Spatiotemporal trends of hemorrhagic fever with renal syndrome (HFRS) in China under climate variation

Significance

The fatality rate of hemorrhagic fever with renal syndrome (HFRS) ranges from less than 1 to 15%, with China reporting around 90% of global cases, though it is also found in Europe and Asia. To predict future HFRS trends, we have developed a statistical model based on high-resolution climate and population data. Our findings show that HFRS will remain a severe risk to mainland China throughout the entire 21st century and that Rattus norvegicus is becoming the most active host. Our work is a nationwide city-level study to predict the spatiotemporal trends of HFRS in mainland China through the end of this century; the results will help to inform policymaking on its future prevention.

Abstract

Hemorrhagic fever with renal syndrome (HFRS) is a zoonotic disease caused by the rodent-transmitted orthohantaviruses (HVs), with China possessing the most cases globally. The virus hosts in China are Apodemus agrarius and Rattus norvegicus, and the disease spread is strongly influenced by global climate dynamics. To assess and predict the spatiotemporal trends of HFRS from 2005 to 2098, we collected historical HFRS data in mainland China (2005–2020), historical and projected climate and population data (2005–2098), and spatial variables including biotic, environmental, topographical, and socioeconomic. Spatiotemporal predictions and mapping were conducted under 27 scenarios incorporating multiple integrated representative concentration pathway models and population scenarios. We identify the type of magistral HVs host species as the best spatial division, including four region categories. Seven extreme climate indices associated with temperature and precipitation have been pinpointed as key factors affecting the trends of HFRS. Our predictions indicate that annual HFRS cases will increase significantly in 62 of 356 cities in mainland China. Rattus regions are predicted to be the most active, surpassing Apodemus and Mixed regions. Eighty cities are identified as at severe risk level for HFRS, each with over 50 reported cases annually, including 22 new cities primarily located in East China and Rattus regions after 2020, while 6 others develop new risk. Our results suggest that the risk of HFRS will remain high through the end of this century, with Rattus norvegicus being the most active host, and that extreme climate indices are significant risk factors. Our findings can inform evidence-based policymaking regarding future risk of HFRS.

Hemorrhagic fever with renal syndrome (HFRS) is a rodent-borne zoonotic disease characterized by high fever, renal insufficiency, and hemorrhage (1, 2). HFRS annually affects 60,000 to 100,000 cases in over 70 countries, primarily in Europe and Asia (2–4). Mainland China accounts for nearly 90% of cases worldwide, with 576,361 HFRS cases from 1995 to 2020 (1, 4–9). Over 1.2 billion Chinese residents live in provinces with documented HFRS cases (10). HFRS is categorized as a category B infectious disease in Chinese law, mandating the prompt reporting of diagnosed cases (9, 11). From 1950 to 2010, approximately 1.4 million HFRS cases were reported in China (12), and even during the COVID-19 pandemic in 2021, the disease persisted, with 2,657 reported cases (13). In China, HFRS is primarily attributed to orthohantaviruses (hereinafter termed as HVs), which are carried by Rattus norvegicus (Seoul virus, SEOV) and Apodemus agrarius (Hantaan virus, HTNV), with cases involving Rattus norvegicus peaking in spring, and those associated with Apodemus agrarius being more prevalent in autumn and winter (14–20). Beyond China, the rodent species responsible for HFRS transmission vary by region (17). Human exposure primarily occurs through contact with infected rodents or inhalation of aerosols and secretions (12). The fatality rate for HFRS varies with the virus strain, ranging from less than 1 to 15% (21). On average, it’s around 5%, with the HTNV showing a higher range of 5 to 15%, while the SEOV has a lower range, typically between 1 and 2% (22, 23). Despite the implementation of integrated intervention measures, such as rodent control, environmental management, and vaccination, which have contributed to a reduction in HFRS incidence in most of China (1, 10), the number of cases remains significant (2). Notably, HFRS has continuously presented or re-emerged in some epidemic areas, including in some metropolises and provincial capital cities (1, 10, 13). However, several spatiotemporal analyses predominantly center around historical cases to elucidate risk factors (2, 5, 14, 24). Additionally, previous prediction analyses in HFRS have incorporated time-series analysis, often with a focus on specific areas (25, 26). Hence, it is imperative to forecast future spatiotemporal dynamics of HFRS cases, facilitating the development of targeted strategies to mitigate its spread, particularly in the post-COVID-19 era. Prediction of these dynamic patterns is essential for effectively reducing the impact of HFRS and addressing this significant public health challenge in China, which affects health, the economy, and society.

