Spatial heterogeneity and its influencing factors of cardiometabolic multimorbidity in a natural community population: a study based on Lingwu city, rural Northwest China

Abstract

Objective

Cardiometabolic multimorbidity (CMM) significantly contributes to the economic burden in China, particularly in rural areas. This study aimed to analyze the spatiotemporal distribution of CMM and identify its primary influencing factors in different townships in Lingwu City, Ningxia, to inform public health policies in Northwest China.

Methods

The standardized prevalence of CMM was investigated using data from Cardiovascular Disease High-Risk Group Early Screening and Comprehensive Intervention Program (2017–2022) conducted in Lingwu City, Ningxia. We applied spatial autocorrelation, cluster analysis, and spatiotemporal scanning to explore the spatiotemporal distribution characteristics of CMM and identify high-risk clusters. Four machine learning algorithms, logistic regression (LR), support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost) were developed using 15 major cardiovascular disease influence factors. The performance of these models was evaluated based on accuracy, precision, recall, and AUC to determine their applicability across different townships in Lingwu City. The optimal model was selected for further analysis using interpretable machine learning algorithms (SHAP analysis) to identify common and key influence factors influencing CMM prevalence across townships.

Results

Among the 11,353 participants, 1,334 individuals (11.8%, 95% CI: 11.2–12.4%) were diagnosed with CMM, with significant variations in influence factors observed across townships (P < 0.05). Trend surface analysis revealed a parabolic geographic distribution of CMM prevalence in Lingwu City, increasing from north to south. Dongta Township exhibited the highest prevalence (16.6%), followed by Chongxing Township and Wutongshu Township. Spatiotemporal scanning identified four high-incidence clusters. The random forest algorithm outperformed others in predicting CMM prevalence across townships. SHAP analysis highlighted differences in the geographic distribution of 15 influence factors. Age, waist circumference, and hypertension were significant influence factors across Lingwu City. Township-specific influence factors included TG and BMI in Dongta; HDL and TG in Chongxing, Haojiaqiao; HDL and TC in Wutongshu and TC, TG, HDL and BMI in Baitugang.

Conclusion

CMM prevalence shows significant geographic variation within Lingwu City, with distinct risk factors across townships. Tailored interventions, based on local needs, should be implemented to reduce CMM prevalence effectively, optimize health resource allocation, and inform public health policies in rural areas of Northwest China.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-24483-5.

Introduction

Cardiometabolic multimorbidity (CMM) refers to the simultaneous occurrence of two or more cardiovascular and metabolic diseases, such as coronary heart disease, stroke, and diabetes [1–4]. CMM is one of the most common modes of comorbidity globally, with its prevalence increasing rapidly [5]. The prevalence of CMM in China was 5.94%, which more than doubled in 5 years [6]. In 2021, cardiovascular diseases accounted for 48.98% of deaths among rural residents and 47.35% among urban residents in China [7]. This rising burden severely impacts patient prognosis, exacerbates health inequalities between urban and rural areas, increases societal healthcare costs, and places significant pressure on medical services [8–11].

Various studies have identified cardiovascular influence factors, including smoking, alcohol consumption, and physical inactivity, as contributors to the increased risk of CMM [12, 13]. Modifiable factors, such as triglycerides (TG), body mass index (BMI), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), total cholesterol (TC), and waist circumference, are particularly crucial in the development and progression of CMM [14]. Additionally, disparities in economic development [15, 16] and medical resource allocation [17–20] contribute to the regional heterogeneity of CMM prevalence [21, 22]. China Million Person Assessment of Cardiac Events (PEACE-MPP) revealed that 10.3% (95% CI: 10.2–10.3) of participants were at high risk for cardiovascular diseases, with county-specific risks ranging from 3.1 to 24.9% [23]. Notably, cardiovascular disease prevalence tends to be higher in rural areas compared to urban areas across China’s eastern, central, and western regions [24, 25].

