91), Model 1 was utilized to predict the occurrence probability of local DF epidemics at the city level in the E–H region. In this region, a total of 295 cities had various occurrence probabilities of local DF epidemics and were then categorized into five groups (Table 5). Among them, 20 cities in the first group (0.8 < probability < 1.0) were indeed affected by local DF epidemics in 2019, demonstrating the highest hitting ratio (100%). Furthermore, the hitting ratios of the other four groups (i.e., 0.6–1.0, 0.4–1.0, 0.2–1.0, and 0.0–1.0) gradually decreased to 95.56%, 79.41%, 67.65%, and 29.15%, respectively. In comparison, the hitting ratios of Model 1 clearly declined during 2020–2022, although only one city (Guangzhou) in 2020 had the highest probability of occurrence of local DF epidemics.

Table 5 The ability of Model 1 to predict the occurrence probability of local dengue fever epidemics at the city level in the E–H region

The cities at various risk levels for local DF epidemics displayed spatial disparities in the E–H region from 2019 to 2022. Most of the cities with the highest risk (20 cities) were located in the southeastern coastal areas (16 cities). Moreover, 25 high-risk and 23 moderate-risk cities were identified in the southeastern and central regions, respectively (Fig. 7). Thus, the above 68 cities were primarily concentrated in the S-QH region. During 2020–2022, the cities at various risk levels were also mainly located in this region (Additional file 1: Fig. S9). These results showed that the risk of local DF epidemics in the E–H region was appropriately predicted at the city level using RF models considering the time windows for local DF transmission.

Fig. 7

The city-level risk of local dengue fever epidemics in 2019 in China. Hu Line The Huhuanyong Line; Q-H Line The Qin Mountain-Huai River Line. Map approval No.: GS (2024) 2127