HFRS trend is a multifaceted outcome shaped by a complex interplay of biotic and host factors, climatic conditions, terrain heterogeneity, and socioeconomic variables. The global climate dynamic significantly affects HFRS epidemiology, where temperature and precipitation are pivotal determinants of disease patterns (7, 12, 27–29). Temperature and humidity have a dual impact, positively influencing HFRS outbreaks, while extreme temperatures, either high or low, are detrimental (12, 28). Additionally, relative humidity not only reflects regional rainfall and temperature levels but also affects the infectivity and stability of hantavirus in the environment, with mountainous regions in China’s humid climates demonstrating the highest incidence of HFRS (12, 24, 30). Concurrently, human activity patterns are subject to the influence of weather conditions and seasonal fluctuations, exerting a direct impact on interactions between humans and rodents (9). Moreover, there is a positive correlation between human population density and the outbreak of HFRS epidemics (31), and increased urbanization rates and higher gross domestic product (GDP) have been linked to elevated HFRS incidence (12). A previous study demonstrated that socioeconomic conditions from the same months are among the influencing factors of the monthly trends in HFRS cases (12). Amid the backdrop of global climate change, gaining an exhaustive comprehension of the intricate interplay between climatic variables and multifaceted determinants influencing the trends of HFRS in China is of paramount importance for effective HFRS mitigation. However, it is worth noting that prior investigations have not undertaken a systematic or dynamic quantification of the effects of changes in climatic factors and population size. Hence, our research aims to systematically and dynamically assess these variables, thereby advancing our scientific understanding of this complex disease.

In this ground-breaking study, we present a comprehensive assessment and prediction of HFRS trends across mainland China, spanning from 2005 to 2098. Our analysis harnesses high-resolution global climate data at a 25-km scale and utilizes recent population data derived from the latest birth rate trends. To address the complexities associated with strong spatial heterogeneity, high correlations among climate variables, and issues of severe zero-inflation and spatiotemporal autocorrelation in HFRS case data, we employed a robust methodology. Generalized additive models (GAMs), conditional minimum average variance estimation (MAVE), and integrated nested Laplace approximation (INLA) were employed. Finally, the primary aim of this research is to provide spatiotemporal prediction of HFRS trends at the municipal level, considering various climate scenarios that extend to the close of the 21st century. This predictive approach is underpinned by the precise identification of HFRS endemic regions and seeks to elucidate high-risk factors arising from the interplay of biotic factors, climate change, and population dynamics. The study aspires to offer a multidimensional perspective on future HFRS risks linked to climate change in China, making a significant contribution to the scientific understanding of this disease.

Results

HFRS Cases and Infection Status, 2005–2020.

From 2005 to 2020, mainland China reported a total of 184,633 cases of HFRS, with the majority concentrated in cities located in Northwest, Northeast, North, and East China (SI Appendix, Fig. S1). The degree of HFRS risk in each of the 356 cities was assessed based on the annual case count. Among these cities, 106 with fewer than one annual case were categorized as having no evident HFRS risk. The remaining 250 cities were grouped into four risk levels: minor risk (1 to 4 cases, 0–25th percentile), low risk (4 to 16 cases, 25–50th percentile), medium risk (16 to 50 cases, 50–75th percentile), and severe risk (more than 50 cases, 75–100th percentile). Cities with a severe HFRS risk were predominantly located in East (17 cities, e.g., Jinan and Tsingtao) and Northeast China (27 cities, e.g., Changchun and Shenyang) (Fig. 1A and SI Appendix, Table S1). Overall, there was a gradual decrease in the number of HFRS cases across mainland China from 2005 to 2020, although this decline was not statistically significant (P > 0.05). Notably, there was a distinct peak in HFRS cases observed between 2012 and 2018.

Fig. 1.

Region division according to the type of magistral HVs host and distribution of cities among risk levels. (A) Based on the seasonal regularity of HFRS cases in China, three endemic region categories can be identified (Apodemus, Rattus, and Mixed). The risk is distinguished by the number of annual cases in the cities. Pictures of Apodemus agrarius and Rattus norvegicus were kindly provided by Dawei Wang. (B) Cases reported from the Apodemus-associated regions (blue bars) primarily occurred in the autumn and winter, while those from Rattus-associated regions (yellow bars) occurred in the spring. In Mixed regions (green bars), cases have been reported in spring, autumn, and winter. (C) Cities in the severe-risk category (75–100th percentile) account for the largest share of HFRS cases in each of the three endemic region categories. The seven geographical regions labeled with Roman numerals are also shown: I, North China; II, Northeast China; III, East China; IV, Central China; V, South China; VI, Southwest China; VII, Northwest China.

