Spatiotemporal heterogeneity and socioeconomic determinants of syphilis incidence in Xining, Northwest China: a geospatial analysis

Abstract

Background

Syphilis remains a persistent public health challenge, and emerging evidence suggests that macro-level socioeconomic determinants contribute to its spatiotemporal heterogeneity. This study analyzed reported syphilis cases from Xining City, China, between 2008 and 2024 to characterize spatiotemporal patterns and quantify the influence of socioeconomic determinants on regional incidence.

Methods

We first examined the temporal trajectory of syphilis using joinpoint regression to identify significant changes in annual case numbers. Spatial patterns were assessed with global and local Moran’s I, and kernel density estimation was used to generate continuous intensity surfaces of case concentration based on geographic coordinates. To address multicollinearity among twelve socioeconomic indicators while accounting for spatial heterogeneity, we employed geographically weighted principal component analysis (GWPCA) using GWmodelS 1.0. Components with eigenvalues greater than one were retained, capturing 74.6–75.5% of variance for economic status, 81.5–85.2% for educational level, 90.4–91.3% for healthcare development, and 56.1–67.2% for communication infrastructure. These spatially varying component scores were then incorporated as predictors in a geographically and temporally weighted regression (GTWR) model. Finally, an autoregressive integrated moving average (ARIMA) model was developed to forecast incidence trends for 2025 to inform future intervention needs.

Results

From 2008 to 2024, syphilis incidence in Xining rose from 1.59 to 6.14 per 100,000. Joinpoint regression identified two phases: a sharp increase during 2008–2013 (APC = 43.46%, p < 0.05) and a slower sustained increase during 2013–2024 (APC = 5.11%, p < 0.05). Incidence rose across all age groups, most rapidly among those aged ≥ 60 years, with sustained increases in both sexes. Spatially, the highest burden concentrated in eastern Xining. Kernel density surfaces revealed a rightward shift and rising peaks, indicating citywide growth and widening regional disparities. Global Moran’s I confirmed significant spatial clustering (I = 0.217, p < 0.05). The Bayesian spatial model further validated this pattern, showing the highest relative risks in the four core urban districts (RR range: 3.17–3.47) and the lowest in Datong County (RR = 1.75), with a spatial random effect variance of 0.32 (95% CI: 0.12–0.55) confirming clustering independent of covariates. GTWR identified positive associations with economic status, healthcare, and communication, while education’s effect varied spatially. The ARIMA model projected a continued decline through 2025, conditional on current efforts.

Conclusion

Syphilis incidence in Xining has increased substantially and exhibits pronounced spatial and temporal heterogeneity, with macro-level socioeconomic determinants significantly shaping transmission dynamics. The temporal trends provide the broader context for understanding these spatial patterns, and the projected decline underscores the need for sustained, spatially tailored public health interventions in high-risk areas.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-026-13237-2.

Text box 1. Contributions to the literature
• This study demonstrates that place of residence and area-level socioeconomic conditions are strongly associated with the spatiotemporal dynamics of syphilis in western China, providing evidence to guide spatially targeted interventions in high-burden, resource-limited settings.
• It is the first to apply GTWR in STI research in highland regions, capturing the dynamic, non-stationary relationships between socioeconomic determinants and syphilis risk over space and time.
• The findings reveal that the protective effect of higher education on syphilis risk is not uniform but context-dependent, underscoring the importance of local conditions in shaping disease transmission and the need for precision public health approaches.

Introduction

Syphilis, a chronic systemic infection caused by Treponema pallidum, progresses through primary, secondary, tertiary, latent, and congenital stages and can lead to severe cardiovascular and neurological complications if untreated [1, 2]. In China, syphilis is a Class B notifiable infectious disease and consistently ranks among the most frequently reported infections nationally [3]. Although penicillin remains highly effective, incidence has remained persistently elevated in recent years, posing ongoing challenges for prevention and control [4, 5]. Qinghai Province reports notably higher syphilis incidence than the national average, suggesting region-specific transmission drivers. Understanding how macro-level socioeconomic determinants shape local transmission patterns is therefore essential for designing targeted interventions and optimizing resource allocation.

Spatial epidemiology provides a framework for characterizing disease distribution and its associations with contextual factors [6–8]. A growing body of evidence indicates that macro‑level socioeconomic factors—such as economic development, urbanization, and transportation and communication infrastructure—play a crucial role in the transmission dynamics of sexually transmitted diseases [9–13]. Higher economic development and urbanization are often linked to increased syphilis incidence, likely due to the conditions they create that facilitate transmission [9]. However, the influence of these factors may vary across space and time due to regional heterogeneity in social contexts [14]. Elucidating such spatiotemporal variation is critical for informing locally adaptive public health strategies [15].

