Spatiotemporal dynamics, clustering, and ARIMA-based short-term prediction of human brucellosis in Xinjiang, China

Abstract

Background

Brucellosis, a global zoonotic disease caused by Brucella spp., poses a severe threat to public health and livestock husbandry in Xinjiang, China’s major pastoral region. Despite control efforts, its recent resurgence and complex spatiotemporal dynamics remain inadequately characterized. This study aimed to analyze brucellosis’ temporal trends, spatial clustering, and short-term forecasts in Xinjiang (2010–2023) to inform targeted interventions.

Methods

Prefecture-level incidence data (2010–2023) were collected, with ARIMA (2,1,1) time-series modeling for forecasting and global/local spatial autocorrelation analysis (Moran’s I, LISA) to identify clustering.

Results

A total of 68,437 cases were reported, showing three distinct phases: upward (2010–2015, peak: 8,820 cases), decline (2016–2020), and rebound (2021–2023). Cases exhibited stable seasonal peaks (May–August, 57.9%–62.1% of annual cases). The ARIMA model demonstrated high predictive accuracy (adjusted R²=0.89), with short-term projections of 10,021 cases in 2024 and 10,583 cases in 2025. Spatially, incidence shifted from negative autocorrelation (2010–2020) to weak positive autocorrelation (2021–2023), with hotspots moving from Northern/Eastern Xinjiang to western regions (e.g., Kizilesu Keerkezi Autonomous Prefecture) and persistent cold spots in southern Xinjiang.

Conclusions

This study elucidates the dynamic spatiotemporal patterns of brucellosis in Xinjiang, shaped by livestock practices, environmental factors, and interventions. The forecasts and clustering insights support region-specific One Health strategies, our findings provide an analytical framework and actionable evidence for precision brucellosis surveillance in pastoral zones across Central Asia.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-026-13220-x.

Background

Brucellosis, a global zoonotic infectious disease caused by bacteria of the genus Brucella, poses a serious threat to public health security and socio-economic development [1]. The World Health Organization (WHO) estimates that there are approximately 500,000 new clinical cases worldwide each year, with the epidemic spreading across more than 170 countries and regions on five continents [2, 3]. This disease not only causes symptoms such as prolonged fever, joint pain, and fatigue in humans, affecting patients’ quality of life [4], but also leads to reproductive disorders in livestock, including abortion and infertility, resulting in substantial economic losses to the animal husbandry industry [5].

China has experienced a long-standing burden of brucellosis, with the national epidemic showing a marked resurgence in recent years; reported human brucellosis cases rose markedly from 4060 (16.09/100,000) in 2019 to 8399 (32.64/100,000) in 2023 [6, 7]. Nationally, the epidemiological trajectory of brucellosis is closely linked to livestock management policy shifts and environmental changes: the rapid expansion of cattle farming with inadequate biosecurity measures led to a nearly 30-fold surge in human cases between 2006 and 2015 [8, 9], while strengthened interventions (infected animal culling, expanded livestock vaccination) drove a temporary national decline from 2016 to 2020. Since 2021, a national rebound has occurred, hypothesized to be exacerbated by COVID-19 pandemic-related public health resource diversion, disrupted routine surveillance, and climate variability impacting animal husbandry practices [10, 11]. As China’s largest pastoral region with an extensive grassland ecosystem and livestock-dependent economy, the Xinjiang Uygur Autonomous Region has consistently ranked among the country’s brucellosis high-incidence areas [8], and its local epidemic trends align with national dynamics while exhibiting distinct regional characteristics. National surveillance data further reveal pronounced spatial heterogeneity in brucellosis transmission across China: reported cases decreased nationally from 47,139 to 37,947 between 2016 and 2019 before rising to 44,036, with divergent regional trends—including a decline in Xinjiang and a concurrent increase in Inner Mongolia during this period [12]. This striking spatial variation across China highlights the limitations of uniform national control strategies and underscores the urgent need for region-specific, data-driven prevention and control measures under the National Brucellosis Prevention and Control Plan.

Although extensive domestic and international research has been conducted on brucellosis—covering pathogen characterization, transmission pathways, diagnostic methods, and vaccine development—substantial gaps remain in understanding its complex spatiotemporal transmission patterns, accurately identifying high-risk areas, and developing effective multisectoral control strategies [13–15]. These gaps are particularly pronounced in ecologically diverse regions with mixed livestock production systems, such as Xinjiang, where environmental, socio-economic, and husbandry-related factors interact in complex ways [16]. Spatiotemporal analysis techniques are widely applied in investigation of infectious diseases to uncover clustering patterns, seasonal trends, and underlying transmission mechanisms [17–20]. In recent years, time-series forecasting models—including classical statistical approaches such as Autoregressive Integrated Moving Average (ARIMA) and Seasonal Autoregressive Integrated Moving Average (SARIMA) models, as well as data-driven methods such as Prophet, neural network models like Long Short-Term Memory (LSTM), and machine learning approaches including Support Vector Regression (SVR) and hybrid ARIMA–SVR models—have been increasingly used to predict infectious disease incidence and support public health decision-making [21–24]. Among these, ARIMA-based models remain widely used due to their solid theoretical foundation, interpretability, and standardized Box–Jenkins modeling framework, while SARIMA models are particularly effective in capturing seasonal patterns commonly observed in epidemiological data [23, 25]. However, existing studies in China have often relied on short study periods or single forecasting frameworks, and although recent work has applied ARIMA, SVR, and hybrid ARIMA–SVR models to predict brucellosis incidence, these studies generally lack integration of long-term data and fine-scale spatial analysis [21]. Building upon and extending previous work, this study employs a comprehensive spatiotemporal analytical framework to investigate human brucellosis in Xinjiang at the prefecture level from 2010 to 2023. Specifically, we integrate spatial autocorrelation analysis with ARIMA-based time-series modeling to examine temporal trends, seasonal characteristics, and spatial clustering patterns of the disease. The novelty of this study lies in the combined application of long-term prefecture-level surveillance data and integrated spatiotemporal methods to simultaneously identify high-risk clusters and generate short-term forecasts, thereby providing a more nuanced understanding of brucellosis dynamics in a major pastoral region of China.