Association of HFRS Cases with the Type of Magistral HVs Host.

Within our analysis of 11 spatial variables using the fitted GAM model (Table 1), the most significant factor explaining the geographic variation in HFRS cases was the type of host responsible for magistral HVs, accounting for 23.68% of the deviance. Consequently, the 356 cities under investigation were categorized into four distinct region types (Fig. 1A): Apodemus-associated endemic regions (referred to as “Apodemus regions”), Rattus-associated endemic regions (“Rattus regions”), regions where both Apodemus and Rattus were associated with endemicity (“Mixed regions”), and regions where the host was unspecified (“Unspecified regions”). These regions exhibited varying total HFRS cases, amounting to 67,021, 38,944, 76,643, and 2,025, respectively (SI Appendix, Table S1). HFRS cases in these four types of endemic regions displayed distinct seasonal patterns (Fig. 1B). In terms of the severity of HFRS risk, the largest proportions of cities with severe risk (defined as having ≥50 annual cases) were identified in the Mixed and Apodemus regions. Collectively, including the severe-risk cities in the Rattus region, these areas contributed to the majority of annual HFRS cases, amounting to 77.8% (Fig. 1C).

Table 1.

Diagnostics for models generated with different spatial variables

Spatial variable	Model formula	ΔUBRE	ΔPD
Type of endemic region	$\begin{array}{c}\hfill \log \left[E\left(Y_{\mathit{it}}\right)\right]=\alpha |r+s_1\left(T_{\mathit{it}}\right)|r+s_2\left(P_{\mathit{it}}\right)|r+\epsilon \hfill \end{array}$	–	–
Climate zone		3,121.44	−2.40
Number of hantavirus genotypes		6,063.88	−4.14
Geomorphic type		7,757.62	−6.46
Infectious disease incidence classification		9,083.60	−7.06
HAQ index classification		11,877.94	−9.49
Ecological zone		12,336.52	−8.25
Economic development zone		12,573.17	−10.01
Impervious surface area ratio classification		13,876.97	−9.47
GDP classification		14,889.39	−11.22
Three-step terrain classification		21,490.81	−14.82
No spatial variable	$\log \left[E\left(Y_{\mathit{it}}\right)\right]=\alpha +s_1\left(T_{\mathit{it}}\right)+s_2\left(P_{\mathit{it}}\right)+\epsilon$	26,318.85	−18.01

HAQ = healthcare access and quality index; GDP = gross domestic product; UBRE = Un-Biased Risk Estimator, which is a scaled Akaike information criterion; PD = percentage of deviance explained, which is a generalisation for a commonly used coefficient of determination; ΔUBRE and ΔPD represent the change in the value of UBRE and PD respectively in the GAM model with a region-specific variable (other than endemic region type) relative to the model with endemic region type alone.

Evaluation of the Effects of Key Climate Variables and Population Size.

Using the MAVE method, we effectively reduced 21 climate variables into a seven-dimensional space, as this configuration yielded the lowest CV(d) value (0.0683), capturing 93.2% of the original information from the 21 climate variables during the 2005–2020 period (Table 2 and SI Appendix, Fig. S4). Within these seven dimensions, we identified seven extreme climate indices as key variables influencing HFRS cases. These included four extreme temperature-related indices (Frost days, Summer days, Ice days, and Tropical nights) and three extreme precipitation-related indices (Number of very heavy rain days, Maximum length of wet spell, and Simple precipitation intensity index). Of these variables, Tropical nights, Summer days, and Simple precipitation intensity index are expected to increase in the future, while Frost days and Ice days are anticipated to decrease. The other two variables, number of very heavy rain days and maximum length of wet spell, display no significant trend (P > 0.05). Furthermore, the analysis of population size revealed varying trends over two time periods between 2021 and 2098. Initially, there is a minor upward trend until 2030, followed by a marked decrease thereafter (SI Appendix, Fig. S5).

Table 2.