Recent advances in spatial epidemiology have seen the increasing application of both Bayesian hierarchical spatiotemporal models and regression‑based approaches to uncover heterogeneity in infectious disease dynamics [16–18]. While Bayesian models are widely used for inference and prediction with robust uncertainty quantification, they are less suited to directly modeling how covariate effects vary continuously across space and time. Traditional ordinary least squares (OLS) regression cannot address spatial heterogeneity, and although geographically weighted regression (GWR) accommodates spatially varying coefficients, it does not incorporate temporal variation [19]. The geographically and temporally weighted regression (GTWR) model addresses both dimensions simultaneously by directly estimating locally varying coefficients in space and time, making it well suited for exploratory analysis of how socioeconomic associations differ across locations and periods. For example, Zhang et al. applied GTWR to examine heterogeneous relationships between socioeconomic and meteorological factors and hand, foot, and mouth disease in Sichuan and Chongqing, China [20]. Yu et al. used GTWR to assess macro‑level determinants of tuberculosis nationally [21]; and Hu et al. demonstrated its utility for mumps surveillance and policy support [22]. Despite the growing use of GTWR in China, syphilis in western regions—particularly in areas with distinctive socioeconomic profiles such as Qinghai Province—remains understudied. To address sample size concerns, we subsequently employed a Bayesian BYM model as a robustness check to validate the stability of our findings.

To address this gap, we selected Xining, the capital of Qinghai, as the study area. Xining has experienced rising syphilis rates in recent years and exhibits socioeconomic characteristics distinct from eastern Chinese cities. Using 17 years of district‑level syphilis case data (2008–2024) and corresponding macro-level socioeconomic indicators, this study aims to: (1) characterize the spatiotemporal evolution of syphilis incidence; (2) quantify the spatially and temporally varying effects of socioeconomic determinants using GTWR; and (3) compare the performance of GTWR with OLS and GWR to demonstrate its advantages in capturing spatiotemporal heterogeneity. The findings are expected to inform targeted, evidence-based prevention strategies and contribute to the broader application of spatiotemporal methods in sexually transmitted disease research.

Materials and methods

Study area

Figure 1 illustrates that Xining lies between 36°02′N and 37°28′N latitude and 100°52′E and 101°54′E longitude. The city governs a total area of 7,660 km², encompassing five urban districts—Chengdong, Chengxi, Chengzhong, Chengbei, and Huangzhong—as well as two counties, Datong and Huangyuan, which together comprise 76 townships. As of the end of 2024, Xining’s permanent resident population stood at 5.39 million, making it the only urban center on the Qinghai–Tibet Plateau with over one million inhabitants. Located within the “Source of Three Rivers” region—often referred to as China’s “Water Tower”—Xining plays a key role as an operational hub for national efforts to strengthen ecological security barriers. The administrative boundary data used in this study were obtained from the National Geospatial Information Public Service Platform (https://cloudcenter.tianditu.gov.cn/administrativeDivision/), under official map approval number GS(2024)0650.

Fig. 1.

Geographic situation of Xining city in China

Data collection and research indicators

The syphilis case data used in this study, spanning from 2008 to 2024, were obtained from the National Notifiable Infectious Disease Reporting System (NNIDRS) of China [23]. All cases were diagnosed and confirmed according to the syphilis diagnostic criteria issued by the Ministry of Health of the People’s Republic of China [24]. In compliance with national regulations, all medical institutions at or above the county level are required to submit an online report within 24 h of confirming a syphilis diagnosis. The reported data are regularly reviewed and quality-controlled by disease prevention and control centers at various levels to ensure completeness, accuracy, and reliability. We created the syphilis case database using Microsoft Excel 2020, with all personal information anonymized and handled in strict adherence to medical ethics to protect patient privacy. In our study, geographic data correspond to the residential addresses reported by individuals at the time of diagnosis.

Previous studies have shown that sexually transmitted disease incidence is influenced by factors such as economic conditions, education, healthcare, and communication infrastructure [25–28]. Accordingly, this study selected four macro‑level domains—economic status, educational level, healthcare development, and communication infrastructure—as potential determinants of the syphilis epidemic in Xining. As summarized in Table 1, economic status was measured using three indicators: local government general budget revenue, per capita disposable income, and per capita urban and rural household savings. Educational level was assessed through the number of regular primary and secondary schools, the number of teachers in these schools, and total student enrollment. Healthcare development was evaluated by the number of health professionals, licensed physicians, and healthcare institutions. Communication infrastructure was measured by the number of fixed‑line telephone subscribers, mobile phone subscribers, and broadband internet subscribers. All data were obtained from sources at the appropriate administrative levels, specifically the Xining Statistical Yearbook, the China Statistical Yearbook for County‑Level Units, and the Qinghai Statistical Yearbook. To ensure spatial alignment, all variables were compiled and aggregated to match the administrative boundaries of Xining’s districts and counties. Although ecological analyses of this type cannot establish causality at the individual level, the use of administrative-unit data from yearbooks that correspond to the geographic scale of analysis minimizes the risk of ecological fallacy and supports the validity of the observed spatial associations.

Table 1.