Materials and methods

Data sources

We collected the incidence of human brucellosis of 14 prefectures in Xinjiang from January 2010 to December 2023. Data from January 2010 to December 2020 were obtained from the website of the National Public Health Data Center (http://www.phsciencedata.cn/Share/), whereas the data from January 2021 to December 2023 were obtained from the website of the Xinjiang Health Commission, and the Xinjiang Statistical Yearbook.

Statistical software

Microsoft Office Excel 2021 (Microsoft Redmond, WDC, USA) and SPSS software (version 23.0, IBM company, New York, USA) were used for the descriptive epidemiological data analysis. ARIMA modeling was conducted using Python (Python Software Foundation, Wilmington, DE, USA). Spatial mapping and spatial autocorrelation analysis were based on ArcGIS software (Version, 10.6, Environmental Systems Research Institute, Redlands, USA).

ARIMA model

To identify the optimal time-series model specification for brucellosis incidence forecasting, we conducted a formal comparison of candidate ARIMA and SARIMA models with a 12-month seasonal cycle (consistent with the annual seasonal pattern of brucellosis in Xinjiang). Model selection was based on three core criteria: the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and adjusted R-squared, which collectively balance model goodness of fit, parsimony, and interpretability. A grid search of plausible ARIMA and SARIMA parameter combinations was performed, and the model with the lowest AIC and BIC values alongside the highest adjusted R-squared was designated the optimal specification for subsequent forecasting and analysis.

The ARIMA model is a widely used approach for forecasting non-stationary time series that exhibit trends or autocorrelation. An ARIMA(p,d,q) model is characterized by three parameters: the autoregressive order p, the degree of differencing d, and the moving average order q.

Let {Y_t​} denote the observed time series. First, the series is differenced d times to achieve stationarity:

where B is the Backshift Operator defined by B^kY_t​=Y_t−k​.

The stationary series {Xt​} is then modeled as an ARMA(p,q) process:

where Φ(B) =  and Θ(B)​=1 + θ_1​B + θ₂​B²+⋯+θ_q​B^q (MA polynomial)( (Box-Jenkins sign convention)),​.

and {εt​} is a white noise with.

It is assumed that εt​ is uncorrelated with past observations Yt − 1​,Yt − 2​,….

For the ARIMA(2,1,1) model applied in this study, the specification is:

Spatial autocorrelation analysis

Spatial autocorrelation analysis, including global and local measures, was used to describe the spatial dependence and assess clustering or dispersion among study areas. Moran’s I index was used to evaluate overall spatial autocorrelation and spatial distribution of the study areas while the local one can be further used to identify local spatial autocorrelation and the specific clustering areas. Moran’s I values range from − 1 to 1, with I > 0 indicating positive spatial autocorrelation, I < 0 indicating negative spatial autocorrelation, and values close to 0 suggesting a random spatial distribution. Larger absolute values of Moran’s I indicate stronger spatial correlation. LISA is used to reflect a geographical phenomenon on a regional unit or the degree of correlation between an attribute value and the same geographical phenomenon or attribute value on the adjacent unit. LISA classified into four types: low–high (L–H), indicating low-incidence areas surrounded by high-incidence areas; high–low (H–L), indicating high-incidence areas surrounded by low-incidence areas; low–low (L–L), representing cold spots; and high–high (H–H), representing hot spots. The spatial autocorrelation analysis was performed using ArcGIS software.

where:

n is the number of spatial units.

w_ij​ is the spatial weight matrix element.

x_i​ is the observed value at spatial unit i.

x̅ is the mean of the observed values.

Ethics statement

This study was approved by the Ethics Committee of Xinjiang Medical University. All data analyzed in this study were de-identified to protect patient confidentiality, and aggregated data were used for the analyses in our study.

Study area

As illustrated in Fig. 1, the Xinjiang Uygur Autonomous Region is located in northwest China, spanning 73°40′-96°18′E and 34°25′-48°10′N, with a “three mountains and two basins” terrain and an arid/semi-arid temperate continental climate marked by large diurnal temperature differences and scarce, variable precipitation—key environmental factors shaping livestock husbandry practices and Brucella persistence in the region. Xinjiang shares borders with eight countries and adjoins Gansu, Qinghai, and Tibet domestically; its transboundary geographic position is strategically relevant to cross-border livestock movement and the control of zoonotic diseases such as brucellosis, a critical contextual factor for this study’s spatial analysis of disease clustering.

Fig. 1.

Map of study area in Xinjiang autonomous region, China

Results

From January 2010 to December 2023, a total of 68,437 cases of human brucellosis were reported in Xinjiang, with the annual reported cases during this period presented in Figs. 2 and 3. From 2010 to 2015, despite annual fluctuations, the number of reported brucellosis cases exhibited an overall upward trend. By 2015, the reported cases reached a peak of 8,820, with the incidence rate surging to 34.96 per 100,000 population, indicating a severe epidemic situation during this period.

Fig. 2.

Trends of annual brucellosis cases and incidence rates from 2010 to 2023 (The red dashed line represents the exponential trend line of the annual incidence)

Fig. 3.

Changes in brucellosis incidence rates in 14 prefectures of Xinjiang from 2010 to 2023

Subsequently, from 2016 to 2020, both the number of cases and the incidence rate decreased year on year. By 2020, the number of reported cases dropped to 3,010, and the incidence rate declined to 11.93 per 100,000, suggesting the initial success of epidemic containment measures.

As illustrated in Fig. 2, an apparent rebound occurred between 2021 and 2023, characterized by a sharp rebound in the brucellosis epidemic. The number of reported cases in these three years was 4,167 (16.60 per 100,000), 5,647 (22.04 per 100,000), and 8,399 (32.64 per 100,000), respectively.