Targeted prevention and intervention strategies are essential to reduce the public health burden of CMM, particularly in rural areas. However, research on the spatial distribution and influencing factors of CMM in rural northwest China remains Limited. The Ningxia Hui Autonomous Region, located in northwest China, has a cardiovascular disease risk rate of 18.85%, with significant regional variation [26]. Lingwu City in Ningxia, characterized by its unique rural-urban gradient and economic disparities, offers an ideal setting for studying the epidemiology of CMM [27, 28]. The clearly defined boundary between urban and rural areas provides a realistic geographic unit to explore the spatial and temporal distribution of CMM using geographic information system (GIS) technology [29]. Integrating epidemiological and environmental data allows for the identification of key risk factors and provides a scientific basis for designing effective prevention and intervention measures. The unique geographical advantages of Lingwu City make it an ideal testing ground for rural health initiatives, with findings that offer valuable insights for improving rural health in northern China.

Traditional methods, such as spatial autocorrelation, Bayesian hierarchical models, and geographically weighted regression, have been used to analyze disease patterns and their determinants. However, these approaches often suffer from limitations, such as loss of three-dimensional spatial information and overfitting when spatial weights are large, leading to distorted estimates [30–32]. This study employs machine learning algorithms to overcome these limitations [33, 34]. By comparing model performance metrics, including accuracy, precision, recall, and AUC, the optimal algorithm for evaluating CMM risk across townships in Lingwu City was identified. The best-performing interpretable machine learning model was then used to assess feature importance, highlighting key risk factors driving CMM prevalence in specific regions [35–37].

Using data from Cardiovascular Disease High-Risk Group Early Screening and Comprehensive Intervention Program conducted in Lingwu between 2017 and 2022, we applied spatial autocorrelation, trend surface analysis, and spatiotemporal scanning statistics to characterize the distribution of CMM. The findings generated regional health profiles, revealed spatial disparities in CMM prevalence, and provided insights for optimizing health resource allocation. These results holded significant implications for the development of tailored health policies and interventions that address regional needs, promote health equity, and contribute to the management and prevention of CMM in alignment with the Healthy China initiative.

Materials and methods

Materials

Study design and participants

The Early Screening and Comprehensive Intervention Program (PEACE MPP) for people at high risk of cardiovascular disease is initiated and organized by the National Health Commission. Implemented since 2014, the project has sampled 252 locations (152 rural counties, 100 urban areas) in China, covering 55% of the mainland’s prefecture-level cities. To ensure the diversity of geographical distribution, economic development and population structure, the typical case sampling method was adopted, and about 8 counties or districts were selected from each of the 31 provinces [23, 38].

This study utilized data from Cardiovascular Disease High-Risk Group Early Screening and Comprehensive Intervention Program in Lingwu City, Ningxia, conducted between 2017 and 2022. A phased random sampling method was employed based on the economic conditions, population distribution, and geographical environment of Lingwu City. The study targeted residents aged 35–75 years who had lived in Lingwu City for at least six months within the previous year. Five towns in Lingwu City, Wutongshu Township, Chongxing Township, Baitugang Township, Haojiaqiao Township, and Dongta Township, were designated as sites for the national early screening and comprehensive intervention project. Within each township, three communities (or villages) were randomly selected, and permanent residents aged 35–75 years were invited to participate in the survey. Recruitment efforts were conducted through community health service centers and township health centers, utilizing community mobilization strategies and media outreach. All participants provided written informed consent prior to enrollment. A total of 11,375 individuals were initially recruited for the study. After screening, 11,353 participants met the inclusion criteria and were included in the final analysis (see Additional files 1, Supplementary Fig. 1).

2.1.2 Study area

Lingwu City is located in central Ningxia Hui Autonomous Region, with geographical coordinates between 106°11’ཞ106°52’ E and 37°35’ཞ38°21’ N. It is situated in the mid-eastern part of the Yinchuan Plain, on the eastern bank of the Yellow River, at the edge of the Mu Us Desert, and at the confluence of the Yinchuan Plain and Ordos Plateau, covering a total area of 4010 square kilometers (Fig. 1). The basic situation of five townships in Lingwu City is shown in Table 1.

Fig. 1.

Geographical region of Lingwu City

Table 1.

