This study, a comprehensive, population-based survey, was conducted in Guangdong Province, Southern China, from April 9 to May 18, 2021. Known as the “Guangdong Sleep Health Study,” it was structured within the framework of the China National Health Survey (24). This study utilized a multi-stage, stratified, cluster sampling method to recruit a representative cohort based on demographic characteristics. Specifically, a three-level composite sampling framework (multi-stage, stratified, and cluster sampling) was implemented to ensure sample representativeness. Initially, we selected Shantou and Meizhou cities in Guangdong Province as the primary study areas. The research region was further divided into coastal (including Chenghai District and Jinping District in Shantou), island (including Nan’ao County in Shantou), and mountainous (including Meijiang District and Jiaoling County in Meizhou) areas, based on three distinct cultural backgrounds: Chaoshan, island, and Hakka.

Recruitment of participants began approximately one week before the formal survey. Participants were invited to schedule their participation via a WeChat mini-program module, which facilitated convenient and efficient appointment booking. With the assistance of local government authorities in the study areas, the research team invited residents from selected towns and communities to participate in the survey.

Initially, 5,838 participants completed the questionnaire survey. Of these, 3,829 completed sleep monitoring, while 2,009 did not. After excluding 133 participants with missing values for MetS, the number of remaining participants was reduced from 3,829 to 3,696. Further exclusions included 68 participants who had received OSA-related treatment, 102 who were using lipid-lowering drugs, and 87 who were using sedative hypnotic drugs. This resulted in 2,649 participants in the non- MetS group and 791 participants in the MetS group. The recruitment and enrollment process is detailed in Supplementary Figure S1 .

The study received ethical approval from the Ethics Committee of Guangdong Provincial People’s Hospital (GDREC2020221H), ensuring adherence to ethical standards. Informed consent was obtained from all participants, guaranteeing their voluntary agreement to participate in the research.

In this study, a reliable type IV wearable sleep monitoring device ( Supplementary Figures S2, S3 ), developed by Chengdu Cloud Care Healthcare Co. Ltd. in Chengdu, China, was used to monitor the sleep patterns of participants (25). This device has been previously described in research published by our group. Briefly, blood oxygen saturation is measured using the photoplethysmographic method, which relies on the absorption characteristics of hemoglobin for infrared and red light. The device records and analyzes effective monitoring time and the number of desaturation events (3%) using built-in automatic algorithms to generate a report. Monitoring is conducted on the thenar eminence of the palm, avoiding areas with veins, scars, spots, or dense hair.

This study compared and validated the Type IV sleep monitoring device against polysomnography at the Sleep Center of Guangdong Provincial People’s Hospital. A total of 196 participants were included, all of whom underwent an all-night PSG and simultaneous monitoring with the Type IV sleep monitoring device to assess its sensitivity and specificity in diagnosing OSA. When using an AHI of ≥5 as the diagnostic criterion for OSA, the sensitivity of the Type IV device was 93%, specificity was 77%, and the area under the curve (AUC) was 0.95. When the AHI threshold was raised to ≥15, the sensitivity was 92%, specificity was 89%, and the AUC remained 0.95 (25).

Among all individuals screened with the Type IV device, we selected 305 participants from Chenghai District and Jinping District in Shantou City and conducted home sleep apnea testing on them. The comparison of the consistency between the Type IV device and HSAT further validated the accuracy of the Type IV device in screening for OSA in a healthy community population. Among the 305 community participants validated with HSAT, the oxygen desaturation index (ODI) measured by the Type IV device showed a strong correlation with the AHI measured by HSAT (R²=0.504, P<0.001). Bland-Altman consistency testing indicated that approximately 93% (284/305) of the data points for the ODI from the Type IV device and the AHI from HSAT fell within the 95% limits of agreement (26).

Due to its ease of use, accessibility, and cost-effectiveness, the Type IV sleep monitoring device has been validated in various populations, including the general population, obese individuals, surgical patients, patients with acquired immune deficiency syndrome, and those with atrial fibrillation (27–30). The use of the Type IV sleep monitoring device alone or in combination with the Stop-Bang questionnaire showed significantly better performance in OSA screening compared to the Stop-Bang questionnaire alone, especially in detecting mild OSA (31). Despite some inconsistencies in the actual AHI estimates, the U.S. Preventive Services Task Force has stated that the Type IV sleep monitoring device generally demonstrates high accuracy in diagnosing OSA (32).

Various parameters such as 3% ODI, nocturnal mean oxygen saturation (MeanSpO2), lowest nocturnal oxygen saturation (MinSpO2), and proportion of time spent with oxygen saturation below 90% (T90) were recorded during the monitoring process. The presence and severity of SDB were determined based on ODI4% levels: normal (<5 events/h), mild (5 to <15 events/h), moderate (15 to <30 events/h), and severe (≥30 events/h) (28).