Across the entire study period (2010–2023), human brucellosis cases in Xinjiang demonstrated a consistent and pronounced seasonal distribution (Fig. 4), with a distinct high - incidence window concentrated in May–August annually. Specifically, the average monthly number of cases reached 694.07 in May and 730.21 in June—representing approximately 3.3 to 3.6 times the average monthly cases observed in December (204.43), the typical low - incidence month.

Fig. 4.

Monthly distribution of brucellosis cases in Xinjiang from 2010 to 2023

Notably, this seasonal pattern remained unaltered even during the epidemic rebound phase (2021–2023). For instance, in 2023, the monthly case counts in May (900) and June (970) remained the highest of the year, while the number of cases in December (392) was substantially lower than those in the spring and summer months. This consistency underscores the stability of the seasonal drivers governing brucellosis transmission in Xinjiang.

An ARIMA (2,1,1) model was constructed to predict monthly human brucellosis cases in Xinjiang, model specification was selected based on minimization of the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) among candidate ARIMA and seasonal ARIMA (SARIMA) specifications.

ARIMA(2,1,1) yielded the lowest AIC (1685.42) and BIC (1697.65) values, and the highest adjusted R² (0.892). SARIMA models incorporating seasonal components exhibited higher information criterion values (AIC: 1698–1721; BIC: 1714–1737) and lower adjusted R² (0.854–0.878), indicating that the inclusion of seasonal parameters did not improve model fit sufficiently to justify their additional complexity for short-term forecasting (Supplementary Table S1).

Model performance was assessed using root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). During the training period (2010–2020), the model achieved an RMSE of 130.7, a MAPE of 25.8%, and an MAE of 98.4, indicating acceptable in-sample fit for epidemiological time-series data (Table 1; Supplementary Figure S2).

Table 1.

Model statistics for ARIMA model

Model	R-squared	RMSE	MAPE(%)	MAE	Ljung-Box Q(18)
					Statistics	DF	Sig.
ARIMA(2,1,1)	0.892	130.7	25.8	98.4	84.045	13	> 0.0001

Full residual diagnostics were conducted to validate the key assumptions of the ARIMA(2,1,1) model. The Ljung–Box Q(18) test yielded a statistic of 84.045 (p < 0.001), indicating significant residual autocorrelation. Inspection of the residual Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) plots revealed no systematic large-magnitude correlations, but the significant Q statistic suggests that the model does not fully capture the temporal structure of the data (Fig. 5).

Fig. 5.

The ACF and PACF of time series after differencing of human brucellosis cases in Xinjiang

Residual normality was assessed via Q-Q plot, histogram, and the Shapiro–Wilk test. Visual inspection showed mild tail deviations from the 45° reference line in the Q-Q plot, with the histogram displaying slight left skew and excess kurtosis (Supplementary Figure S1). The Shapiro–Wilk test rejected the null hypothesis of strict normality (W = 0.987, p = 0.014), though skewness and kurtosis were both within the range commonly considered acceptable for approximate normality (|skewness| < 2, |kurtosis| < 2). Residuals were centered near zero (mean = 0.72, SD = 131.2) with evidence of mild heteroscedasticity over the study period; collectively, these residual characteristics reflect minor departures from strict ARIMA assumptions, with no meaningful impact on the model’s ability to reproduce observed brucellosis incidence trends and seasonal patterns. These diagnostics collectively indicate that the ARIMA(2,1,1) model satisfies the core assumptions only partially. Forecast Confidence Intervals should therefore be interpreted with caution.

To evaluate model robustness to real-world surveillance data limitations, sensitivity analyses were conducted using truncated training periods. The model was re-calibrated on three reduced datasets: 2010–2020, 2010–2018, and 2010–2016, and validated on the subsequent 24 months.

When trained on 2010–2020 data and validated on 2021–2023, forecast accuracy deteriorated markedly (RMSE = 357.0, MAPE = 21.5%). Performance also declined for the 2010–2018 training period (RMSE = 146.9, MAPE = 25.8%) and further for 2010–2016 (RMSE = 189.4, MAPE = 28.3%). Adjusted R² values decreased from 0.892 (full model) to 0.850–0.842 for truncated models (Supplementary Figure S2).

These results demonstrate that the ARIMA(2,1,1) model is sensitive to the length of the training period and the model exhibits reduced predictive accuracy under altered epidemiological conditions, as observed during the 2021–2023 rebound phase. The model should therefore be applied with caution in non-stationary settings.

Despite these limitations, the ARIMA(2,1,1) model accurately reproduced the general trajectory of brucellosis cases during the training period. During the 2010–2015 upward phase, the model predicted an average annual increase of 18.7%, closely aligning with the observed increase from 386 cases (2010) to 8,820 cases (2015). The predicted peak value (8,792 cases) showed a relative error of 0.32%. For the 2016–2020 decline phase, the model estimated an average annual decrease of 19.2%, with a predicted 2020 case count of 3,045 versus 3,010 observed (relative error: 1.2%).

The model also preserved the stable seasonal distribution of brucellosis, with predicted May–August case proportions (58.2%–61.7%) closely matching observed values (57.9%–62.1%). Predicted average case counts for peak months (May: 689.3; June: 725.6) were within 3% of observed averages (May: 694.1; June: 730.2).

Based on the full 2010–2023 model, we generated continuous short-term (2024–2025) annual/monthly projections for human brucellosis cases in Xinjiang (Fig. 6; Table 2). The model projected 10,021 and 10,583 annual cases for the 2024 and 2025 short-term forecasting window of the ARIMA model, respectively. As expected for time-series models, uncertainty increases with longer extrapolation horizons.

Fig. 6.

The ARIMA model fitted and forecasted time series of reported cases of human brucellosis in Xinjiang

Table 2.

