Introduction

Echinococcus granulosus (E. granulosus), is a parasite that can cause a serious zoonotic disease called cystic echinococcosis (CE) [1]. Its adult tapeworm parasitizes carnivores and its larvae parasitizing various herbivorous livestock and humans. The distribution of CE is extensive and has expanded in parallel with the development of global livestock resources, thus emerging as a significant global public health concern. The World Health Organization (WHO) has identified it as one of the 21 neglected tropical diseases (NTDs) [2].

The life cycle of E. granulosus involves intermediate hosts, such as cattle and sheep, and definitive hosts, typically dogs [2]. Infected canines excrete the parasite’s eggs in their feces, thereby contaminating soil, grasslands, and water sources. Intermediate hosts acquire the infection through grazing on contaminated lands [3,4]. Humans serve as accidental hosts and can become infected by ingesting eggs through contaminated food or water or through direct contact with fecal matter from infected dog. Once ingested, the eggs hatch, migrate through the bloodstream, and lodge in organs, forming cysts that cause severe pathological damage [5].

CE leads to substantial economic losses in both the livestock industry and human healthcare [3,6–9]. In livestock, the disease causes organ damage such as the liver and lungs, reducing the quality and quantity of meat, milk, and wool production. Additionally, it hampers reproductive rates while increasing costs associated with handling infected viscera and deceased animals, and can even result in export bans on animals and their products [10,11]. For humans, the economic burden stems from medical treatments, rehabilitation efforts, as well as a decline in life quality caused by disability. In certain endemic regions worldwide such as Argentina, Peru, East Africa, Central Asia, and China, human incidence rates can exceed 50 per 100,000 person-years while prevalence rates ranging from 5% to 10% have been observed [12]. Globally, CE results in an estimated annual economic losses of USD 3 billion and accounts for approximately one million disability-adjusted life years (DALYs) [13].

China bears a significant portion of the global CE burden, with nearly 400,000 DALYs annually, representing 40% of the global total [14]. This disease burden is primarily concentrated in western provinces such as Tibet, Gansu, Qinghai, Ningxia, Xinjiang, Inner Mongolia and Sichuan. Since the 1950s, over 35,000 human CE cases have undergone surgical treatment in China. A study conducted in Ningxia in 2008 revealed that the average economic burden for CE patients undergoing surgery exceeded USD 1,500 [15,16]. In addition, it is estimated that around 30 million livestock are affected by CE in China resulting in an average economic cost of approximately USD 140 million [17–19].

The prevalence of CE is associated with environmental conditions and the distribution of animal hosts [20]. Environmental factors, such as climate and land use, can influence the transmission cycle of E. granulosus by affecting egg survival and the interaction between animal hosts and humans [21–24]. Several studies have demonstrated a significant correlation between CE prevalence and temperature, precipitation, elevation and landscape [21,23,25,26]. Changes in environmental conditions that facilitate the transmission between definitive and intermediate hosts may increase the risk of human CE [24,27]. Moreover, public awareness of CE prevention plays a crucial role in mitigating the risk of infection [28,29]. Lack of knowledge about preventive measures can exacerbate infection rates, highlighting the necessity for future strategies aimed at enhancing awareness [30]. In fact, a global projection based on burden of disease estimates has indicated that, without intervention, the number of CE cases could exceed 200,000 by 2030, representing a 13.63% increase from 2019 [31]. To address these challenges comprehensively, the World Health Organization has established impact-oriented global targets within the “2021-2030 Roadmap for Neglected Tropical Diseases”, which encompasses bolstering CE control efforts in 17 highly endemic countries by 2030 [32].

The implementation of a comprehensive control plan for CE is currently underway in China, with the aim of achieving a 98% awareness rate among primary and secondary school students in endemic regions by 2030 [33]. Despite these commendable efforts, there remains a dearth of studies examining the projected risk of CE in response to future global changes, as well as simulations exploring the potential impact of increased public awareness on CE transmission risk.

Therefore, in order to bridge this knowledge gap, we developed spatial risk models for CE in animal hosts (cattle, sheep, and dogs) and humans based on the 2012–2016 echinococcosis survey data in China [19]. Using Bayesian Additive Regression Trees (BART), we predicted the future distribution of CE infection risk in western China by the 2030s, while considering future climate and social-economic changes. Furthermore, we simulated potential CE risk scenarios under both unchanged and increased awareness rates. This will aid relevant government agencies in strengthening their surveillance and early warning capabilities for CE in China, providing a scientific foundation and strategies for targeted prevention and control measures in the future.

Materials and methods

Modeling methods

Analytical models.