Measurement indicators of influencing factors

Macro-influencing factor	Measure Index
Economic status	X1: Local government general budget revenue
	X2: Per capita disposable income
	X3: Per capita urban and rural household savings
Educational level	X4: Number of regular primary and secondary schools
	X5: Number of teachers in regular primary and secondary schools
	X6: Enrollment in regular primary and secondary schools
Healthcare development	X7: Hospital beds per capita
	X8: Number of health professionals
	X9: Number of licensed physicians
Communication infrastructure	X10: Number of fixed-line telephone subscribers
	X11: Number of mobile phone subscribers
	X12: Number of broadband internet subscribers

Joinpoint regression analysis

Joinpoint regression models evaluate changes in incidence trends by identifying segmented patterns that fit the data significantly better than a single linear trend derived from Poisson regression or standard time series approaches [29]. This method determines the optimal number of joinpoints—time points at which the trend direction changes significantly—thereby capturing meaningful shifts in incidence over time. Specifically, it pinpoints the calendar years in which statistically significant abrupt changes in temporal trends occur. We used Annual Percent Change (APC) to quantify syphilis number variations within specific time intervals and calculated the Average Annual Percent Change (AAPC) across the entire study period to assess long-term trends. Statistical significance of the overall trend was determined by evaluating whether the 95% confidence interval of the AAPC excluded zero, corresponding to a p-value less than 0.05. For this analysis, we used Joinpoint software (version 4.9.1.0; National Cancer Institute, Rockville, Maryland, USA; https://surveillance.cancer.gov/joinpoint/download).

Global spatial autocorrelation

Global spatial auto-correlation analysis is conducted at the national scale to assess whether syphilis incidence across China’s provincial-level administrative units exhibits statistically significant spatial clustering [30]. This approach compares each province’s observed incidence rate with the national average to quantify the overall degree of spatial association. ArcGIS 10.8 was used to perform global spatial auto-correlation analysis. The primary measure used is Moran’s I index, defined as:

where n is the number of spatial units, and are the incidence values at locations i and j, denotes the national mean, and represents the spatial weight between units i and j. Moran’s I ranges from − 1 to + 1: values near + 1 suggest strong positive spatial clustering, those near − 1 indicate dispersion, and values around 0 imply a random spatial pattern.

Bayesian spatial analysis

Given the limited number of districts (n = 7) available for analysis, we adopted a two-stage analytical approach to ensure robust inference. First, we conducted exploratory spatial data analysis using global Moran’s I to assess overall spatial auto-correlation, while acknowledging the statistical power limitations inherent to such a small sample. Second, to obtain reliable risk estimates and overcome the instability associated with small-area counts, we employed a Bayesian hierarchical model using the Besag-York-Mollié (BYM) specification [31]. This approach accounts for both spatial correlation and unstructured heterogeneity by borrowing information from neighboring areas, thereby stabilizing estimates. GTWR was not pursued at this spatial scale due to the insufficient number of spatial units required for stable local estimation, consistent with methodological recommendations in the literature. Penalized complexity (PC) priors were assigned to the precision parameters of the random effects. The model was fitted using the INLA package (R version 4.5.2). Posterior means and 95% credible intervals for the relative risks were mapped to identify areas with significantly elevated risk. Convergence was assessed by inspecting the stability of hyperparameter estimates [32].

Kernel density estimation

Kernel density estimation (KDE) is a nonparametric method that examines syphilis incidence patterns without assuming a specific data distribution. By using observed values, KDE captures key distribution characteristics—such as central tendency, shape, spread, and modality—which are crucial for analyzing spatial imbalances. In this study, we applyed KDE to analyze syphilis trends in Xining over time. Following Wang et al. [33], we compare temporal density curves to shifting dynamics: shifts in the distribution’s position reflect changes in average incidence, while variations in peak height and width indicate township level disparity. The spread of the distribution highlights the gap between high-and low-incidence regions, and multiple peaks suggest increased epidemic polarization. We used MATLAB 8.2 for analysis and visualization. The method’s core principle is as follows:

where K represents the kernel function, h represents bandwidth parameter, n represents sample size, and represents data points.

Geographically weighted principal component analysis

GWPCA was used to generate composite spatial indices from the macro-level socioeconomic variables described above. This method extends conventional PCA by accounting for spatial nonstationarity. While standard PCA captures the global structure of a dataset, it assumes a stationary covariance structure and may therefore mask local variations [34]. GWPCA addresses this limitation by constructing a local variance–covariance matrix for each observation based on its geographic coordinates, thereby accommodating spatial heterogeneity [35]. This approach enables several analytical functions: identification of regions with homogeneous multivariate characteristics, exploration of local eigenvalues to assess spatial variation, and evaluation of variable contributions through location‑specific eigenvectors. It also reduces the dimensionality of the original variables into a set of uncorrelated local components [36]. GWPCA was performed using the GWmodelS software package. A detailed description of the equation is provided in Supplementary File 1.

Correlation analysis and collinearity diagnosis

Multicollinearity occurs when two or more predictor variables are highly correlated, making it difficult for a model to isolate their individual effects on the dependent variable. If unaddressed, this issue can bias parameter estimates, reduce the reliability of regression coefficients, and compromise the validity of conclusions drawn from the analysis. In this study, we first assessed pairwise associations between macro-level determinants and syphilis incidence using Spearman’s rank correlation in R 4.5.2. To minimize the influence of multicollinearity on model estimation, we employed an iterative elimination approach by first computing variance inflation factors (VIFs) for all candidate variables and identifying those with VIF greater than 10 as indicating substantial multicollinearity; subsequently removing the variable with the highest VIF in sequential steps while refitting the model after each elimination until all remaining variables had VIF values below the threshold. The final predictor set was then selected based on model fit criteria (AIC/BIC), cross-validation prediction errors, and epidemiological relevance to achieve an optimal balance among statistical robustness, predictive accuracy, and result interpretability.