Monthly and annual predicted human brucellosis cases in Xinjiang for 2024–2025 by the ARIMA(2,1,1) model

Month	2024		2025
	Predicted Cases	95% Confidence Interval	Predicted Cases	95% Confidence Interval
January	339	[129, 549]	511	[61, 961]
February	401	[106, 696]	514	[27, 1001]
March	559	[219, 899]	601	[86, 1117]
April	771	[413, 1130]	749	[217, 1280]
May	986	[624, 1348]	919	[381, 1458]
June	1152	[790, 1514]	1072	[532, 1612]
July	1232	[869, 1595]	1172	[631, 1712]
August	1211	[848, 1575]	1197	[657, 1738]
September	1100	[737, 1464]	1147	[607, 1688]
October	930	[563, 1298]	1038	[496, 1579]
November	746	[364, 1129]	898	[350, 1446]
December	594	[183, 1006]	765	[204, 1326]
Total annual predicted cases	10,021		10,583

Global spatial autocorrelation characteristics

Global spatial autocorrelation analysis was performed to quantify the overall spatial association of brucellosis incidence in Xinjiang from 2010 to 2023, using the Moran’s I index, Z-score, and P-value (Table 3).

Table 3.

Global autocorrelation of Moran’s I values of brucellosis in Xinjiang (2010–2023)

Year	Moran’s I	Z-Score	P-value
2010	-0.103755	-0.140804	0.88
2011	-0.068707	0.042925	0.96
2012	-0.14803	-0.403837	0.68
2013	-0.173831	-0.465579	0.64
2014	-0.212519	-0.806351	0.42
2015	-0.33778	-1.332906	0.18
2016	-0.368356	-1.436604	0.15
2017	-0.316098	-1.203742	0.23
2018	-0.402701	-1.716163	0.08
2019	-0.493232	-2.094909	0.03
2020	-0.36118	-1.383726	0.16
2021	0.057138	0.691142	0.48
2022	0.013637	0.47244	0.64
2023	0.044508	0.614421	0.54

From 2010 to 2020, the Moran’s I indices for brucellosis incidence were consistently negative, ranging from − 0.493 (2019) to − 0.069 (2011). This indicates an overall negative spatial correlation in the disease’s spatial distribution, meaning high-incidence areas were predominantly surrounded by low-incidence areas and vice versa, with no large-scale contiguous clustering observed. During 2010–2018, the absolute values of Z-scores were all < 1.96 and P-values > 0.05, suggesting the spatial distribution of brucellosis was statistically close to random during this period. A notable and statistically significant turning point occurred in 2019 (Z-score = − 2.095, P = 0.03) – the only year with a significant negative Moran’s I value – driven by two key regional epidemiological and intervention-related factors: (1) the implementation of targeted, prefecture-specific brucellosis control measures in core northern pastoral high-incidence areas (e.g., Tacheng, Altay) starting in 2018, which caused a sharp local decline in incidence, and (2) concurrent sporadic outbreaks in adjacent low-incidence industrial/agricultural regions (e.g., Karamay, Urumqi) that amplified the spatial contrast between high- and low-incidence areas. This combination of targeted local control and heterogeneous sporadic transmission created a pronounced high-low adjacent spatial pattern, leading to the only statistically significant negative autocorrelation in the study period. For example, Tacheng (a northern pastoral area) recorded an incidence rate of 58.3 per 100,000, while neighboring Kelamayi (an industrial city with minimal livestock farming) had an incidence rate of only 1.51 per 100,000, highlighting a stark spatial contrast.

After 2020, the spatial autocorrelation pattern changed: Moran’s I indices shifted to positive values, ranging from 0.014 (2022) to 0.057 (2021). However, none of these values were statistically significant (Z-scores: 0.472–0.691; P-values: 0.48–0.64). While the global Moran’s I values remained non-significant across all years except 2019 (indicating no large-scale, region-wide spatial clustering/dispersion at the autonomous prefecture level), the directional shifts in Moran’s I values (from consistently negative 2010–2020 to weakly positive 2021–2023) reflect meaningful temporal changes in the underlying spatial distribution pattern of brucellosis incidence. A negative Moran’s I (2010–2020) denotes a dominant “high-low adjacent” microscale pattern (isolated high-incidence pastoral areas surrounded by low-incidence non-pastoral regions), while a shift to weak positive Moran’s I (2021–2023) signals the emergence of localized spatial aggregation (e.g., contiguous high-Confidence hotspots in western Xinjiang identified via LISA analysis) – a pattern that is not yet sufficient to drive global statistical significance, but represents a substantive epidemiological shift in disease transmission geography. This positive shift indicates a weakening of the earlier “high-low adjacent” pattern and a tentative trend toward spatial aggregation. However, the lack of statistical significance suggests that the emerging positive spatial correlation remained weak, with no formation of large-scale contiguous hotspots or cold spots during 2021–2023.

Temporal-spatial distribution trends of regions

Combined with the incidence maps (Fig. 7), the high-incidence areas of human brucellosis in Xinjiang displayed a dynamic transmission expansion pattern—with Northern Xinjiang as the initial epidemiological core, followed by a spatial expansion into Eastern Xinjiang and select peripheral regions, and eventual stabilization with persistent regional disparities. This characterization of “expansion” is not based solely on changes in incidence geographies, but on convergent evidence for cross-regional brucellosis transmission including (1) the synchronous temporal rise in incidence in Eastern Xinjiang following peak transmission in Northern Xinjiang; (2) ecological and socioeconomic connectivity between core high-incidence pastoral areas and expanding regions; and (3) alignment with known Brucella transmission pathways. Northern Xinjiang (Tacheng, Altay) served as the initial epidemiological core—a pattern mechanistically linked to this region’s status as Xinjiang’s primary livestock-producing hub, with sheep density exceeding 60 per km² and traditional extensive grazing practices that lack biosecurity measures. Unprotected handling of infected animals and consumption of unpasteurized dairy products created a sustained reservoir of human infection, driving the initial surge in incidence. The incidence rate in Tacheng increased from 24.31 per 100,000 in 2010 to 159.83 per 100,000 in 2014, while that in Altay rose from 25.38 per 100,000 in 2010 to 59.28 per 100,000 in 2014. By 2014, both regions had entered the highest incidence interval (59.28–159.83 per 100,000) and together accounted for over 60% of all high-incidence cases in Xinjiang, confirming their role as the primary source of brucellosis transmission in the region.

Fig. 7.