Absolute values of coefficients for the selected MAVE dimensions

Climate variables	X1	X2	X3	X4	X5	X6	X7
Temperature variables
Frost days	0.216*	0.216	0.755*	0.443*	0.455*	0.674*	0.237
Summer days	0.167	0.181	0.362*	0.810*	0.114	0.511*	0.042
Tropical nights	0.892*	0.786*	0.143	0.350*	0.452	0.095	0.086
Ice days	0.255*	0.399*	0.346*	0.020	0.157	0.244	0.137
Temperature	0.030	0.000	0.013	0.008	0.031	0.009	0.022
Maximum air temperature	0.017	0.006	0.010	0.001	0.013	0.001	0.011
Minimum air temperature	0.017	0.000	0.010	0.010	0.007	0.013	0.018
Number of cold days	0.003	0.003	0.006	0.003	0.008	0.006	0.001
Number of cool days	0.002	0.005	0.003	0.001	0.007	0.016	0.004
Number of warm nights	0.005	0.002	0.010	0.002	0.017	0.022	0.010
Number of hot days	0.004	0.001	0.003	0.002	0.005	0.002	0.003
Warm spell duration index	0.013	0.007	0.000	0.002	0.017	0.044	0.030
Cold spell duration index	0.020	0.008	0.010	0.015	0.009	0.030	0.007
Precipitation variables
Number of very heavy rain days	0.189	0.168	0.015	0.068	0.511*	0.342*	0.744*
Simple precipitation intensity index	0.145	0.044	0.063	0.103	0.205	0.082	0.342*
Maximum length of wet spell	0.061	0.328*	0.015	0.078	0.493*	0.242	0.484*
Precipitation	0.018	0.024	0.038	0.014	0.001	0.067	0.012
Total annual very heavy rain days	0.023	0.051	0.004	0.049	0.046	0.110	0.021
Maximum length of dry spell	0.018	0.023	0.039	0.002	0.011	0.054	0.078
Max 1-day precipitation	0.019	0.026	0.020	0.009	0.011	0.046	0.046
Max 5-day precipitation	0.025	0.037	0.033	0.004	0.004	0.093	0.001

Details of the climate variables can be found in SI Appendix, Table S2. “X” refers to the dimensions in MAVE. “*” indicates the top three variables with the greatest absolute coefficients and hence the largest contributions in each respective dimension.

Spatiotemporal Prediction of HFRS Cases, 2021–2098.

Using the INLA model, we have created dynamic city-level maps projecting HFRS cases across mainland China from 2021 to the close of the century. This model integrates various factors, attributing 3.1% to fixed effects such as endemic regions and population size, while treating climate factors (47.3%), spatial components (35.3%), and the temporal component (14.4%) as random effects (SI Appendix, Fig. S19). The analysis indicates an overall stability in HFRS cases for the 2021–2098 period within the 250 cities examined. However, it notably forecasts a significant increase in annual HFRS cases for 62 of these cities (with β > 0, P < 0.05, and annual cases ≥ 1, as detailed in SI Appendix, Table S1 and Fig. S22). To maintain precision, cities with fewer than 1 annual case were excluded from this projection. Of the cities predicted to experience an increase in HFRS cases, 27 are in the Rattus regions, 15 in Apodemus regions, and 14 in Mixed regions. Moreover, the model anticipates significant declines in HFRS cases for 53 cities (with β < 0 and P < 0.05, as detailed in SI Appendix, Table S1). Interestingly, many cities exhibit distinct seasonal fluctuations in HFRS cases (SI Appendix, Fig. S20). Importantly, the spatial distribution of projected annual cases between 2021 and 2098 closely resembles the observed pattern from 2005 to 2020 (SI Appendix, Figs. S1 and S21).

Rattus Regions Are Predicted to be the most Active.

The Rattus regions are collectively foreseen as the most active areas for future HFRS cases, characterized by the highest growth rate and a statistically significant increasing trend (β = 5.35 cases/year, P < 0.001) (Fig. 2A). In contrast, while the Apodemus regions are still anticipated to maintain their status as the primary endemic areas, with a higher annual case count compared to the Rattus regions, there is no significant projected future trend (β = 0.97 cases/year, P > 0.05), and a similar nonsignificant trend is seen in the Mixed regions (β = −2.01 cases/year, P > 0.05). Additionally, the most obvious rise in HFRS cases is expected to occur in severe-risk cities in the Rattus region (Fig. 2B).

Fig. 2.

Trend and change rate of HFRS cases in the cities of each endemic region category and the five risk levels. (A) Rattus regions (β = 5.35 cases/year, P < 0.001) will be the most concerning compared to Apodemus regions (β = 0.97 cases/year, P > 0.05) and Mixed regions (β = −2.01 cases/year, P > 0.05). The shaded area is the 95% CI. (B) The rate of change is calculated as follows: (annual average cases from 2021 to 2098 − annual average cases from 2005 to 2020)/annual average cases from 2005 to 2020. Increases in the total HFRS cases of severe cities are seen in all three endemic region categories, with the most remarkable rise in Rattus regions.