The Bayesian additive regression trees (BART) is used in this study for species distribution modeling, and subsequent model predictions are made using the dbarts package implemented in the R package embarcadero [35].

Compared to other classification and regression tree (CART) methods, BART is an integrated method that combine Bayesian theory with the cumulative tree model, which applies prior information to the parameters of the cumulative tree model [36,37]. In the single tree model (single binary regression tree), tree T is a binary tree consisting of a set of internal node decision rules and a set of leaf nodes. M={μ₁, μ₂, ……, μ_b} represents a set of parameter values associated with all b leaf child nodes of T. Specifically, the segmentation rule of non-leaf nodes is a binary segmentation of the value domain space of the input variables, in the form of {x ∈ A} and {x ∉ A}, where A is a subset of the input variable. Each is made from top to bottom to be associated with a leaf node of tree T, and its associated parameter μ_i is the response value of the input x. Given T and M, the model of a single tree can be expressed as:

(1)

Where g (x; T, M) is a function of mapping {μ_i∈ M}, ε is the Gaussian noise of the number of samples, and ε follows N (0, σ²). Thus, the additive tree model can be expressed as:

(2)

Where T_j represents each binary regression tree, M_j represents the parameter set of leaf nodes associated with T_j, x is mapped to the {μ_ij∈ M} by the function g (x; T_j, M_j). When the number of trees, m, is greater than 1, each μ_ij merely represents a fraction of the final predicted value. When g (x; T_j, M_j) has only one variable x, each μ_ij merely represents the main effect result of this variable. When g (x; T_j, M_j) depends on multiple variables x, μ_ij is expressed as the result of the interaction of multiple variables x [38].

Since the parameters (T₁, M₁), (T₂, M₂), ……, (T_m, M_m) and ε determine the BART model, prior information needs to be added to those parameters. For a given sample dataset, the joint posterior distribution of the model parameters can be obtained by applying a Bayesian approach and then using sampling to obtain the parameter values. BART model parameter training uses Bayesian Backfitting MCMC (Markov Chain Monte Carlo) algorithm, which is a combination of Bayesian Backfitting fitting strategy and MCMC repeated sampling algorithm.

The BART model mentioned above is based on continuous response variables, while the BART model can also use logit link to deal with binary classification problems, whose formula is:

(3)

where Φ is the cumulative distribution function of the standard normal distribution. [39]

Modeling and future risk predictions.

To mitigate the impact of high correlation among the 19 bioclimatic variables on the accuracy of prediction results, we initially performed Principal Component Analysis (PCA) to assess the correlation structure and reduce redundancy among variables. Based on the variable loadings in the principal components and the cumulative variance explained (99%), we selected nine representative variables: BIO2, BIO6, BIO8, BIO9, BIO12, BIO13, BIO14, BIO15, and BIO19. These variables capture the major environmental gradients while retaining ecological interpretability.

We further conducted a correlation analysis to identify variable pairs with a correlation coefficient surpassing 0.8, with the results of the analysis displayed in S1 and S2 Figs. Subsequently, we conducted Variance Inflation Factor (VIF) analysis, with the results presented in S3 Table. Specifically, variables exhibiting a VIF value exceeding 10 or a correlation coefficient surpassing 0.8 were identified as having multicollinearity issues and subsequently excluded [40,41]. Then, the BART model was employed to predict the risk of CE in three main animal hosts (cattle, sheep, and dogs) at the township level. Each model utilized a distinct combination of variables selected based on VIF analysis and correlation analysis (S4 Table). Subsequently, using selected ecoclimatic, geographical, and socio-economic variables, we first estimated the infection risk for dogs, cattle, and sheep. Then, by incorporating these predicted animal hosts infection risks into the human CE modeling analysis, we predicted the risk of human infection with CE. The previously processed data on human CE positive and negative points were used as independent variables in the township-level modeling analysis. Additionally, in the human CE modeling process, two strategies were considered: one including awareness rate and another excluding it.

The data from each animal host and human cases was randomly split into an 80% training set and a 20% test set. The method of 10-fold cross-validation was used to fit 100 models by repeating the process 100 times. The final estimation results and overall predictive performance are represented by the average values of the Relative Contributions (RC) of all predictors, the Area Under the Receiver Operating Characteristic Curve (AUC), Root Mean Square Error (RMSE), and F1-score, which were calculated across 100 models on the test sets. RMSE serves as a measure of the average magnitude of the errors, with lower values indicating a better fit between the predicted and actual values. The F1-score acts as a balanced measure of a model’s accuracy, combining precision and recall, especially useful in situations with imbalanced classes. Lastly, the standard deviation was calculated based on 100 predictions to represent the model uncertainty.