Temporal-spatial distribution map of brucellosis incidence in Xinjiang from 2010 to 2023 (A-N)

The subsequent expansion to Eastern Xinjiang (Bayingolin Mongolian Autonomous Prefecture, Changji Hui Autonomous Prefecture) was a spillover of transmission from the Northern Xinjiang core, supported by three lines of evidence. First, the temporal trajectory of incidence in Eastern Xinjiang lagged 1–2 years behind the Northern Xinjiang peak (2014), a classic pattern of zoonotic pathogen spillover from core to peripheral regions. Second, Bayingolin and Changji are geographically and economically connected to Tacheng and Altay via major livestock trade routes and transregional grazing corridors, with annual movement of over 1 million head of livestock between these regions—a well-documented driver of Brucella dissemination, as infected livestock act as mobile reservoirs. Third, Eastern Xinjiang’s ecological characteristics enhance Brucella survival in soil and water, creating a suitable environment for pathogen persistence and human exposure following introduction from the northern core. The rate in Bayingolin Mongolian Autonomous Prefecture increased from 14.81 per 100,000 in 2010 to 58.42 per 100,000 in 2014, and that in Changji Hui Autonomous Prefecture jumped from 1.79 per 100,000 in 2010 to 47.01 per 100,000 in 2014; both regions exceeded the incidence interval of 16.85–59.28 per 100,000, forming a northern-eastern high-incidence transmission belt shaped by livestock-mediated pathogen spillover.

In contrast, Hetian and Kashi (in Southern Xinjiang) remained in the lowest incidence interval (< 4.72 per 100,000) consistently, with incidence rates ranging from 0.1 to 1.38 per 100,000 in Hetian and 0.2 to 2.29 per 100,000 in Kashi, showing no clear evidence of diffusion.

A range of potential factors may contribute to this persistently low reported incidence, and no single driver can be definitively confirmed with the current surveillance data. These include structural factors such as limited access to formal healthcare and diagnostic capacity in remote southern pastoral and agricultural communities, which may lead to underdiagnosis and underreporting of mild or subclinical brucellosis cases. Additionally, Southern Xinjiang enforces strict cross-regional livestock quarantine protocols that limit the movement of potentially infected animals from high-incidence northern/eastern pastoral regions into the south. Socioeconomic and husbandry practice differences are also notable: the region’s economy is dominated by small-scale crop agriculture rather than extensive livestock grazing, with sheep and cattle densities (< 20 per km²) far lower than northern pastoral cores (> 60 per km²). Environmental barriers further shape transmission risk, including the arid desert climate of Southern Xinjiang (annual precipitation < 100 mm) and extreme diurnal temperature fluctuations, which reduce the environmental persistence of Brucella spp. in soil and water sources—key reservoirs for pathogen transmission to humans. Collectively, these overlapping factors create a unique regional context for brucellosis transmission, and the low reported incidence likely reflects a combination of true reduced transmission risk and potential surveillance/diagnostic biases, rather than uniform absence of the pathogen.

After 2015, the incidence rates in core high-incidence areas (Northern/Eastern Xinjiang) decreased by more than 50%—Tacheng’s rate dropped from 151.44 per 100,000 in 2015 to 33.82 per 100,000 in 2020, and Altay’s rate fell from 73.65 per 100,000 in 2015 to 22.82 per 100,000 in 2020. This decline is directly attributable to the implementation of China’s National Brucellosis Control Program, including expanded livestock vaccination (S2 Strain, heat-stable for pastoral conditions), compensated culling of infected animals, and public health education on protective practices—interventions that target the animal reservoir and human exposure pathways, the two core pillars of brucellosis control in the One Health framework. Nevertheless, these regions continued to exhibit relatively high incidence levels (24.27–33.82 per 100,000), reflecting residual transmission in small-scale, unregulated pastoral operations that were not fully covered by control measures—a common challenge in pastoral brucellosis control globally.

Hami (in Eastern Xinjiang) exhibited the most significant fluctuations in incidence during the mid-term: its rate surged from 7.94 per 100,000 in 2014 to 55.62 per 100,000 in 2015 (entering the interval of 55.62–77.13 per 100,000) and then declined to 33.18 per 100,000 in 2020. This volatility is characteristic of transitional farming-pastoral ecosystems, where mixed livestock raising and agricultural practices create variable human-animal interaction risks, and sporadic livestock movement from high-incidence neighboring regions leads to periodic reintroduction of Brucella.

During 2021–2023, the spatial pattern of brucellosis incidence in Xinjiang stabilized while maintaining marked regional disparities, a reflection of the dual impact of sustained control measures and persistent ecological/socioeconomic heterogeneity. Tacheng and Altay (Northern Xinjiang) remained high-incidence areas, with incidence rates fluctuating within 24.27–33.82 per 100,000, indicating that core transmission zones persisted due to residual infection in the animal reservoir. Hami’s incidence rate continued to show volatility but did not return to its 2015 peak, remaining within 24.27–33.82 per 100,000, as control measures reduced (but did not eliminate) spillover transmission. In contrast, Hetian (1.19–1.76 per 100,000) and Kashi (3.45–5.58 per 100,000) (Southern Xinjiang) showed no significant increase in incidence throughout the entire study period (2010–2023). This stark North-South disparity is not a random spatial pattern, but a predictable outcome of differences in livestock density, ecological suitability for Brucella, and human-animal interaction practices—key determinants of brucellosis transmission identified in global and national epidemiological studies.

Local spatial clustering and outlier characteristics

From 2010 to 2023, the spatial distribution of brucellosis hotspots and cold spots in Xinjiang showed clear temporal dynamics, with most regions being categorized as Not Significant. In 2011 (Fig. 8-A), the Altay region emerged as a hot spot with 90% Confidence, although the clustering was relatively weak. In 2013 (Fig. 8-B), the Kashi region was identified as a Cold Spot with 90% Confidence, while most other regions remained not significant, indicating that the spatial clustering pattern was still at an early stage. In 2015 (Fig. 8-C), the Karamay region became a hot spot with 90% Confidence, whereas the Kashi region continued to exhibit a cold spot with 90% Confidence, showing a more differentiated spatial pattern. In 2016 (Fig. 8-D), the cold spot with 90% Confidence areas expanded to include the Kashi and Hetian regions, where the long-term incidence rates remained below 5.58 per 100,000, maintaining a stable low-incidence status without evidence of hotspot infiltration, while most other regions were not significant.