Five townships in Lingwu City

Townships	Location (in the city)	Area (sq. km)	Population (10,000)	Landform	Climate type	Major industry	Per capita disposable income (Yuan) in 2023
Wutongshu	Northwest	158	2.8	Plane-dominated	Medium-temperate arid continental climate	Main rice production	22,337
Dongta	Southeast	252	2.3	Plane-dominated	Temperate continental monsoon climate	Main Lingwu long jujube	23,269
Chongxing	Southern	128.43	5.0	Hilly and plain areas	Medium-temperate arid continental climate	Food production and vegetable industry	21,700
Haojiaqiao	South central	36	1.9	Hilly and gully	Medium-temperate arid continental climate	Main vegetable, leek	20,549
Baitugang	Southeast	691.92	1.7	High hilly area	Warm temperate continental monsoon	Forage and forage management	21,238

Procedures

We assessed 15 major risk factors associated with CMM, employing the procedures and methodologies of the National Cardiovascular Disease High-Risk Group Early Screening and Comprehensive Intervention Program. Data collection involved administering questionnaires, conducting physical examinations, and performing laboratory tests among local residents. All activities were carried out by trained and qualified personnel. The work flow chart of the survey site is shown in Additional files 1, Supplementary Fig. 2.

There are 6 indicators in the questionnaire, marital status: participants were categorized as either married or others; education level: classified based on whether the highest education level attained was high school or above; smoking: quantified by the number of cigarettes smoked per day and categorized into current smokers or non-smokers; drinking: evaluated using a combination of drinking frequency and the standard daily alcohol intake, dividing participants into current drinkers or non-drinkers; physical activity: the total hours per day spent in occupational, leisure, transportation-related, and household activities were summed to classify individuals as physically active or not; annual household income: annual gross household income, post-tax, was recorded and categorized based on whether the average annual income exceeded 50,000 yuan. The physical examinations included measurements of height, weight, and waist circumference using an electronic height and weight gauge and a soft tape measure. Blood pressure was measured using an HPP-1300 sphygmomanometer (Omron China). Each measurement was performed twice, with intervals of more than 1 min. If the difference between the two systolic blood pressure readings exceeded 10 mmHg, a third measurement was conducted, and the average of the last two results was recorded. For laboratory tests, fasting fingertip blood samples were collected. Blood glucose levels were measured using the PD-G001 Glucose Meter (Baijier), and lipid profiles, including total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL), and low-density lipoprotein cholesterol (LDL), were analyzed using the Cardiocheck PA Rapid Lipid Detector. The basic variables are described in Additional files 2, Supplementary Table 1.

Outcomes

CMM is defined as the coexistence of at least two cardiovascular and metabolic conditions, such as type 2 diabetes, coronary heart disease, and stroke [1–4, 39–42]. Before the project investigation, the qualified personnel (village doctors, etc.) who have participated in the training would issue a notice in advance to the participants who met the inclusion criteria, asking them to carry the disease diagnosis certificate of nearly one year. These factors were combined to determine whether participants had the corresponding diseases. This included reviewing participants’ medical histories (e.g., diabetes, hypertension, and cardiovascular diseases), medication use, and family history. These self-reported details were corroborated with diagnosis certificates and physical examination data collected as part of the National Basic Public Health Service Program. This comprehensive approach allowed for precise confirmation of the presence of CMM-related conditions.

Geographic information data

The base layer of the Ningxia map was sourced from the National Basic Geographic Information System [43].

Methods

Basic descriptive analysis

To ensure that differences in the geographic distribution of CMM prevalence are not influenced by variations in the population age distribution, the indirect standardization method was used to calculate age-standardized prevalence rates for the five townships, referencing the 2022 population of Lingwu City.

Spatial autocorrelation analysis

Global Moran’s I and local Moran’s I were used as indicators of spatial autocorrelation. Moran’ s I ranges from − 1 to 1, with I > 0 indicating positive correlation. The closer the value is to 1, the higher the spatial clustering. I < 0 indicates negative spatial correlation, while I = 0 indicates random distribution [44]. Local Moran’s I includes four clusters: high-high (HH), low-low (LL), high-low (HL), and low-high (LH).