Geographically and temporally weighted regression

The GTWR model extends the GWR framework by incorporating temporal dynamics, utilizing spatiotemporal kernel functions to simultaneously capture variations in regression coefficients across space and time, thereby more accurately representing the dynamic heterogeneity of disease risk factors. This extension enhances model flexibility in handling time-dependent data, enabling precise and dynamic mapping of evolving trends [37, 38]. By producing localized, fine-scale coefficient estimates for specific locations and time points, GTWR achieves high-precision fitting for datasets exhibiting strong temporal dependencies. Consequently, it improves the accuracy and reliability of predictions and offers a more robust analytical framework for research and evidence-based decision-making. The model provides the most granular representation of coefficient dynamics by simultaneously modeling their variation across geographic space and time. Its mathematical formulation is given as:

In this formulation, denotes the observed value of the incidence of syphilis for the i sample. The spatiotemporal coordinates represent the geographic location of counties in Xining and time point of observation . The term is the local intercept, and represents the set of coefficients for predictor , which vary across both space and time; is the error term. A key strength of the GTWR model is its capacity to capture spatiotemporal heterogeneity in regression parameters through these adaptive, location- and time-specific coefficients. Model calibration relies on a spatiotemporal weight matrix W-a diagonal matrix that assigns a unique weight to each observation based on its proximity in both space and time. Optimal bandwidths for weighting are selected via cross-validation to maximize goodness-of-fit, using locally weighted least squares estimation. All data preprocessing was performed in R (4.5.2), and the GTWR analysis was implemented with the GTWR extension in ArcGIS 10.8.0 software [39].

ARIMA model

The ARIMA (Autoregressive Integrated Moving Average) model is ideal for time series data exhibiting trends, seasonal patterns, or autocorrelation [23]. It handles non-stationarity by incorporating three components: autoregressive (AR), differencing (I), and moving average (MA), which together capture the lagged dependencies and random fluctuations in the data. In this investigation, the ARIMA model was performed using R 4.5.2 software. The model is represented as:

denotes a white noise sequence with a zero mean. Our model is characterized by three parameters: p, d, and q, which represent the order of the auto-regressive part, the number of differencing steps required to achieve stationarity, and the order of the moving average part, respectively. When the series exhibits seasonal patterns, a multiplicative seasonal extension—denoted as ARIMA(p, d,q)(P, D,Q)_s—is used, where P, D, and Q represent the seasonal auto-regressive, differencing, and moving average orders, and s is the length of the seasonal cycle.

Analytical framework

The analysis proceeded in three stages. First, we conducted a descriptive temporal analysis using joinpoint regression to identify significant changes in the annual trend of syphilis cases and ARIMA modeling to forecast incidence for 2025–2026. Building on this temporal perspective, we then examined the spatial distribution of incidence through choropleth mapping and global spatial autocorrelation statistics, and applied kernel density estimation to assess shifting trends. To validate the stability of the global spatial autocorrelation identified, we incorporated the BYM model. Based on these spatial characterizations, we finally employed geographically weighted principal component analysis and spatial regression models to identify the socioeconomic determinants underlying the observed spatial patterns.

Results

The spatial-temporal evolution of the incidence of syphilis in Xining, China from 2008 to 2024

Between 2008 and 2024, reported syphilis cases and incidence rates in Xining City showed a consistent upward trend, reflecting a progressive increase in disease burden. Despite occasional declines in certain years, the overall pattern of increase was evident. The reported incidence rate rose from 1.59 per 100,000 in 2008 to 6.14 per 100,000 in 2024, reaching the highest level observed during the study period. In terms of spatial distribution, incidence rates generally increased across all districts and counties in the city. Chengdong, Chengzhong, Chengbei, Chengxi, and Huangzhong districts formed the high incidence areas, with Chengdong District having the highest rate. In contrast, Datong County and Huangyuan County consistently maintained relatively low incidence rates. The annual incidence maps are presented in the supplementary material in Figure S1. The spatiotemporal heterogeneity pattern of syphilis incidence in the city is shown in Fig. 2. A detailed summary of the demographic characteristics of syphilis patients in Xining is also provided in the supplementary material in Table S1.

Fig. 2.

The spatial-temporal evolution of the incidence of syphilis in Xining, China from 2008 to 2024

Joinpoint regression analysis

Figure 3 shows the temporal trend of syphilis case numbers in Xining from 2008 to 2024. Trend analysis in Fig. 3a indicates a significant overall increase in the annual number of syphilis cases, with both stages were statistically significant (P < 0.05). The number of syphilis cases in both males and females also showed a continuous upward trend. Age group analysis revealed an increase in the number of syphilis cases across all age groups, with the most rapid growth observed in those aged 60 and above. Figure 3b shows that before 2013, the APC was as high as 43.46%, and after 2013, the growth rate (APC = 5.11%) slowed but continued to increase. Overall, the trend in all age groups was consistent with the citywide trend, showing a long-term increase.

Fig. 3.

Changing trend in the number of syphilis in Xining, China, 2008–2024: (a) changing trend in syphilis cases in different gender in Xining, China, 2008–2024; (b) changing trend in syphilis cases in different age groups in Xining, China, 2008–2024

Spatial autocorrelation analysis and Kernel density estimation

The Moran’s I index was used to analyze the spatial auto-correlation of syphilis incidence in Xining from 2008 to 2024. As shown in Table 2, the global spatial auto-correlation test revealed significant positive spatial correlation for the average annual syphilis incidence (Moran’s I = 0.217). Positive correlations (Moran’s I > 0, p < 0.05) were observed in 2008–2016 and 2020.