Fig. 8.

Spatial distribution of hotspots and cold spots of Brucellosis in Xinjiang, 2010–2023 (A: 2011; B: 2013; C: 2015; D: 2016; E: 2021; F: 2022; G: 2023)

During 2021–2023, the spatial clustering pattern evolved further: In 2021 (Fig. 8-E), the western region (e.g., parts of Kizilsu Kirghiz Autonomous Prefecture) emerged as a hot spot with 90% Confidence (orange) adjacent to a hot spot with 95% Confidence (red), marking the first appearance of a contiguous high-Confidence hotspot cluster in western Xinjiang. Meanwhile, northern-central areas maintained a “Cold Spot with 90% Confidence” status, reflecting a more polarized spatial differentiation. In 2022 (Fig. 8-F), the western hotspot cluster expanded, with the “Hot Spot with 95% Confidence” area enlarging, corresponding to a local incidence surge (reaching 42.7/100,000 in this region), while other areas remained predominantly not significant. By 2023 (Fig. 8-G), the western hotspot cluster showed a slight contraction: the Hot Spot with 95% Confidence area narrowed, and the adjacent region shifted to Hot Spot with 90% Confidence, while the northern-central cold spot persisted.

From 2010 to 2023, the H-L outliers, defined as the high-incidence areas surrounded by low-incidence neighbors, were a prominent feature of the spatial distribution of brucellosis in Xinjiang. In 2014 (Fig. 9-A), Tacheng (marked in yellow) was the core H-L outlier area, with an incidence rate of 159.83 per 100,000, surrounded by low-incidence areas such as Karamay (3.59 per 100,000, marked in white) and Urumqi (4.72 per 100,000), demonstrating a strong spatial contrast between the high-incidence center and its low-incidence surrounding. During 2015–2017 (Fig. 9-B-D), Karamay (marked in white) became a focal L-H outlier area, where the incidence rate in its surrounding areas remained relatively low, while Karamay itself showed obvious high-incidence characteristics compared with the surrounding areas, forming a clear L-H outlier pattern. From 2018 to 2019 (Fig. 9E-F), Tacheng (marked in yellow) again emerged as the core H-L outlier area, surrounding Karamay (marked in white), reflecting the recurrence of the pattern of high-incidence center surrounded by low-incidence areas similar to that in 2014, though with potential differences in intensity and spatial extent. In 2020 (Fig. 9-G), Tacheng (marked in yellow) remained an H-L outlier area, while Karamay in the surrounding area was classified as the Not Significant (gray) category, indicating a shift in local spatial relationships and changes in spatial heterogeneity of brucellosis incidence. By 2021 (Fig. 9-H), the spatial outlier pattern shifted markedly: the western region (e.g., Kizilsu Kirghiz Autonomous Prefecture) emerged as a H-H cluster (black), representing a contiguous high-incidence area surrounded by other high-incidence regions. This marked the first formation of a concentrated high-high cluster in Xinjiang, corresponding to a local incidence peak (reaching 48.3/100,000 in this area). In 2022 (Fig. 9-I), the H-H cluster expanded geographically, with adjacent regions also transitioning to H-H clusters, indicating a strengthened spatial aggregation of high-incidence areas in the west.

Fig. 9.

Results of local spatial autocorrelation analysis of Brucellosis in Xinjiang, 2010–2023 (A: 2014; B: 2015; C: 2016; D: 2017; E: 2018; F: 2019; G: 2020; H:2021; I: 2022)

Discussion

Scientific analysis and accurate prediction of human brucellosis incidence are critical for understanding the disease’s spatiotemporal transmission dynamics, identifying high-risk regions and populations, and designing evidence-based, region-specific One Health control strategies—especially in ecologically and socioeconomically diverse pastoral regions such as Xinjiang, where brucellosis remains a major public health and livestock husbandry threat. The spatiotemporal dynamics of brucellosis in Xinjiang (2010–2023), characterized by an initial concentration in Northern Xinjiang, expansion to Eastern Xinjiang and select regions, and eventual stabilization with persistent regional disparities, are closely associated with the interactions of human activities, animal husbandry practices, and environmental conditions. A key nuance of the 2021–2023 stabilization phase is that global Moran’s I analysis confirmed the overall spatial distribution of incidence remained statistically random (P > 0.48) at the autonomous region scale, while local indicator spatial association (LISA) analysis detected small-scale, contiguous high-Confidence hotspot clusters in western Xinjiang (e.g., Kizilsu Kirghiz Autonomous Prefecture). This discrepancy highlights that the observed clustering during 2021–2023 is localized to specific subregions and does not constitute a region-wide spatial pattern—an important distinction for interpreting the spatial epidemiology of brucellosis in Xinjiang. Such multi-factorial transmission patterns (with localized rather than global aggregation) are well documented for brucellosis and other zoonoses in pastoral settings, where livestock density, livestock management systems, and environmental suitability (e.g., temperature and vegetation) jointly shape human infection risk [26–32].

These dynamics align with the One Health framework [16], reflecting both local epidemiological characteristics and broader global trends of zoonotic disease transmission in pastoral ecosystems [2, 20]. Our analysis provides detailed evidence for region-specific control strategies based on long-term, prefecture-level surveillance data, consistent with findings in spatial epidemiology studies emphasizing the value of fine-scale analysis [21].

The initial concentration of high-incidence areas in Northern Xinjiang, specifically Tacheng and Altay, corresponds to China’s core livestock-producing hub. With sheep density exceeding 60 per km² and traditional extensive grazing practices [8], unprotected handling of infected animals and consumption of unpasteurized dairy products may facilitate Brucella transmission. This finding aligns with Wang et al.’s [9] national-level analysis, which identified large-scale livestock farming without adequate biosecurity as a key driver of epidemic surges between 2006 and 2015. Environmental and husbandry factors such as pasture grazing, herd mobility, and livestock output have been shown to be significantly associated with increased human brucellosis risk in China, particularly in traditional pastoral areas [26–28, 33, 34].