Trend surface analysis

To visualize the spatial-temporal changes in CMM prevalence in Ningxia from 2017 to 2022, the three-dimensional spatial trend surface of average CMM prevalence in different regions was analyzed, revealing the spatial distribution characteristics of clustering areas. Disease index values were considered as points located at geometric centers, with diseases on the Z-axis and geographic positions on the X and Y axes, scattered in three-dimensional space. Scatter projections on the XZ and YZ planes were analyzed to represent trend changes in north-south (latitude) and east-west (longitude) directions, respectively. Fitting the data, estimated values reflected changes and development trends within the entire region.

Center of gravity migration and spatiotemporal scanning statistics

The center of gravity migration method provides a quantitative tool for analyzing spatial distribution characteristics across different time points or regions, offering an intuitive reflection of the overall macro-geographical structure and its evolving characteristics within the study area. An extension of the spatial scanning statistical method, the spatiotemporal scanning method integrates both geographic and temporal dimensions to detect spatial and temporal aggregation or anomalies [45, 46]. During the scanning process, this method calculates the Log Likelihood Ratio (LLR) for each region within each window and identifies the window with the highest LLR as the maximum likelihood cluster. If the LLR value is statistically significant and the p-value is small, this indicates the region may represent a high-risk cluster. Additionally, the Relative Risk (RR) assessment further quantifies the risk level within the region. Spatiotemporal scanning analysis not only reveals patterns of aggregation over time within geographic space but also more precisely identifies the location of spatial clusters and captures aggregation patterns across the entire spatiotemporal range [47, 48].

Prediction model and analysis of influencing factors

To predict CMM, we developed four machine learning models: Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). LR, a linear classification model, employs an S-shaped function to estimate the probability of the target event. SVM is versatile and can handle both linear and non-linear classification and regression tasks. RF, a supervised learning algorithm, generates multiple decision trees using bootstrapping, where the classification of new instances is determined by majority voting across these trees. XGBoost, an optimized implementation of gradient boosting based on decision trees, excels in both regression and classification tasks [49, 50]. The dataset was divided into training and test sets in a 7:3 ratio, and 10-fold cross-validation was applied to ensure robust model evaluation. On the test set, the performance of all models in risk differentiation was assessed using the Receiver Operating Characteristic (ROC) curve. Multiple metrics, including Accuracy, Precision, Recall and F1 scores, were used to evaluate the predictive performance of the models, the calculation is as follows:

To further analyze the contribution of each feature to the model predictions, we utilized the SHapley Additive exPlanation (SHAP) method. SHAP constructs an additive explanatory model that quantifies the contribution of each feature. Compared to alternative interpretability methods (e.g., local interpretable model-agnostic explanations), SHAP demonstrates greater efficiency and reliability in providing robust interpretability of machine learning models [35–37, 51].

The dependent variable in all models was CMM, with the following independent variables: gender, age, marital status, education, annual household income, smoking, drinking, physical activity, hypertension, TC, LDL, HDL, TG, waist circumference, and BMI.

Statistical analysis

Statistical descriptions were performed using IBM SPSS Statistics 26.0 software. Measurement data are presented as means ± standard deviations (). Comparisons between groups were conducted using independent sample t-tests. Count data are presented as frequencies and percentages (%), with group comparisons made using chi-square tests. A three-dimensional trend surface analysis, spatial autocorrelation, and hot spot analysis were conducted using ArcGIS 10.6. Statistical significance was considered at P < 0.05. Machine learning prediction models and influencing factor analyses were constructed using Python 3.9.0.

Results

Basic demographic characteristics

The 11,353 participants included in this study were the primary screening population in the early screening and intervention program for people at high cardiovascular risk, of these, 1,334 (11.8%, 95%CI: 11.2%, 12.4%) had CMM. Additionally, influence factors varied significantly between different townships (P < 0.05) (Table 2).

Table 2.