More Cities Are Predicted to Possess Severe Risk.

Our projections for future HFRS risk in mainland China indicate that 254 cities will remain at risk. Significantly, our research reveals that 80 cities will face severe HFRS risk (Fig. 3), surpassing the initial count of 62 cities. In contrast to the period from 2005 to 2020, the geographic distribution of cities with severe HFRS risk exhibited significant changes. East China saw an increase in affected cities, rising from 17 to 29, while Central China also experienced a surge, escalating from 7 to 12 cities. In contrast, Northeast China observed decreasing from 27 to 24 cities with severe HFRS risk. The most pronounced increase in severe HFRS risk has been observed in the Rattus regions, where the number of cities has risen from 13 to 22, marking a substantial 40.9% growth. In a similar vein, Apodemus regions have experienced an increase from 19 to 21 cities, representing a noteworthy 10.5% surge, while Mixed regions have seen a rise from 30 to 37 cities, reflecting a 23.3% increase (Figs. 1A and 3A). Next, our comprehensive analysis examines the dynamics of risk-level transitions, revealing that 50 cities have transitioned to a higher-risk status, with their distribution spanning all regions of China (Fig. 3A). Significantly, 22 cities exhibit a dual classification, concurrently falling into both the severe-risk and higher-risk categories, which we will hereinafter refer to as “emerging severe-risk cities” (Fig. 3A). Among this subset, 12 cities, representing 54.5% of the total, are primarily situated in East China and are evenly distributed between the Rattus region (six cities) and the Mixed region (six cities). The remaining 10 cities are dispersed across five distinct regions in China, encompassing six cities in the Rattus region, six cities in the Mixed region, and two cities in the Apodemus region. Notably, a total of 12 cities among the emerging severe-risk cities are located within the Rattus region. Additionally, our analysis anticipates that cities categorized as medium, low, and minor risk levels (nonsevere) will either decrease or remain stable when compared to their initial counts spanning the period from 2005 to 2020 (Fig. 3B). Furthermore, 12 cities are predicted to experience a reduction in their future risk levels, and these cities are distributed throughout mainland China. Interestingly, there are six cities that had no HFRS risk from 2005 to 2020 but are expected to develop HFRS risk in the future. (hereinafter labeled as emerging risk cities, Fig. 3A). These emerging risk cities are distributed across one in the North, two in the East, two in the Southwest, and one in Northwest China. Simultaneously, two cities have transitioned to a no HFRS risk.

Fig. 3.

Severe and rising HFRS risk cities and changes in the risk level during 2021-2098. (A) Risks are predicted to rise in 50 cities, and 80 cities are predicted to belong to the severe HFRS risk category from 2021 to 2098. (B) Twenty-two cities will escalate from medium to severe risk, and 58 cities will remain in the severe-risk category, adding up to 80 cities with severe HFRS risk. Meanwhile, 13 cities will escalate from low to medium risk, 9 from minor to low risk, and 6 from no risk to minor risk. The seven geographical regions labeled with Roman numerals are also shown: I, North China; II, Northeast China; III, East China; IV, Central China; V, South China; VI, Southwest China; VII, Northwest China.

Discussion

The presented study is a nationwide study to predict the spatiotemporal trends of HFRS in mainland China through the end of this century and used high-resolution climate and population data to assess the impacts of multiple factors. We produced city-level, temporally dynamic maps of HFRS cases across mainland China, covering the period from 2021 through 2098. Comprehending the intricate relationship between climatic variables and factors affecting HFRS cases is of significant scientific importance in the context of global climate change. The study pinpointed HVs host species as crucial in explaining the varied distribution of HFRS in China, identifying three endemic region categories and an unspecified region category. It emphasized the Apodemus and Rattus-associated regions as key to understanding HFRS spatial differences, offering insights into factors affecting HFRS trends. Spatiotemporal analysis indicates a significant increase in HFRS cases from 2021 to 2098, with approximately a quarter of cities reaching severe-risk levels, and 22 cities at emerging severe-risk. The Rattus region is predicted to replace the Apodemus region as the most active endemic region. To ensure the accuracy of the prediction model, we established calibration data and training data in the historical HFRS model, and our results fit well with actual data. These findings highlight the significance of quantifying spatiotemporal effects, particularly in the context of global climate change, for the effective prevention and control of HFRS in China.