The subsequent expansion to Eastern Xinjiang, including Bayingolin Mongolian Autonomous Prefecture and Hami, likely reflects spillover from core pastoral regions mediated by environmental and socioeconomic factors.

Higher NDVI in Eastern Xinjiang supports increased livestock carrying capacity [8], while moderate temperatures (10–15 °C) and precipitation (200–400 mm) enhance Brucella survival in soil and water [20]. Studies indicated that brucellosis incidence correlates with ecological variables such as vegetation cover and climate, due to their influence on livestock distribution and pathogen survival, contributing to broader geographic spread [29, 35, 36].

Pronounced incidence fluctuations in Hami, from 7.94 per 100,000 in 2014 to 55.62 per 100,000 in 2015, mirror patterns in other mixed farming, –pastoral regions [6], highlighting the vulnerability of transitional ecosystems to zoonotic spillover.

The stabilization phase during 2021–2023, characterized by a rebound in overall incidence and persistent regional disparities, underscores the complex interplay of residual transmission from suboptimal control in small-scale unregulated farms [8, 9] and hypothesized external stressors (e.g., COVID-19-related disruptions to surveillance/interventions, climate variability)—factors identified in prior zoonotic disease research as potential modifiers of brucellosis transmission, but not quantitatively evaluated in this study. Declines in incidence in Northern and Eastern Xinjiang after 2015 correlate with the implementation of China’s National Brucellosis Control Program, including expanded livestock vaccination and culling of infected animals [9], yet residual transmission in unregulated farming systems sustained relatively high incidence in these core pastoral regions (24.27–33.82 per 100,000), a challenge also reported in Mongolia’s pastoral communities [8]. Notably, the current study does not provide empirical evidence for a causal link between COVID-19 disruptions/climate variability and the 2021–2023 rebound in Xinjiang; these remain testable hypotheses for future research integrating meteorological, public health resource, and brucellosis incidence data.

In contrast, Southern Xinjiang (Hetian, Kashi) maintained consistently low brucellosis incidence (< 5.58 per 100,000) from 2010 to 2023, a pattern shaped by both genuine reduced transmission risk and potential surveillance/diagnostic limitations, rather than a complete absence of exposure. Key drivers of lower transmission include an agriculture-dominant economy with livestock densities < 20 per km² a stark contrast to Northern Xinjiang’s pastoral systems—and an arid climate (annual precipitation < 100 mm) that curtails Brucella spp. environmental persistence [4]. Strict cross-regional livestock quarantine also blocks the introduction of infected animals from high-incidence northern/eastern areas. Additionally, underreporting may occur due to limited healthcare and diagnostic access in remote southern communities, which can miss mild/subclinical cases. Socioeconomic differences further reduce risk: small-scale backyard livestock rearing in the south involves far less contact with large herds or raw animal products than northern extensive grazing and trading practices. This North-South disparity underscores the need to account for both epidemiological and structural factors when interpreting brucellosis surveillance data in ecologically and socioeconomically diverse regions like Xinjiang.

This north–south disparity aligns with Liu et al.’s [7] nationwide analysis, emphasizing that zoonotic disease control strategies must account for regional ecological and socioeconomic heterogeneity.

The shift of brucellosis hotspots from Northern to Western Xinjiang (Kizilsu Kirghiz Autonomous Prefecture) and consolidation of cold spots in Southern and North-Central Xinjiang demonstrate evolving transmission dynamics shaped by One Health interactions. Western Xinjiang’s emergence as a contiguous high-Confidence hotspot cluster (2021–2023) may be driven by cross-border livestock trade, related to Xinjiang’s role in the Belt and Road Initiative.

Adjacent Central Asian countries (Kazakhstan, Kyrgyzstan) are endemic for brucellosis [2], and unregulated cross-border animal movement has been linked to zoonotic spillover in similar regions [19]. Abdrakhmanov et al.’s [19] analysis of foot-and-mouth disease outbreaks in Kazakhstan further supports the role of transboundary livestock trade in amplifying spatial aggregation of zoonotic pathogens.

Persistent cold spots in Southern Xinjiang (Hetian, Kashi) reflect the combined effects of low livestock density and environmental barriers. The arid climate limits Brucella survival in soil (< 1 month under dry conditions) [4], and strict cross-regional livestock quarantine may prevent introduction of infected animals [8]. These observations align with Corbel’s [4] global review, which identified arid ecosystems and limited livestock movement as key factors suppressing brucellosis transmission. Consolidation of cold spots in North-Central Xinjiang after 2016 correlates with improved biosecurity in large-scale farms, showing that targeted interventions can maintain sustained low-incidence status.

The transition of spatial outliers from single-point H-L (Tacheng/Karamay) to contiguous H-H clusters (western Xinjiang) reflects the evolutionary trajectory of brucellosis transmission, from localized, herd-driven outbreaks to environmentally mediated regional spread. Tacheng’s recurring role as an H-L outlier (2014, 2018–2019), where high-incidence pastoral areas were surrounded by low-incidence industrial/agricultural regions (Karamay, Urumqi), is consistent with Zhou et al.’s [16] One Health study linking H-L outliers to fragmented land use and uneven biosecurity implementation. The emergence of H-H clusters in western Xinjiang (2021–2022) marks a shift toward sustained, community-level transmission, likely influenced by environmental contamination (infected livestock waste), cross-border livestock movement, and sustained human–animal interactions.

This transition is comparable to brucellosis dynamics in Ethiopia’s camel-herding communities, where isolated outbreaks evolved into regional clusters due to inadequate waste management and limited vaccination coverage [4]. The temporary increase in spatial heterogeneity prior to overall incidence decline highlights the importance of adapting interventions to transmission stages: early detection and culling for H-L outliers, and integrated environmental disinfection, vaccination, and community education for H-H clusters.