Basic characteristics of the study population

	Lingwu city N (%)	Baitugang township N (%)	Chongxing township N (%)	Dongta township N (%)	Haojiaqiao township N (%)	Wutongshu township N (%)	P-value
N	11,353(100%)	2125(18.7%)	2095(18.5%)	1805(15.9%)	3191(28.1%)	2137(18.8%)
Gender							< 0.001
Female	4702 (41.4%)	974 (45.8%)	824 (39.3%)	734 (40.7%)	1333 (41.8%)	837 (39.2%)
Male	6651 (58.6%)	1151 (54.2%)	1271 (60.7%)	1071 (59.3%)	1858 (58.2%)	1300 (60.8%)
Age	54.7 ± 9.537	52.3 ± 11.0	55.7 ± 9.3	58.0 ± 8.3	53.5 ± 9.0	55.0 ± 9.0	< 0.001
Marriage							< 0.001
No	464 (4.1%)	66 (3.1%)	95 (4.5%)	48 (2.7%)	131 (4.1%)	124 (5.8%)
Yes	10,889 (95.9%)	2059 (96.9%)	2000 (95.5%)	1757 (97.3%)	3060 (95.9%)	2013 (94.2%)
Education							< 0.001
Below high school	10,871 (95.8%)	2103 (99.0%)	1980 (94.5%)	1680 (93.1%)	3059 (95.9%)	2049 (95.9%)
High school and above	482 (4.2%)	22 (1.0%)	115 (5.5%)	125 (6.9%)	132 (4.1%)	88 (4.1%)
Annual household income							< 0.001
≤ 50,000yuan/year	10,048 (88.5%)	2008 (94.5%)	1883 (89.9%)	1655 (91.7%)	2507 (78.6%)	1995 (93.4%)
> 50,000 yuan/year	1305 (11.5%)	117 (5.5%)	212 (10.1%)	150 (8.3%)	684 (21.4%)	142 (6.6%)
Smoking							< 0.001
No	9665 (85.1%)	1829 (86.1%)	1851 (88.4%)	1570 (87.0%)	2685 (84.1%)	1730 (81.0%)
Yes	1688 (14.9%)	296 (13.9%)	244 (11.6%)	235 (13.0%)	506 (15.9%)	407 (19.0%)
Drinking							< 0.001
No	11,213 (98.8%)	2113 (99.4%)	2092 (99.9%)	1768 (98.0%)	3159 (99.0%)	2081 (97.4%)
Yes	140 (1.2%)	12 (0.6%)	3 (0.1%)	37 (2.0%)	32 (1.0%)	56 (2.6%)
Physical activity							< 0.001
No	2779 (24.5%)	1047 (49.3%)	533 (25.4%)	185 (10.2%)	259 (8.1%)	755 (35.3%)
Yes	8574 (75.5%)	1078 (50.7%)	1562 (74.6%)	1620 (89.8%)	2932 (91.9%)	1382 (64.7%)
Hypertension							< 0.001
No	8613 (75.9%)	1742 (82.0%)	1522 (72.6%)	1188 (65.8%)	2540 (79.6%)	1621 (75.9%)
Yes	2740 (24.1%)	383 (18.0%)	573 (27.4%)	617 (34.2%)	651 (20.4%)	516 (24.1%)
TC	4.3 ± 1.0	4.2 ± 1.0	4.2 ± 0.9	4.3 ± 0.9	4.3 ± 1.0	4.4 ± 0.9	< 0.001
LDL	2.2 ± 0.8	2.2 ± 0.8	2.0 ± 0.7	2.3 ± 0.8	2.2 ± 0.8	2.2 ± 0.7	< 0.001
HDL	1.3 ± 0.4	1.3 ± 0.4	1.4 ± 0.4	1.3 ± 0.3	1.3 ± 0.4	1.3 ± 0.4	< 0.001
TG	2.0 ± 1.1	1.7 ± 1.0	2.1 ± 1.2	1.9 ± 1.0	1.9 ± 1.1	2.4 ± 1.3	< 0.001
Waist circumference	87.1 ± 9.3	83.4 ± 8.4	90.3 ± 10.3	88.7 ± 9.4	86.2 ± 8.6	87.8 ± 8.9	< 0.001
BMI	25.6 ± 3.3	25.2 ± 3.2	26.4 ± 3.4	25.3 ± 3.2	25.5 ± 3.3	25.7 ± 3.2	< 0.001
CMM							< 0.001
No	10,019 (88.2%)	1930 (90.8%)	1841 (87.9%)	1437 (79.6%)	2905 (91.0%)	1906 (89.2%)
Yes	1334 (11.8%)	195 (9.9%)	254 (12.1%)	368 (20.4%)	286 (9.0%)	231 (10.8%)