The assessment of variable importance in BART models is typically measured by calculating the number of times a given variable is used in the complete posterior probability sampling of the trees. Variable response curves are generated by applying the BART model to predict each variable over its range of values. The standard deviation of variable importance and the confidence intervals of variable response curves are calculated through 100 repeated runs of the model. Besides, the R package embarcadero includes a special tool that can draw spatial partial dependence plots and reclassify prediction raster based on partial dependence plots of predictor variables, displaying the relative suitability of different areas for a single covariate [35].

To further assess the impact of future change on CE risk, we utilized projected climate data from four GCM models under three scenarios to predict the risk of CE infection in animal hosts and humans in the 2030s based on the BART model. By averaging the future prediction results derived from the four GCM models, we obtained more robust predictions. This process can provide a comprehensive evaluation of future risk distributions.

Results

Analysis of driving factors for CE

The ensembled BART models for predicting the presence of CE in animal hosts and humans demonstrated good performance (Table 1). The human CE risk model, which incorporated the awareness rate, showed improved performance compared to the model excluding awareness, with higher AUC and F1 scores and lower RMSE values. As a result, subsequent analyses focused on the results of the model incorporating the awareness rate.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Model accuracy for predicting CE infection risk.

https://doi.org/10.1371/journal.pntd.0013182.t001

Predictive models for animal hosts were constructed based on selected variables, with the importance of each variable illustrated in S3 Fig. S4-S6 Figs provide the response curves for the top six variables influencing the infection risk in each animal host. In the human CE risk model, the awareness rate emerged as the most influencing factor (mean importance: 7.52% ± 0.15%), followed by elevation (mean importance: 7.46% ± 0.17%), and the infection risks of dogs and sheep. Fig 1 presents the response curves for the top six variables in the human CE risk model. As shown, awareness rate has a negative effect on human CE risk, while elevation, dog infection risk, and sheep infection risk are positively correlated with human CE. In addition, these risk models also exhibited low uncertainty levels as shown in S7 Fig.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Response curve for the six most important risk factors in the human CE risk model (including awareness rate).

The blue shading gives a 95% credible interval from the posterior distribution of the Bayesian model.

https://doi.org/10.1371/journal.pntd.0013182.g001

The spatial partial dependency plots of the six most important variables for the animal infection models were generated to visualize the effect of individual covariates across different geographic areas (S8-S10 Figs). Fig 2 presents the spatial partial dependence plots for the six most important variables related to human CE risk. The plots show that higher-elevation areas, particularly on the Qinghai-Tibet Plateau, exhibit increased disease risk, and regions with higher infection risks in sheep and dogs correspond to higher human CE risk. In contrast, areas with higher awareness rates demonstrate significantly reduced CE risk.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Spatial partial dependence plots in human CE risk model.

Variables are given in descending order of the first six places of model importance. Values are unitless partial effects on predicted probabilities that range from 0.6 to 0.8. Note: Base map data from Map World (https://map.tianditu.gov.cn/).

https://doi.org/10.1371/journal.pntd.0013182.g002

Current distribution of CE infection risk

Fig 3 illustrates the current spatial distribution of predicted CE infection risk for cattle, sheep, dogs, and humans. The analysis shows that high CE infection risk for animal hosts is predominantly concentrated in Qinghai, Gansu, and Xinjiang. Dogs exhibit high-risk areas extending into Yunnan, while cattle infection risk is mainly found in Qinghai, northern Sichuan, Tibet, and northeastern Inner Mongolia. Similarly, human CE infections are most prevalent in Tibet, Qinghai, Xinjiang, and northern Sichuan.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The current spatial distribution of predicted CE infection risk based on the modelling analysis.

(a)cattle, (b)sheep, (c)dog, (d)human CE. Note: Base map data from Map World (https://map.tianditu.gov.cn/).

https://doi.org/10.1371/journal.pntd.0013182.g003

To quantify the extent of risk, a threshold of 0.5 was established to define high-risk and low-risk areas for human CE infection. The population proportion in potential high-risk areas relative to the total population in each region was calculated (Table 2). For example, Tibet and Qinghai exhibit the highest proportions of their populations residing in potential high-risk areas, with 75.93% and 46.09%, respectively. Other regions such as Xinjiang, Gansu, and Ningxia also show significant population proportions residing in potential high-risk areas (>5% of all populations).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. The population proportion in the provinces of western China residing in potential high-risk areas.

https://doi.org/10.1371/journal.pntd.0013182.t002

Future distribution of human CE risk with unchanged and increased awareness rates

The predicted risk of CE among cattle, sheep, and dogs under three future scenarios (SSP2-4.5, SSP3-7.0, and SSP5-8.5) is shown in S11 Fig. Comparisons between the future predictions with current infection risk are presented in S12 Fig. Additionally, the future predictions for both animal hosts’ and humans’ future predictions exhibit low uncertainty, as shown in S13 and S14 Figs, respectively.