Our research broadens the scope from previous provincial-level HFRS studies, aiming to provide a national-level exploration of HFRS in China. Previous studies have often relied on geographical information systems (GIS) and spatial autocorrelation analyses (9, 30, 32–34). These studies also utilized models such as Seasonal Difference Space-Time Autoregressive Integrated Moving Average (SD-STARIMA) and seasonal autoregressive fractionally integrated moving average (SARFIMA), which encountered difficulties in reconciling temporal and spatial aspects (25, 26). In our study, we employed the INLA model, which efficiently harmonizes both dimensions, adeptly addressing challenges in HFRS case data (35). Our findings provide compelling evidence of an imminent surge in HFRS cases, particularly in regions predominantly inhabited by Rattus host species. This projection encompasses 27 scenarios, revealing regional disparities in HFRS risk. Emerging risk cities are spread across various regions, with substantial increases concentrated in Rattus regions. These findings emphasize the significance of our rodent-city classification system for HFRS prediction in China. Our analysis forecasts changes in HFRS risk across China: 62 cities are on an increasing trend, while 80, primarily in Northeast, East, and Central China, are categorized as severe-risk areas. A notable escalation to higher risk levels is observed in 50 cities, with 22 experiencing both severe and higher-risk categories, as emerging severe-risk cities, mainly in East China. Additionally, six cities in North, East, Southwest, and Northwest China have been marked as emerging HFRS risk cities. Cities with Apodemus as the primary host are expected to maintain relatively stable HFRS trends. We stress the importance of focusing on emerging and severe cities due to heightened HFRS risk. In summary, our study reveals a marked shift in the geographical distribution of the disease, suggesting an elevated risk in areas characterized by a high prevalence of Rattus species. This necessitates targeted public health preparedness. Conversely, areas with Apodemus as the primary host species are expected to maintain relatively stable HFRS trends. These insights guide the development of precise public health strategies in response to the evolving landscape of this disease.

To assess HFRS spatial variation in China, we applied models incorporating biotic, environmental, and socioeconomic variables. This led to a host-based classification of cities for HFRS into Apodemus, Rattus, Mixed, and Unspecific regions. This approach combines observed seasonality in human HFRS cases with their association with host species, deepening our understanding of HFRS epidemiology (36). HFRS cases show distinct spatial and seasonal patterns. Apodemus (HTNV type) is mainly found in field habitats with a higher case fatality rate, while Rattus (SEOV type) is associated with urban areas, with their cases peaking in spring and having a lower case fatality rate (17, 22, 23). It is found that urbanization would contribute to ecologically favorable niches for Rattus and may elevate future HFRS risk in Rattus regions (37–39). However, Urbanization, socioeconomic shifts, and agricultural advancements significantly affect hemorrhagic fever with renal syndrome (HFRS), particularly in Apodemus and Rattus host systems (40–43). A study indicates that advancements in agricultural practices and public health measures have led to a notable decrease in rodent density, thereby altering the epidemiological landscape of HFRS (43). These findings underscore the necessity for further comprehensive studies to elucidate these complex relationships and their implications in HFRS management. Besides, climate change would bring about rising temperatures and shifting precipitation patterns, thus affecting the survival and abundance of rodent hosts (37, 38). While specific areas may experience shifts in HV host species (23, 24), as reported by Fang et al. regarding the transition from HTNV to SEOV in HFRS transmission in Shandong Province due to local climate factors (30), the overall HV host ranges across China are generally expected to remain stable throughout this century. Thus, maintaining a consistent nationwide rodent-city classification system is vital for predicting HFRS in China, aiding in the development of effective prevention and control strategies against the disease.

Our study emphasizes climate’s vital role in HFRS. Empirical evidence supports climate's influence on HFRS, including temperature, precipitation, and humidity (8, 9, 44, 45). Past research worldwide consistently highlights climatic variables, especially precipitation and humidity, affecting HFRS (24, 30, 46). As climate change intensifies extreme weather events, it’s expected to impact HFRS (9, 47). Our study systematically investigates the climatic factors affecting HFRS, utilizing the MAVE model to dissect intervariable correlations and provide insights into the climatic impacts. This comprehensive analysis emphasizes extreme climate indices and mean variables, identifying extreme temperature and precipitation as key risk factors, which aligns with existing research on HFRS dynamics (47, 48). This reveals the changing HFRS landscape in China due to climate change, stressing the need for tailored control and prevention strategies.

It is crucial to acknowledge the limitations of this study. First, our primary focus was on assessing the impact of climate change on projected HFRS trends. While we considered socioeconomic factors in regional divisions, we didn’t extensively explore their interactions with HFRS cases. Second, climate change can influence the distribution and population dynamics of the two host species, potentially altering endemic regions. Third, there may be underreporting of asymptomatic patients not seeking medical attention. Nevertheless, our study lays a foundational framework for anticipating future HFRS risk in mainland China.