Spatial and temporal distribution of CMM

Spatial distribution of CMM prevalence

The prevalence of CMM in the five townships of Lingwu City was as follows: Dongta Town (16.6%), Chongxing Town (9.9%), Wutongshu Town (8.5%), Baitugang Town (7.6%), and Haojiaqiao Town (2.8%), with the highest prevalence in Dongta Town (Fig. 2A). To further investigate the spatiotemporal clustering of CMM in Lingwu City’s five townships, the Local Moran’s I method was used. The “low-high” clustering areas were mainly located in the northern region of Lingwu, such as Wutongshu Town, this may reflect that the prevalence of CMM in Baitugang Town is not evenly distributed in space, which may be a transition from high prevalence to low prevalence (Fig. 2B). The hot spot analysis results were consistent (Fig. 2C).

Fig. 2.

Spatial Distribution of CMM Prevalence (A) Prevalence (B) Spatial Clustering Model (C) Cold hot spot analysis

Three-Dimensional trend surface analysis

The three-dimensional trend surface analysis indicated a spatial trend of CMM in Lingwu City, showing a parabolic trend from north to south, with higher rates in the north, and from east to west, with higher rates in the east. This pattern demonstrated a significant spatial orientation, with the northeastern region being a hotspot for CMM (Fig. 3).

Fig. 3.

Three-Dimensional Trend Surface Analysis

Statistical results of center of gravity migration and spatiotemporal scanning

In early 2017, the center of gravity for the overall prevalence of CMM was primarily located in Chongxing Township. By 2018 and 2019, the center of gravity began to shift from its initial position toward the vicinity of Haojiaqiao Township. In 2020, it moved further toward Wutongshu Township, Likely due to a significant increase in prevalence within the township, which rapidly increased its influence on the overall distribution pattern. In 2021 and 2022, the center of gravity shifted prominently toward Dongta Township, suggesting that the town played a dominant role in shaping the regional prevalence of CMM during this period and exerted a key influence on the overall distribution pattern (Fig. 4A).

Fig. 4.

Spatial and temporal distribution characteristic (A) Center of gravity migration (B) Spatiotemporal aggregation scan analysis

The results of the spatiotemporal scanning analysis are illustrated in Fig. 4B, Additional files 2, Supplementary Table 2. Four significant clusters of CMM were identified in Lingwu City. During January to December 2017, Wutongshu Township (RR = 1.19, LLR = 24.50, P < 0.0001) and Chongxing Township (RR = 1.19, LLR = 23.87, P < 0.0001) were identified as the most probable cluster areas. For the period from January to December 2018, Haojiaqiao Township was identified as the most likely cluster area (RR = 1.30, LLR = 54.49, P < 0.0001). Similarly, during the period from January to December 2021, the most significant cluster was in Baitugang Township (RR = 48.07, LLR = 2853.52, P < 0.0001).

Prediction model and analysis of influencing factors of CMM

Model performance

We employed four machine learning algorithms to predict CMM prevalence across various townships. The performance metrics for these models are presented in Fig. 5 and Additional files 2, Supplementary Tables 3–8.

Fig. 5.

ROC Curves for CMM Models in Different Regions (A, F) Baitugang Town Training and Testing Set (B, G) Chongxing Township Training and Testing Set (C, H) Dongta Township Training and Testing Set (D, I) Haojiaqiao Township Training and Testing Set (E, J) Wutongshu Township Training and Testing Set.

In the training set, the RF model demonstrated exceptional performance in Dongta Township (AUC = 0.999, 95% CI: 0.998–0.999) and Wutongshu Township (AUC = 1.000, 95% CI: 0.999–1.000). For Baitugang Township (AUC = 0.959, 95% CI: 0.945–0.959), Chongxing Township (AUC = 0.974, 95% CI: 0.965–0.974), and Haojiaqiao Township (AUC = 0.981, 95% CI: 0.975–0.981), the XGBoost model achieved optimal performance. In the test set, the RF model outperformed other algorithms across all five townships. Based on these results, we identified the RF algorithm as the optimal model for predicting CMM prevalence in the townships of Lingwu City.