The future human CE risk predictions were conducted using two strategies: Strategy A, which assumed a constant awareness rate, and Strategy B, which assumed an increased awareness rate of 90%. We compared the results of these strategies to evaluate the impact of enhanced awareness on future CE risk. Figs 4 and S15 illustrate the future spatial distribution of CE infection risk under both strategies. The population proportion and size residing in potential high-risk areas in various western provinces by the 2030s were quantified and compared to those in the 2010s (S5 and S6 Tables). Table 3 highlights the changes, suggesting that with increased awareness, the population in potential high-risk areas may decrease by 6.99 to 7.92 million individuals, accompanied by a decrease in the proportion of the at-risk population by 2.72% to 3.35% compared to the scenario of constant awareness. This substantial reduction was particularly evident in provinces such as Qinghai, Xinjiang, and Gansu, where the enhancement of local awareness coincided with a decrease in populations residing in potential high-risk areas. Specifically, in Qinghai, the high-risk population decreased by 0.54 to 1.11 million individuals, with a reduction in proportion from 11.85% to 22.28%; in Xinjiang, it decreased by 2.04 to 2.19 million individuals, with a reduction in proportion from 10.38% to 11.93%; and in Gansu, it declined by 1.67 to 2.39 million individuals, with a reduction in proportion from 3.27% to 9.63%.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Changes in population proportion (%) and population size (Million) residing in high-risk areas with increased awareness rate (Strategy B - Strategy A).

https://doi.org/10.1371/journal.pntd.0013182.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Future predicted infection risk distribution and changes of human CE in western China based on the modelling analysis.

(Strategy B: Awareness Rate Increased). Note: Base map data from Map World (https://map.tianditu.gov.cn/).

https://doi.org/10.1371/journal.pntd.0013182.g004

The data suggest a correlation between increased awareness rate and a reduction in the overall number of at-risk individuals, with a particularly strong association observed in provinces with larger high-risk populations, where both the number and the proportion of populations at risk tend to decrease.

Discussion

The zoonotic infection of CE has had a significant global impact on public health, with China being the country with the highest worldwide prevalence [42]. In 2016, the Chinese government released the “Healthy China 2030” blueprint, with the objective of strengthening the prevention and control against major infectious diseases [43]. To implement this blueprint, China CDC issued a comprehensive prevention and control plan for combating key parasitic diseases in 2024, which included specific requirements regarding the awareness rate of CE related knowledge among primary and secondary school students, to effectively control the spread of CE across all endemic counties nationwide [33]. Thus, understanding the changes in the risk of CE prevalence in the context of global changes serves as a fundamental prerequisite, while understanding the impact of increased awareness on the risk of CE is key for achieving precise disease control.

Our results indicate that climatic conditions, including temperature and precipitation, significantly influence the transmission risk among various hosts (Figs S6–S8 and 1). This could be attributed to the vulnerability of E. granulosus transmission to climatic factors, as variations in temperature and humidity can affect the survival and suitability of eggs in the external environment, thereby impacting disease transmission [27,44,45]. Notably, a previous study found no clear relationship between temperature and CE risk in Chile, suggesting that the influence of climate factors on CE prevalence may vary across regions [46]. In addition, it is evident that the risk of CE is positively correlated with grassland and forest (S3 Fig). This correlation is likely due to the role of these landscapes as natural habitats for intermediate hosts such as cattle and sheep. These habitats provide favorable conditions for the survival of both host and parasite survival, facilitating disease transmission [25,47].

In addition to considering various environmental factors, we also included biological factors, primarily the risk of infection in the three main animal hosts for CE, into our human CE risk model. Our analysis demonstrated a nonlinear positive correlation between infection rates in animal hosts and the risk of human CE. Intermediate hosts such as sheep and cattle primarily acquire infections through ingesting feed and water contaminated with eggs from dog feces [5]. Subsequently, when definitive hosts like dogs consume infected sheep or cattle, they become infected and excrete parasite eggs into the environment, thereby increasing the contamination level and increasing the risk of human infection [48].