According to the Chinese Guidelines for the Prevention and Treatment of Type 2 Diabetes (2020 Edition), metabolic syndrome is defined by the presence of at least three of the following components (33):

In this study, physical examinations included measurements of height, weight, neck circumference, waist circumference, blood pressure, and body composition analysis. All measurements were conducted by trained technicians and repeated twice to ensure accuracy.

Height was measured with participants standing next to a vertical wall on a level floor. During measurement, participants removed any hats and shoes, standing upright with the head, shoulders, buttocks, and heels touching the measuring instrument. The measuring plate was positioned parallel to the wall and precisely at the top of the participant’s head. Height was recorded in centimeters (cm), accurate to 0.1 cm.

Neck circumference and waist circumference were measured at the upper edge of the thyroid cartilage and 1 cm above the navel, respectively. Participants breathed naturally, avoiding deep inhalation, neck muscle tension, or abdominal contraction. A soft tape measure was placed close to the skin without compression, and measurements were recorded to the nearest 0.1 cm.

Weight was measured using the TANITA BC-420 body composition analyzer (manufactured in Japan). The instrument was placed on a flat surface and connected to a power source. Participants removed shoes and socks and stood with both feet evenly placed on the measurement plate. The researcher selected the appropriate measurement mode and input participants’ gender and height. The instrument completed the measurement within seconds, displaying parameters including weight, body fat percentage, muscle mass, and water content. Body Mass Index (BMI) was calculated by dividing weight by the square of height.

Blood pressure was measured three times using the Omron HEM-907 digital blood pressure monitor (manufactured in Japan), with each measurement taken 1 minute apart and the average value recorded. Before measurement, participants were advised to avoid stimulants such as caffeine and tobacco, intense exercise, and to rest quietly for 5 minutes. During measurement, participants sat comfortably with both feet flat on the floor and their left arm resting on a table or support. The cuff was positioned on the left arm at heart level, and the device automatically inflated to record real-time blood pressure data. After measurement, the device deflated and displayed the results, including systolic pressure, diastolic pressure, and heart rate.

For each participant, sterile techniques were employed to extract 9 milliliters of fasting blood samples following an 8-hour fast, typically from the left anterior arm vein. Blood samples were collected in a specific sequence: first using the yellow blood collection tube, followed by the purple blood collection tube, which contained anticoagulants to prevent coagulation. After separation, the samples were stored at -20°C to ensure stability. During transportation, ice bags and sealed containers marked with biological hazard indicators were used to maintain safety and integrity from the collection point to the laboratory.

The analysis of the blood samples was conducted based on required indicators and methods. Conventional blood analysis was performed using the XT-4000i automated instrument, while specific biochemical analyses required special reagents and equipment. The routine blood analysis included measurements of red blood cells, white blood cells, platelets, and hemoglobin. Blood lipids were analyzed, including total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), and lipoprotein(a). Other biochemical indicators measured were fasting blood glucose, high-sensitivity C-reactive protein (hsCRP), alanine aminotransferase (ALT), aspartate aminotransferase (AST), urea, creatinine, and uric acid.

This study utilized a standard questionnaire consisting of four parts. Each part of the questionnaire required signatures and special annotations from respondents, ensuring that final verifiers could quickly confirm the completion of the survey. The first part collected sociodemographic information, including age, gender, education level, and marital status. The second part gathered self-reported professional diagnoses of diseases such as hypertension, coronary heart disease, diabetes, and dyslipidemia, along with the disease history of individuals and their family members. The third part documented participants’ lifestyle habits, including smoking history, drinking history, and physical exercise. The fourth part recorded on-site physical examination results.

Smoking: Smoking continuously for more than six months. Quitting smoking refers to not smoking for at least six months after previously smoking. Drinking: Regularly drinking at least twice a month, with consumption of at least one bottle of beer or two liang of Baijiu/wine. Abstinence refers to not drinking for at least half a year. Physical Exercise: Engagement in physical exercise in the past year, averaging more than 20 minutes per session. Frequency categories are 5–7 days per week, 3–4 days per week, 1–2 days per week, ≤ 3 days per month, or never exercising. Hypertension: Use of antihypertensive drugs and/or systolic or diastolic blood pressure ≥ 140/90 mmHg. Diabetes: Fasting blood glucose ≥ 7 mmol/L or use of diabetes medications (oral hypoglycemic drugs and/or insulin).

The questionnaire was administered by a professional physician through face-to-face interviews. On-site supervisors conducted quality control inspections to ensure the completeness and authenticity of each questionnaire. At the end of each day, the questionnaires were scanned and converted into electronic format using specialized software, with data input checked for accuracy.

Participants with snoring and other sleep problems were evaluated using specific sleep questionnaires:

Epworth Sleepiness Scale (ESS): Developed by Murray W. Johns in 1991, this tool assesses daytime drowsiness tendencies with 8 descriptive questions, each scored from 0 to 3. Total scores range from 0 to 24, with a score of ≥ 9 indicating daytime sleepiness (34).

Pittsburgh Sleep Quality Index (PSQI): Compiled by sleep medicine experts from the University of Pittsburgh, the PSQI includes 19 items covering various aspects of sleep quality. Scores range from 0 to 21, with a score of ≤ 5 indicating good sleep quality and > 5 indicating poor sleep quality (35).