Predictors of CMM in different regions of Lingwu City

Given the excellent performance of the RF algorithm, we provide explanations based on the findings of this machine learning model. To elucidate the importance of each feature within the model, the ranked importance of input features is provided in Additional files 1, Supplementary Fig. 3. Figure 6 presented the SHAP summary plot, where predictors are ranked by importance. The x-axis represents SHAP values, indicating the influence of each predictor, while the y-axis shows the predictors. Higher SHAP values suggest a stronger likelihood of contributing to CMM. Red dots represent higher predictor values, whereas blue dots signify lower values.

Fig. 6.

SHAP Interpretative Factors for CMM in Different Regions (A) Lingwu City (B) Baitugang Township (C) Chongxing Township (D) Dongta Township (E) Haojiaqiao Township (F) Wutongshu Township

Analysis of the top five most significant factors in the three towns with the highest CMM prevalence rates (Chongxing Township, Dongta Township, and Wutongshu Township) revealed age and waist circumference as consistent and critical predictors, with high SHAP values underscoring their importance as key risk factors. Conversely, education level, drinking status, and marital status contributed minimally to CMM prediction.

Beyond age and waist circumference, the key predictors for CMM varied across townships, In Baitugang Township, significant factors included TG, TC, BMI, and HDL. In Chongxing and Haojiaqiao Township, TG, and HDL were influential. Dongta Township highlighted TG, and BMI as critical predictors. Similarly, in Wutongshu Township, HDL and TC were the most significant predictors. These findings highlight the importance of developing township-specific strategies to address unique risk factor profiles effectively (Fig. 7). The contributions of individual features to CMM prediction across Lingwu City and its townships are illustrated in Additional files 1, Supplementary Fig. 4.

Fig. 7.

Relationship between influencing factors of CMM in different regions

Discussion

Currently, limited research has explored the regional heterogeneity and influencing factors of rural CMM in northwest China. This study addresses this gap by investigating the temporal and spatial distribution patterns of CMM and identifying key risk factors for its prevalence across different townships in Lingwu City, Ningxia, using population samples from Cardiovascular Disease High-Risk Group Early Screening and Comprehensive Intervention Program.

Lingwu City’s geographical landscape, characterized by hilly and plain terrains with an arid (monsoon) continental climate in the mid-temperate zone, serves as a representative rural area in northwest China. The city’s agricultural system, which focuses on grain processing, grass and livestock farming, and cultivation of melons and vegetables, forms the foundation of its socioeconomic profile. Additionally, the population distribution reflects naturally formed human settlements, representing homogenous social and geographical communities shaped by longstanding production and living practices.

The findings revealed that the average prevalence of CMM in Lingwu City was 11.8%, with significant variability across townships. For instance, Dongta Township exhibited a notably higher prevalence of 16.6%, exceeding the national average [52]. This aligns with Xu et al.'s report of a 11.6% prevalence in Guangdong Province based on the China Patient-Centered Cardiac Event Evaluation Million People Project (China-Peace MPP) [53]. Similarly, the China-PAR project highlighted considerable regional differences in cardiovascular risk across China, with higher proportions of high-risk groups observed in northeastern and northern regions compared to southern areas [23]. These findings suggest that environmental and socioeconomic disparities play a pivotal role in influencing cardiovascular disease.

The analysis of major risk factors at the community level in Lingwu City identified clusters of key factors, particularly age, waist circumference, and hypertension, which were prevalent among populations with high CMM prevalence. Consistent with the China Kadoorie Biobank (CKB) study, which reported a 15.8% comorbidity rate that increased with age, our findings underscore aging as a critical risk factor for CMM [54]. Older individuals face heightened vulnerability to multiple chronic conditions. Additionally, increased waist circumference, indicative of visceral fat accumulation, is strongly associated with cardiovascular and metabolic risks [55–57], independent of BMI [58]. Hypertension, a central component of CMM, significantly contributes to its development and progression [59, 60].