A key finding of this study is the notable association between public awareness and the reduction of human CE risk. Our models show that awareness is the most significant factor influencing human infection risk. Notably, increased public health awareness is linked to a marked decrease in human infection risk. This finding suggests that enhanced public education, particularly in high-risk areas, is essential for disease prevention. As the awareness assessment in this study such as knowledge of transmission routes, hygiene practices, interactions with dogs, post-slaughter livestock handling, and attitudes toward dog deworming, improvements in these areas of awareness are associated with a lower risk of human CE, suggesting the potential value of targeted public health education in disease prevention. Previous research has also demonstrated that educational campaigns can significantly reduce CE prevalence, thereby reinforcing the importance of integrating public health education into disease control strategies [30].

The future risk predictions presented in this study emphasize the potential role of increased awareness in mitigating the burden of CE in western China. Under Strategy B, when public awareness increases to 90%, the modeled risk can decrease by 4.12% compared to the 2010s baseline. On the other hand, relative to Strategy A, which assumes no change in awareness, the reduction in population at-risk might be up to 3.35%. This reduction is particularly pronounced in high-risk regions such as Tibet, Qinghai, and Xinjiang, where the population proportion residing in high-risk areas could drop by as much as 22.28% in comparison to Strategy A. Conversely, without significant improvements in public awareness, it is possible that regions such as Xinjiang, Gansu, and Yunnan are projected to experience increasing CE risk due to the lack of sufficient awareness, which leads to the continued spread of infection among animal hosts. This could result in an increase of as much as 8.43% in CE risk, while Strategy B is expected to minimize this risk by 11.93% in contrast to Strategy A.

These results emphasize the potential contribution of public health awareness to the prevention and control of CE. By improving public awareness of CE, the proportion of the population residing in areas identified as high risk is projected to decline, thereby contributing to a reduction in overall community CE risk. Additionally, the findings also highlight the importance of integrating public health education into disease control strategies. Awareness interventions that focus on key behavioral pathways such as reducing close contact with dogs, adopting hygienic practices after livestock slaughter, and promoting regular dog deworming have the potential to produce measurable reductions in infection risk. When public health awareness is increased, host animals management may also be more standardized [30], which in turn could help limit zoonotic transmission of CE and reduce future risks through preventive measures. In terms of implementation, such interventions are supported by existing government efforts. For instance, national plans under the “Healthy China 2030” initiative have explicitly promoted CE awareness among primary and secondary school students in endemic counties. This highlights the feasibility of integrating educational campaigns into existing infrastructure [33,43]. To ensure sustained impact, however, factors such as accessibility, community engagement, and cost-effectiveness, especially in rural and remote areas, must also be considered. Further research is needed to evaluate the practical implementation and long-term policy relevance of these interventions.

It is important to note that the public awareness data used in this study come from cross-sectional surveys and are not linked to individual infection status. As such, the associations found reflect ecological relationships rather than causality. While the projections suggest possible benefits of increased awareness under certain scenarios, more evidence from longitudinal or interventional studies is needed to confirm its actual impact on CE risk.

While this study provides valuable insights into the present and projected risk of CE in western China, several limitations should be considered. First, the study does not account for the distribution of stray dogs, which are key hosts for E. granulosus, while our study acknowledges the exclusion of stray dogs as a limitation, future work should explore targeted interventions such as stray dog population monitoring and control to better capture the full transmission dynamics. Incorporating such data in future studies could provide a more comprehensive assessment of CE transmission risk. Additionally, this study focused primarily on western China, where most CE cases are reported. Expanding the analysis to include eastern regions and other areas with emerging CE cases would enhance the generalizability of the findings. Moreover, while the importance of public awareness interventions is emphasized, future studies should further evaluate their practical implementation in rural and resource-limited settings.

Conclusion

This study evaluated the impact of various risk factors on CE infection in animal hosts and humans in western China, integrating echinococcosis survey data with geographical, ecoclimatic, socio-economic, and biological variables. By applying Bayesian additive regression tree (BART) models, we predicted current CE infection risk for three animal hosts and humans, and further projected future risk distributions under the SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios. The projections considered two strategies that accounted for both constant and increased public awareness regarding CE prevention. Notably, under higher awareness conditions, the proportion of the population residing in potential high-risk areas could decrease by approximately 2.72% to 3.35% by 2030, highlighting the potential of public education in mitigating infection risk. These findings highlight the critical role of incorporating awareness-related interventions into CE control strategies. Additionally, this research offers a valuable framework for assessing other zoonotic diseases and emphasizes the need for adaptive management strategies as climate and socio-economic conditions change.

Supporting information