Table 2.

Global spatial autocorrelation analysis of syphilis incidence in Xining City, China in 2008–2024

Year	Moran’I	S.D.	Z-score	P-value
2008	0.169	0.016	2.589	0.009
2009	0.158	0.017	2.474	0.013
2010	0.157	0.018	2.424	0.015
2011	0.149	0.018	2.329	0.020
2012	0.102	0.018	1.995	0.046
2013	0.125	0.018	2.175	0.030
2014	0.150	0.018	2.376	0.018
2015	0.137	0.018	2.285	0.022
2016	0.120	0.018	2.126	0.034
2017	0.117	0.024	1.828	0.068
2018	-0.017	0.012	1.384	0.166
2019	-0.015	0.020	1.073	0.283
2020	0.037	0.009	2.130	0.033
2021	-0.095	0.022	0.484	0.628
2022	-0.003	0.008	1.756	0.079
2023	-0.344	0.008	-1.911	0.056
2024	-0.102	0.020	0.464	0.164
Average	0.217	0.023	2.512	0.012

To examine the spatiotemporal evolution of syphilis incidence in Xining, we applied KDE in MATLAB. The resulting three-dimensional surface (Fig. 4) visualizes changes in the distribution of incidence values over time, with the X-axis representing syphilis incidence, the Y-axis representing year, and the Z-axis representing kernel density. Over the study period, the density curve shifted progressively rightward, indicating an overall increase in incidence. The persistent right-skewed shape of the distribution reflects substantial inter-regional variation and the existence of areas with markedly higher incidence. The emergence and intensification of multiple peaks over time suggest a trend toward multipolar differentiation, with growing disparities in incidence across the city. Increasing peak heights and sharper curve morphology further indicate widening spatial inequality. These patterns collectively illustrate both the overall upward trend and the increasing spatial polarization of syphilis incidence in Xining.

Fig. 4.

Kernel density estimation of syphilis cases in Xining, China, 2008–2024

Robustness check: Bayesian spatial model

The BYM model confirmed significant spatial heterogeneity in syphilis risk across the seven districts. Table 3 presents the posterior mean relative risks (RR) at the district/county level. The posterior probability of elevated risk (RR > 1) was 1.00 for all districts, indicating strong evidence that each district’s risk exceeds the regional average. The four core urban districts—Chengzhong, Chengbei, Chengxi, and Chengdong—exhibited the highest relative risks, with posterior means ranging from 3.17 to 3.47. In contrast, Datong County showed the lowest relative risk (RR = 1.75, 95% CrI: 1.70–1.79). This spatial gradient (core urban > suburban > rural) was fully consistent with the high-risk areas identified by both global Moran’s I and the GTWR model, while providing more reliable and stabilized risk estimates by accounting for spatial dependence and borrowing strength across neighboring units. The spatial random effect variance had a posterior mean of 0.32 (95% CI: 0.12–0.55), confirming the presence of meaningful spatial clustering independent of the covariates. Model diagnostic criteria indicated a good fit (DIC = 80.86, WAIC = 78.70, CPO = 0.001). The stability of the model was further supported by the convergence of hyperparameter estimates and the consistency of results across alternative prior specifications.

Table 3.

Posterior summaries of district-specific RR from the Bayesian BYM model

district	RR	RR lower	RR upper	P
Chengdong	3.17	3.05	3.29	1
Chengzhong	3.47	3.32	3.63	1
Chengxi	3.32	3.17	3.47	1
Chengbei	3.33	3.19	3.48	1
Datong	1.75	1.70	1.79	1
Huangzhong	2.38	2.30	2.46	1
Huangyuan	2.33	2.21	2.47	1

Geographically weighted principal component analysis and multicollinearity diagnosis

To reduce the dimensionality of the 12 socio-economic indicators while accounting for spatial heterogeneity, we employed GWPCA using GWmodelS 1.0 software. GWPCA extends traditional PCA by allowing the principal component structure to vary across space, thereby capturing locally specific relationships among variables. Prior to analysis, all indicators were standardized to zero mean and unit variance to ensure comparability. The optimal number of components was determined based on the cumulative proportion of variance explained, with components having eigenvalues greater than one retained. The analysis revealed that the first principal component effectively captured the essential information from the original variables, explaining over 60% of the total variance across all four domains. Specifically, the cumulative variance contribution of the first principal component ranged from 74.56% to 75.52% for economic status, from 81.53% to 85.22% for educational level, from 90.41% to 91.30% for healthcare development, and from 56.07% to 67.19% for communication infrastructure (Table 4). The local loadings for each component were mapped to assess spatial variation in the influence of individual indicators. The component scores for each district were then extracted and used as explanatory variables in subsequent models. To assess multicollinearity among these factors, we conducted a multicollinearity test on the first principal component using SPSS 27.0. The results indicated that tolerance values for all four domains were greater than 0.1, and the variance inflation factor (VIF) values were below 10, confirming that multicollinearity was not a concern. This approach ensures that the derived dimensions are empirically driven by the data rather than subjectively defined, while also accounting for spatial non-stationarity in the underlying socio-economic structure.