This study also explored the relationship between CMM risk factors and regional environmental and socioeconomic characteristics. Contrary to the protective effects of high socioeconomic status (SES) combined with a healthy lifestyle reported in previous studies [61], our findings revealed an association between higher socio-economic situation and increased CMM prevalence [17, 62]. This paradox may reflect greater exposure to unhealthy lifestyle factors, such as frequent dining out, consumption of high-sugar and high-fat foods, and insufficient physical activity among individuals with higher socio-economic situation [63, 64]. Dongta township, a prominent area in Lingwu City, shares urban development and lifestyle similarities with neighboring Chongxing and Wutong townships, which may further influence these patterns. Additionally, disparities in local economic development, education levels, medical access, and rural residents’ health awareness likely contribute to the observed trends [65, 66].

Further analysis revealed that TG, BMI, HDL, and TC were also closely associated with CMM. Low HDL, an independent risk factor for CMM progression [67], and elevated TC, TG levels and BMI significantly increased CMM risk [68, 69]. Although education level, alcohol consumption, and marital status contributed minimally to the model’s explanatory power, this likely reflects the homogeneity of these factors across Lingwu’s population. Most participants were married, rarely consumed alcohol, and attained high school education, with few attained higher education, limiting their independent influence on CMM prevalence.

Significant geographical disparities in CMM prevalence exist across Lingwu City, with varying primary influencing factors among townships. These findings highlight the need for geographically tailored intervention strategies that consider environmental and socioeconomic contexts to effectively mitigate the burden of CMM. Such strategies should aim to reduce disparities in primary healthcare access and quality while addressing key risk factors to improve cardiovascular health outcomes in rural northwest China.

The current study has several limitations. First, for diagnosing CMM and assessing cardiovascular disease risk factors, a combination of investigator self-report and objective tests was used. Self-reporting may be subject to memory bias and reporting bias, while objective tests may be constrained by accessibility and accuracy limitations. Second, although the five townships selected for the cardiovascular high-risk screening project in Lingwu City represent the region’s overall population, a more comprehensive and representative sample could be achieved by including a broader range of townships. Additionally, while machine learning algorithms were used to predict the prevalence of CMM in different townships of Lingwu City, future research could benefit from using data from other towns in northwestern rural areas for external validation. Finally, environmental indicators were not collected in this study. Future research will address this gap by gathering environmental data from areas near industrial zones to assess the impact of various environmental factors on the prevalence of CMM.

Conclusion

By analyzing the prevalence of CMM and its influencing factors across different townships in Lingwu City, this study created regional health profiles, revealing the spatial distribution of CMM and highlighting the differences in factors affecting the health of residents in various areas. It provides a scientific foundation for optimizing the spatial allocation of health resources, aiming to reduce disparities in resource distribution, service capacity, and health outcomes between urban and rural areas, as well as among different populations. Specifically, this research offers valuable insights for improving the management and prevention of CMM in rural areas, promoting the adoption of advanced healthcare concepts and enhancing service capabilities, ultimately fostering health equity and improving the overall health of society.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Author contributions

All authors contributed to the conceptualisation and planning of the analyses. Wei Gong and Yuxin Zhao: Writing– original draft, Methodology; Jianping Shi and Siyu Ma: Writing– review & editing, Methodology, Data curation; Xiaoxiao Hu, Manya Ma, and Xiuna Li: Methodology, Data curation; Jinlong Shi and Jianjun Yang review & editing, Conceptualization.

Funding

This work was supported by the National Natural Science Foundation of China (No.82060597), Ningxia Natural Science Foundation (No.2024AAC03240), the Key research and development project in Ningxia (No.2021BEG03031), the Key research and development project in Ningxia (2022BEG01001).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare no competing interests.

Clinical trial number

Not applicable.

Ethics approval

The central ethics committee at the National Center for Cardiovascular Disease (NCCD) approved the pilot (Trial Registration Number NCT02536456).

Institutional review board statement

Not applicable.

Informed consent statement

Not applicable.

Conflict of interest

The authors declare no competing interests.