Conclusion

This study evaluated the potential health risks associated with climate change and HFRS and provides a city-level prediction of HFRS trends in mainland China through the end of this century. Our predictions indicate that annual HFRS cases will increase significantly in 62 of 356 cities in mainland China. Rattus regions are predicted to be the most active, surpassing Apodemus and Mixed regions. Eighty cities, mainly in Northeast, East, and Central China, face severe HFRS risk, including 22 emerging severe-risk cities primarily located in East China and Rattus regions after 2020, while six others develop new risk from no risk. Meanwhile, the dynamics of extreme climate indices represent significant contributors to HFRS trends in the future. These findings provide policymakers with essential insights for developing effective prevention and control strategies.

Materials and Methods

Data.

This study collected data of HFRS cases reported between 2005 and 2020 from the National Centre for Disease Control and Prevention of China, which follow the same standard among Chinese cities according to the law (SI Appendix, Table S1) (11). The cases were observed in 327 cities across mainland China, with another 29 cities reporting no HFRS cases (for a total 356 cities). The spatial data associated with HFRS cases consisted of 11 variables in total (SI Appendix, Figs. S6–S16) and were organized in five panels as follows: 1) The biotic panel, which included the primary types of HVs host in the endemic region and the associated HVs subtypes. Endemic region types were named after the primary host species in the region. The distinguishing indexes of HFRS endemic regions mainly include the community structure and population density of local rodent hosts, virus-carrying rate of dominate rodent species, epidemic seasonality, serotype of HVs, and recessive infection rate (36). In Apodemus regions, HFRS primarily occurs in autumn and winter. In Rattus regions, HFRS occurs in spring. In Mixed regions, cases are reported in both of those time periods. Cities with no apparent seasonal pattern or reference were classified as Unspecified regions. The process of identifying affiliations of each 356 cities in the division of endemic region is introduced in supporting text in SI Appendix. Moreover, the distribution ranges of Apodemus agrarius and Rattus norvegicus in China are depicted in SI Appendix, Figs. S17 and S18, respectively. 2) The environmental panel, which included the climate zones and ecological zones. 3) The topographical panel, which included the geomorphic type and the three-step terrain classification. 4) The health panel, which included the healthcare access and quality index (HAQ index) and infectious disease incidence classification. 5) The socioeconomic panel, which included the economic development zone, impervious surface area ratio classification, and GDP classification.

Additional information on the definitions and selection of spatial variables is provided in SI Appendix.

Data on climate change are simulated data from 2005 through 2098 (SI Appendix, Table S2). The 21 climate variables extracted for this period were of two types (49): extreme climate indices and mean climate variables. Extreme climate indices include the maximum air temperature (TXx), minimum air temperature (TNn), frost days (FD), summer days (SU), ice days (ID), tropical nights (TR), number of hot days (TN10p), number of cool days (TX10p), number of warm nights (TN90p), number of hot days (TX90p), warm spell duration index (WSDI), cold spell duration index (CSDI), very heavy rain days (R20mm), total annual precipitation from very heavy rain days (R99p), maximum length of a dry spell (CDD), maximum length of a wet spell (CWD), maximum 1-d precipitation (Rx1day), maximum 5-d precipitation (Rx5day), and precipitation intensity index (SDII). Mean climate index variables consisted of the monthly mean temperature (MMT) and monthly precipitation (MP). Detailed information on the climate variables can be found in SI Appendix. Projected temperature and precipitation data were derived from RegCM4.4 climate simulations over the CORDEX–East Asia domain (https://esgf-data.dkrz.de/search/cordex-dkrz/). All projected temperature and precipitation data originated from the average global climate scenarios under the three representative concentration pathways (RCPs) RCP2.6, RCP4.5, and RCP8.5 (SI Appendix).

The population data were likewise of two types: historic data covering the years 2005 through 2020 and projected data for the years 2021 through 2098 (SI Appendix, Figs. S2 and S5H). The annual gridded population was used to represent the historic population size (50). Projected population size data are based on the fertility rate and included low-, medium-, and high-fertility rate groups (51).

GAM Model for Spatial Variables Section.