Table 4.

Results of GWPCA and collinearity diagnosis

Factor	GWPCA					Collinearity Diagnosis
	Local Proportion of Variance (%)
	Min	1st Qu	Median	3rd Qu	Max	Tolerance	VIF
Economic status	74.560	74.753	74.817	75.512	75.523	0.580	1.723
Educational level	81.525	82.469	84.951	85.044	85.218	0.130	7.668
Healthcare development	90.414	90.492	90.515	91.256	91.298	0.183	5.473
Communication infrastructure	56.065	61.538	61.919	62.345	67.188	0.291	3.436

Geographically and temporally weighted regression analysis

To explore the influence of various factors on syphilis incidence and evaluate the suitability of the model selection, syphilis incidence was set as the dependent variable, with economic status, education level, healthcare development, and communication infrastructure as independent variables. Four models—OLS, GWR, TWR, and GTWR—were developed using ArcGIS 10.8, and the results are shown in Table 5. The GTWR model exhibited the smallest residual squares, sigma, and AICc, while the adjusted R² was the highest, indicating that GTWR provided the best fit and highest explanatory power.

Table 5.

Comparison of model parameters

	OLS	GWR	TWR	GTWR
Bandwidth	——	0.119	0.285	0.128
Residual Squares	314042.093	171,218	222,223	82616.2
Sigma	——	37.932	43.214	26.349
AICc	1285.208	1250.48	1261.72	1207.07
R2	0.547	0.755	0.682	0.882
R2Adjusted	0.547	0.747	0.671	0.878

P.S. Both R² and adjusted R² represent the goodness of fit of the equation, with higher values indicating better fit. AICc refers to the corrected Akaike Information Criterion, with lower values indicating a better fit of the data. RSS stands for the residual sum of squares, where smaller values suggest a better model fit. A “−” denotes the absence of a value. The bandwidth value is used to estimate the local bandwidth, controlling the smoothness of the model

To further explore the influence of various factors on the incidence of syphilis, we employed a GTWR model and visualized the regression coefficients, as shown in Table 6, Figs. 5 and 6. Economic status had a consistently positive effect across all regions, with higher levels of economic development correlating with higher syphilis incidence rates, a trend maintained from 2008 to 2024. The impact of educational level varied by region. In the Eastern, Central, Western, and Northern districts, higher educational attainment was associated with lower syphilis incidence, whereas in Datong County, Huangzhong District, and Huangyuan County, a positive correlation was observed. Between 2008 and 2014, and again from 2020 to 2024, higher educational levels correlated positively with syphilis incidence, while from 2015 to 2019, the relationship was negative.Healthcare development generally had a positive impact on syphilis incidence in most regions, with the exception of Datong County, where a negative correlation was found. From 2008 to 2019, healthcare development showed a positive impact, but from 2020 to 2024, it had a negative effect. Communication infrastructure, with the exception of Datong County, showed a positive impact on syphilis incidence in all regions. From 2008 to 2011, communication infrastructure had a negative impact, while from 2012 to 2024, it showed a positive effect.

Table 6.

Results of GTWR

Factor	Mean	Min	Med	Max	Proportion of p-values<0.05
Economic status	36.547	-12.824	31.680	99.058	100.00%
Educational level	6.890	-84.533	16.873	80.165	100.00%
Healthcare development	21.806	-105.552	36.219	101.160	100.00%
Communication infrastructure	10.372	-52.207	15.897	62.926	99.14%

Fig. 5.

Distribution of regression coefficients for predictors of syphilis incidence in Xining, China: (a) Across geographic areas; (b) Over time (2008–2024)

Fig. 6.

The distribution of regression coefficients for factors on the incidence of syphilis from 2008 to 2024 in Xining, China: (a) Economic status; (b) Educational level; (c) Healthcare development; (d) communication infrastructure

The ARIMA model forecast results

The model was developed using the monthly syphilis incidence rates in Xining city from January 2008 to December 2024, as shown in Fig. 7. Model parameters were automatically identified with R software version 4.5.2, and the ARIMA(3,1,0)(1,0,1) model was selected as the best fit, resulting in an AIC of 1635.25 and a BIC of 1654.76. The Ljung-Box test yielded a Q-statistic of 11.944 (p = 0.450), confirming the residuals follow a white noise process. The static R² value of 0.959 indicates that the model explains a large portion of the variance in the original incidence trends. Using the ARIMA(3,1,0)(1,0,1) model, the monthly incidence rates in Xining for 2025 were predicted.

Fig. 7.

Forecast of syphilis incidence rate for 2025 in Xining, China

Discussion

Syphilis is a highly transmissible chronic infection caused by Treponema pallidum with variable clinical manifestations. Following its re-emergence in China in the late 1980s, the epidemic has evolved from localized outbreaks to a sustained increase, subsequently developing into widespread transmission [40]. It has become one of the most prevalent STDs in the country and poses significant public health challenges [41]. To elucidate the spatiotemporal dynamics and regional heterogeneity of syphilis transmission, our study analyzed the data on syphilis cases in Xining from 2008 to 2024 and applies a combination of geospatial analysis and the GTWR model to explore trends in incidence, spatial clustering patterns, and key determinants.