Our findings emphasize the need for context-specific One Health strategies tailored to regional differences. In Northern and Eastern Xinjiang, scaling up livestock vaccination (S2 Strain, heat-stable for pastoral conditions) to at least 90%, implementing annual serological screening with compensated culling [6], and promoting protective measures during lambing are recommended. In Southern Xinjiang, maintaining strict cross-regional livestock movement controls, consolidating cold spot status through public health education, strengthening surveillance for H-H clusters, and collaborating with Central Asian countries on regional monitoring are advised.

A key limitation of this study is its use of Xinjiang-wide aggregated ARIMA modeling, which may mask fine-scale regional transmission heterogeneities across prefectures. Stratified regional time-series modeling was unfeasible due to incomplete, non-standardized prefecture-level surveillance data for 2010–2023, including missing early-phase monthly records and non-unified case reporting criteria. Though prefecture-level spatial autocorrelation analysis supplemented temporal modeling, it relied on annual (not monthly) data, precluding capture of regional temporal dynamics critical for short-term forecasting. While stratified ARIMA models for aggregated high/low-incidence regional groups were conducted, these still aggregate cross-prefecture data and fail to reflect prefecture-specific transmission drivers. Future research should develop disaggregated prefecture-level forecasting models, enabled by complete, standardized long-term monthly surveillance data for Xinjiang, to refine region-specific One Health interventions.

Xinjiang’s brucellosis dynamics align with global trends in pastoral regions, where livestock density, land use, and climate interact to shape transmission [2, 20]. The “concentration–expansion–stabilization” trend mirrors patterns in Inner Mongolia [7] and Mongolia [8], while the H-L to H-H shift resembles patterns in Ethiopia [4] and Kazakhstan [19]. These consistencies suggest the One Health framework proposed here can be applied to other pastoral ecosystems, with adaptation to local ecological and socioeconomic contexts.

Several important limitations of the ARIMA(2,1,1) forecasting model should be acknowledged. First, violation of model assumptions. Residual diagnostics revealed autocorrelation (Ljung–Box p < 0.001) and non-normality (Shapiro–Wilk p = 0.014), indicating incomplete capture of the data-generating process. Notably, Perfect normality is not strictly required for ARIMA forecasting; these deviations were modest in magnitude and did not compromise the model’s core utility for short-term brucellosis incidence forecasting evidenced by the model’s high adjusted R², close alignment of predicted and observed seasonal/annual trends, and acceptable in-sample fit metrics. Forecast intervals should therefore be interpreted with caution. Second, limited robustness. Sensitivity analyses showed deterioration in forecast accuracy when the model was trained on truncated datasets. RMSE increased from 130.7 (full model, in-sample) to 357.0 when trained on 2010–2020 and validated on 2021–2023, demonstrating poor generalizability under changing epidemiological conditions. We recommend that the long short-term memory (LSTM)/Bayesian Structural Time Series (BSTS) models be used for performing time series analysis with long-term dependencies. Third, absence of exogenous predictors. The model relies solely on historical case counts and omits known drivers of brucellosis transmission (e.g., livestock density, vaccination coverage, climate, mobility). This likely contributes to residual autocorrelation and limits the model’s utility for mechanistic inference or intervention planning. Fourth, spatial aggregation. Province-level data may mask heterogeneity at finer scales, potentially obscuring local transmission dynamics relevant for targeted control.

To address these limitations, future studies should integrate exogenous variables, evaluate non-linear and machine learning approaches, and develop forecasting models at finer spatial resolutions. Despite these constraints, the ARIMA(2,1,1) model provides a transparent, reproducible baseline for short-term forecasting under stable conditions.

Conclusion

The spatiotemporal patterns of brucellosis in Xinjiang (2010–2023) reflect complex interactions among livestock practices, environmental conditions, and public health interventions. The shift from isolated hotspots to contiguous clusters underscores the need for adaptive surveillance and integrated interventions. Leveraging spatial epidemiology to target high-risk regions and transmission stages enables Xinjiang to achieve sustainable brucellosis control, serving as a model for zoonotic disease management in pastoral economies globally, in line with One Health priorities.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Bahetibieke·Tuohetaerbaike and Mehdi Borhania: Conceptualization, data collection, statistical analysis, and draft preparation. Wenbao Zhang: Methodology, spatial analysis supervision, and manuscript revision. Hao Wen: Project supervision, funding acquisition, study design, and final manuscript approval. All authors read and approved the final version of the manuscript.

Funding

This study was sponsored by Natural Science Foundation of Xinjiang Uygur Autonomous Region: Study on the Dynamic Mechanism and Intervention of Pulmonary Fibrosis Mediated by cGAS-STING Signaling Pathway via Autophagy-Inflammation Imbalance (Project No.: 2025D01C321), University Scientific Research Program of Xinjiang Uygur Autonomous Region (2025): Construction of Multi-omics Atlas of “Tuberculosis Resisters” and Study on Their Specific Immune Mechanisms (Project No.: XJEDU2025J061), Key Research and Development Program of Xinjiang Uygur Autonomous Region: Study on Etiology, Epidemiological Characteristics and Transmission Laws & Mechanisms of Major Zoonoses between Humans and Animals (Project No: 2024B03021-1). Also, Research and Innovation Team for the Development of Antibody and Small-Molecule Therapeutic Drugs Against Important Zoonoses in Xinjiang (No. 2023TSYCTD0017). The funder had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

The data used in this study are available from public and institutional sources. Data on human brucellosis incidence from January 2010 to December 2020 were obtained from the National Public Health Data Center (http://www.phsciencedata.cn/Share/). Data from January 2021 to December 2023 were retrieved from the official website of the Xinjiang Health Commission and the Xinjiang Statistical Yearbook. The ARIMA model code and spatial analysis datasets used in the study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of Xinjiang Medical University. All data analyzed were de-identified to protect patient confidentiality, and aggregated surveillance data were used throughout the study. No individual-level data or identifying information was accessed, so informed consent from participants was not required. The study was conducted in compliance with the Declaration of Helsinki and relevant national ethical regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.