To select a spatial variable capable of effectively and accurately explaining the spatial heterogeneity of HFRS cases, we compared the ability of 11 potential HFRS-associated spatial variables to geographically divide HFRS endemic regions via GAM models. The key spatial variable that explains the spatial heterogeneity of HFRS cases should generate the best model performance, with the lowest unbiased risk estimator (UBRE) and the highest percentage of deviance explained (PD), when the GAM model is adjusted for monthly mean temperature (MMT) and monthly precipitation (MP) (52) as follows:

$$\log \left[E\left(Y_{\mathit{it}}\right)\right]=\alpha |r+s_1\left(T_{\mathit{it}}\right)|r+s_2\left(P_{\mathit{it}}\right)|r+\epsilon .$$

where $i=1,\cdots ,I(I=356)$ represents the city, and $t=1,\cdots ,T(T=12)$ represents the month. $E\left(Y_{\mathit{it}}\right)$ denotes the expected HFRS cases and $r$ the tested spatial variables (e.g., types of endemic region). $\alpha$, $T_{\mathit{it}}$, $P_{\mathit{it}}$, and $\epsilon$ are the model intercept, MMT, MP, and unexplained stochastic residuals, respectively. $s_1$ and $s_2$ are the smoothing functions (thin plate regression spline function with four knots) for MMT and MP, respectively. The operator $|$ represents interaction; hence, $\alpha |r$ is the HFRS-associated region-specific intercept, quantifying the average level of HFRS cases in different HFRS-associated regions.

MAVE Model for Evaluating Climate Variables.

To better extract the information contained in the original climate variables (53), we applied the MAVE method to generate HFRS-associated climate variables, which could be divided into two groups (49) based on their relationships with the temperature and precipitation indices. The variable generation involved transforming the original climate variables into an equal number of orthogonal dimensions using linear equations. All climate variables were then integrated and combined to form the effective dimension reduction (EDR) space. To better identify the important climate variables that influence the number of HFRS cases, we selected the optimal subset of dimensions with the lowest CV(d) value. The larger absolute coefficients in linear equations related to that optimal subset were used to identify the significant climate variables.

Integrated Nested Laplace Approximation Model for Spatiotemporal Prediction.

We utilized INLA to construct a spatiotemporal model to predict HFRS trends from 2021 to 2098. This method can efficiently account for the zero-inflated problem as well as the similarities among spatial neighborhoods and consecutive times, which are common in epidemiological data (54, 55). We then used the model to generate spatiotemporal predictions of HFRS cases based on three population scenarios (low, medium, and high) and three climate scenarios (RCP2.5, 4.5, and 8.5), for 27 total climate-population scenarios and took the average of the 27 sets of results as the final prediction (see SI Appendix or details). It is the arithmetic mean with a uniform prior for the 27 combinations. In this study, $E\left(Y_{\mathit{it}}\right)$ is modeled with additive components via the log link function as follows:

$$\log \left[E\left(Y_{\mathit{it}}\right)\right]=\alpha |r+{b\eta }_{\mathit{it}}+{\sum }_j^m{f}_j\left(x_{it,j}\right)|r+{\upsilon }_i+{\nu }_i+{\varphi }_t|r+\epsilon .$$

where ${\eta }_{\mathit{it}}$ is the log-transformed population size with linear coefficients given by b; $f_j()|r$ is the region-specific nonparametric function (first-order random walk) for the j-th MAVE dimension $x_j$, implicitly indicating the nonlinear effects of the climate variables; m is the total number of dimensions selected using the MAVE method; ${\upsilon }_i$ and ${\nu }_i$, respectively, represent the city-level spatially structured and unstructured random effects jointly specified as a Besag–York–Mollie (BYM) model (35); and ${\varphi }_t|r$ is the region-specific seasonal effect to account for seasonality (second-order random walk to capture dependency between months). The INLA model was calculated via the R package INLA (http://www.r-inla.org). In addition, we performed a Theil–Sen median trend analysis (56) to determine the slope (hereafter denoted as $\beta$) for the time series analysis of different levels of HFRS cases for the period of 2021 through 2098 and applied the Mann–Kendall test (35) to determine the significance level of those changes. This allowed us to quantify a significantly increasing trend (P < 0.05) in HFRS cases.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (TXT)

Dataset S02 (TXT)

Dataset S03 (TXT)

Dataset S04 (TXT)

Dataset S05 (TXT)

Dataset S06 (TXT)

Dataset S07 (TXT)

Dataset S08 (TXT)

Data, Materials, and Software Availability

All the relevant codes used for statistical analyses are readily available and attached to this article. The R code used for statistical analyses is listed in Datasets S1–S8. The data can be acquired by contacting xu_lei@tsinghua.edu.cn only for scientific uses.