From 2008 to 2024, reported syphilis incidence in Xining exhibited an overall upward trend with periodic fluctuations. A marked increase between 2008 and 2014 coincided with improvements in surveillance capacity and the implementation of the national Syphilis Prevention and Control Plan (2010–2020), which enhanced case reporting through a web-based infectious disease system and expanded testing access [42]. In 2015, incidence declined for the first time, following the nationwide rollout of the Comprehensive Intervention Program for Prevention of Mother‑to‑Child Transmission of HIV, Syphilis, and Hepatitis B [43–45]. Subsequent years saw continued fluctuations but a sustained upward trajectory. The COVID-19 pandemic and associated control measures, particularly extended lockdowns in 2022, temporarily disrupted healthcare access and reduced testing willingness, contributing to a short-term decline in reported cases, a pattern observed elsewhere in China [46–48]. An ARIMA model developed in this study projects a decrease in syphilis incidence by 2025, suggesting initial effectiveness of current prevention strategies, although the model does not account for potential disruptions such as outbreaks or changes in disease dynamics [49, 50]. Spatial analysis revealed persistent clustering of syphilis in eastern Xining, an area characterized by rapid urbanization, high population density, and substantial migration—factors that collectively facilitate transmission [50]. The consistency between the classical spatial analyses and the Bayesian BYM model confirms that the observed spatial patterns are robust and not attributable to small sample size or overfitting. This concordance supports the prioritization of core urban districts for targeted interventions. The GTWR analysis revealed spatial heterogeneity in syphilis incidence across Xining, with macro factors exerting region-specific influences. Economic status demonstrated a consistently positive regression coefficient with syphilis incidence, attributable to multiple pathways: improvements in economic conditions typically coincide with an increase in living standards, particularly during urbanization, which leads to higher consumer spending and greater social interactions, potentially resulting in more frequent sexual activity. In this context, risky sexual behaviors, such as unprotected sex and frequent partner changes, may facilitate the transmission of sexually transmitted diseases like syphilis [51–53]. Additionally, the existing economic disparities, where the gap between the rich and the poor continues to widen, often push low-income and impoverished groups into high-risk sexual behaviors [54]. These populations, including sex workers [55], MSM [56], or those without adequate healthcare [57], tend to have lower health awareness and may lack effective syphilis prevention measures, thus contributing to the spread and higher incidence of syphilis. It is also noteworthy that Xining, as a tourism-oriented city with frequent population movement, experiences increased interpersonal contact opportunities, which have further exacerbated the risk of syphilis transmission.

Spatial analysis indicates the distribution of syphilis incidence patterns in Xining, with elevated rates persistently observed in eastern regions. These areas feature accelerated economic development and urbanization, high population density, substantial migrant populations, and rapid societal shifts in attitudes and lifestyles, collectively increasing the prevalence of high-risk sexual behaviors and close contact frequency, which facilitate pathogen transmission [51]. The GTWR analysis revealed spatial heterogeneity in syphilis incidence across Xining, with macro factors exerting region-specific influences. Economic status demonstrated a consistently positive regression coefficient with syphilis incidence, attributable to multiple pathways: improvements in economic conditions typically coincide with an increase in living standards, particularly during urbanization, which leads to higher consumer spending and greater social interactions, potentially resulting in more frequent sexual activity. In this context, risky sexual behaviors, such as unprotected sex and frequent partner changes, may facilitate the transmission of sexually transmitted diseases like syphilis [52–54]. Additionally, the existing economic disparities, where the gap between the rich and the poor continues to widen, often push low-income and impoverished groups into high-risk sexual behaviors [55]. These populations, including sex workers [56], MSM [57], or those without adequate healthcare [58], tend to have lower health awareness and may lack effective syphilis prevention measures, thus contributing to the spread and higher incidence of syphilis. It is also noteworthy that Xining, as a tourism-oriented city with frequent population movement, experiences increased interpersonal contact opportunities, which have further exacerbated the risk of syphilis transmission.

Secondly, the impact of the regression coefficient for educational level on syphilis incidence in Xining varies across regions. In the urban districts of Chengdong, Chengzhong, Chengxi, and Chengbei, the coefficient for educational level is negative, indicating that a higher educational level is associated with a lower risk of syphilis reports. This may be due to individuals with higher education generally possessing enhanced health literacy and risk awareness, leading to greater adoption of protective sexual behaviors that lower infection susceptibility [59]. However, in Datong County, Huangzhong District, and Huangyuan County, educational level shows a positive association, potentially reflecting increased screening participation among higher-educated individuals, resulting in more comprehensive case detection and reporting [60].

In addition, Healthcare development is a crucial factor influencing the incidence of syphilis in Xining, with its regression coefficient generally showing a positive effect, which indicates that stronger healthcare systems enhance epidemiological surveillance capacity and case detection [58, 61]. Higher healthcare levels typically indicate more abundant resources, including better timeliness and accessibility of diagnostic facilities [62]. Regions with robust medical resources exhibit higher reported infections due to improved service accessibility and timeliness, which expanded syphilis testing coverage and enabled earlier identification of cases [63]. This effect is more pronounced in areas characterized by limited resources and weaker baseline surveillance infrastructure. However, in Datong County, healthcare development is negatively associated with syphilis incidence, possibly due to a low baseline transmission levels and effective public health interventions such as expanded screening and standardized treatment protocols that successfully interrupted transmission chains, allowing enhanced healthcare capacity to translate directly into reduced disease burden.

Finally, the regression coefficient of Communication Infrastructure shows an overall positive association with the reported syphilis incidence in Xining, primarily attributable to its indirect influence on social behavior patterns. The widespread adoption of internet and mobile technologies has increased the use of social platforms, reduced spatiotemporal constraints on interpersonal interactions, and elevated the prevalence of seeking sexual partners online [64, 65]. Such relationships often involve information asymmetry and anonymity, leading to reduced condom use and unstable sexual partnerships that heighten infection risk [66]. Additionally, Communication Infrastructure facilitates the concealment of commercial sex activities by enabling the more discreet organization and operation of illegal transactions through digital platforms, thereby expanding the transmission pathways for syphilis and contributing to higher reported incidence rates [67].

In summary, based on the current syphilis situation in Xining, the following targeted syphilis prevention strategies are proposed: Firstly, there should be increased financial and technical investment in the western regions, particularly in high-incidence counties, to improve the primary healthcare system. Efforts should focus on enhancing the accessibility of syphilis screening, diagnosis, and standardized treatment, particularly through the implementation of rapid testing at township and community health facilities, thereby reducing undetected infections and reporting gaps [68]. Secondly, proactive screening and health follow-up should be strengthened for mobile populations, including migrants, youth, and commercial service workers, in areas with high population mobility [69, 70]. This should be achieved through intersectoral collaboration mechanisms that integrate syphilis control into routine health management to disrupt interregional transmission. Simultaneously, rapid testing and health management mechanisms should be promoted in areas with dense migration to sever the interregional transmission chain [71]. Finally, sexual and reproductive health education, incorporating syphilis and STDs awareness, should be systematically embedded in school curricula to improve adolescent self-protection capabilities [72]. Targeted interventions should be carried out in high-risk groups, promoting core preventive measures such as condom use and regular testing, along with focused awareness campaigns and psychological support, to establish a unified health education system based on knowledge, trust, and action, thereby improving overall prevention and control efforts.

Limitations

This study has several limitations. First, the analysis was conducted at the district level (n = 7), which constrains spatial inference. While global Moran’s I provided exploratory insight, its statistical power is limited with so few units. Methods requiring denser spatial data, such as kernel density estimation and geographically weighted regression, were therefore not appropriate. To address these concerns, we employed a BYM spatial model as our primary analytical tool, which stabilizes estimates through spatial borrowing and confirmed the patterns observed in exploratory analysis. Second, as an ecological study, we examined only macro-level socioeconomic determinants and did not incorporate meteorological, behavioral, or physiological variables that may influence syphilis transmission. Our findings should therefore be extrapolated to other areas with caution. Future studies with finer spatial resolution and broader individual-level determinants would enable more comprehensive analysis.

Conclusion

This study demonstrates spatial and temporal heterogeneity in syphilis distribution across county-level administrative units. The application of the GTWR model to examine associations between macro-level socioeconomic determinants and syphilis incidence-level socioeconomic determinants cannot fully account for the complexity of syphilis transmission, our findings provide quantitative evidence of their influence at refined spatiotemporal scales. The results can assist public health authorities in prioritizing preventive measures based on regional differences and guide the rational allocation of public health resources by the government.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AIC

Akaike Information Criterion

APC

Age-Period-Cohort model

ARIMA

Autoregressive Integrated Moving Average

DIC

Deviance Information Criterion

GTWR

Geographically and Temporally Weighted Regression

GWPCA

Geographically Weighted Principal Component Analysis

GWR

Geographically Weighted Regression

HFMD

Hand, Foot and Mouth Disease

KDE

Kernel Density Estimation

NNIDRS

National Notifiable Infectious Disease Reporting System

OLS

Ordinary Least Squares

PC

Principal Component

RR

Relative Risk

STDs

Sexually Transmitted Diseases

TWR

Temporally Weighted Regression

WAIC

Widely Available Information Criterion

Author contributions

Wei Li designed the study. Data collection was carried out by Ying Zhao and Lulu Fu. Data analysis was conducted by Yitaoren, Yongkai Shi and Jinxiong Lin. The manuscript was written by Yitao Ren, with all authors contributing to its revision. All authors approved the final manuscript.

Funding

This work was supported by “the leading academic team established by the Xining Center for Disease Control and Prevention” (Document No. Ningjikyeye [2025] 21).

Data availability

The patient-level data cannot be made publicly available due to the presence of sensitive personal information. Requests for additional information may be directed to the corresponding author.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Review Committee of the Xining Center for Disease Control and Prevention (Approval No. QHXNCDCLLSC–2025020) and conducted in accordance with the principles of the Declaration of Helsinki (2024 version). The study used retrospective, city-level aggregated surveillance data on syphilis cases, which contained no personally identifiable information and were classified as low-risk research involving anonymized secondary data. Given the fully anonymized and aggregated nature of the dataset, the Ethics Review Committee waived the requirement for written informed consent. All data handling and analysis were performed under institutional supervision, with strict measures in place to protect data confidentiality and privacy.

Consent for publication

Not applicable.

Author information

Not applicable.

Competing interests

The authors declare no competing interests.