Insomnia Severity Index (ISI): Developed by Morin and Barlow in 1993, this scale assesses insomnia severity over the past week with 7 items, each scored from 0 to 4. Scores ≤ 7 indicate no insomnia, while scores > 7 indicate insomnia (36).

To evaluate the association between sleep parameters and the prevalence of MetS, an overall sleep quality score was calculated by combining all four sleep parameters, ranging from 0 (best) to 8 (worst), as detailed in Supplementary Table S1 . In this study, the overall sleep score was considered both as a continuous variable and a categorical variable: scores > 5 were considered the worst sleep quality, 3–5 indicated poor sleep quality, and < 3 indicated the best sleep quality (37).

Quantitative data are presented as median ± interquartile range, and categorical data are presented as numbers (percentages). The comparison of quantitative data was performed using the independent samples t-test, while categorical data were compared using the chi-square test.

To evaluate the association between nocturnal hypoxia parameters, sleep symptoms, sleep patterns, and metabolic syndrome, univariate and multivariate logistic regression models were used. Model 1 is the unadjusted model, excluding covariates. Model 2 is the minimally adjusted model, adjusting only for sociodemographic variables. Model 3 is the fully adjusted model, adjusting for all covariates. To ensure the accuracy of the data analysis process, a sensitivity study was conducted. Oxygen saturation parameters were divided into quartiles, and the P-value for trend analysis was calculated to confirm the findings regarding oxygen saturation parameters as continuous variables and to explore any potential nonlinear relationships.

Due to the limitations of logistic regression models in handling nonlinear relationships, we used restricted cubic splines (RCS) to address potential nonlinear relationships between oxygen saturation parameters and metabolic syndrome. Subgroup analysis was performed using stratified binary logistic regression models, and the likelihood ratio test was used to investigate the impact of modifying subgroup indicators. Mediation analysis was conducted to determine the extent to which inflammation indices mediated the effect of the ODI on metabolic syndrome. The mediation analysis employed three equations to examine the relationships between the ODI, the mediator (inflammation indices), and the dependent variable (MetS). The analyses were adjusted for age, sex, marital status, education level, economic status, dietary factors (meat, seafood, egg, dairy, bean, vegetable, fruit, pickles), smoking, drinking, tea consumption, activities, and exercise. The proportion of mediation was used to evaluate the degree of the mediation effect.

Latent class analysis was performed using the ‘poLCA’ package in R software to fit 2 to 10 latent cluster solutions. The optimal number of clusters was determined by the smallest Bayesian Information Criterion (BIC). Once the optimal number of classes was determined, all variables were tested to identify significant differences among the clusters.

Statistical analyses were performed using the R software package (version 4.1.2) and Empower Stats (X&Y Solutions, Inc., Boston, MA).

In the total population of 3,440 individuals, 791 (22.99%) have MetS ( Table 1 ). Among the total population, 1,025 (29.80%) are male, and 2,415 (70.20%) are female. Individuals with MetS are generally older, have higher BMI and waist circumference, and exhibit significant differences in gender distribution, marital status, education level, and income compared to those without MetS. Additionally, they show distinct differences in lifestyle habits such as smoking and drinking, as well as in biochemical measurements including TG, HDL-C, LDL-C, and FPG levels.

MetS scores show a significant increasing trend with the severity of SDB. The prevalence of MetS also increases significantly with the severity of SDB ( Figure 1 ).

The results show that the adjusted odds ratio (OR) for the risk of MetS in males with severe SDB is 4.50 [2.1, 9.6], and in females, it is 3.60 [1.4, 9.5] ( Table 2 ). Further interaction analysis revealed that the difference in adjusted OR between gender subgroups was not statistically significant (p-value for gender difference = 0.302).

In males, the severity of SDB is independently associated with an increased risk of MetS in those under 60 years old (p-value for age group difference <0.001). In females, the severity of SDB is significantly associated with MetS in those aged 60 years and above (p-value for age group difference = 0.028) ( Table 3 ). When classified by menopausal status, the severity of moderate to severe SDB in postmenopausal women is associated with metabolic syndrome (2.2 [1.3, 3.5], p = 0.002) ( Supplementary Table S2 ).

After adjusting for confounding factors such as age and gender, hypoxia burden indicators are associated with risk of MetS ( Table 4 ). Using the low-risk group (Q1) as the reference, the risk of MetS for quartiles Q2 to Q4 of indicators reflecting hypoxia burden, such as the ODI, T90 was significantly increased (P<0.01) ( Supplementary Table S3 ).

In various subgroups, hypoxia burden indicators are associated with an increased prevalence of MetS in most subgroups ( Supplementary Tables S4, 5, 6, 7 ). The nonlinear relationship between hypoxia burden indicators and prevalence of MetS was evaluated using RCS ( Figure 2 ). The plot showed an inverted U-shaped association between continuous average arterial oxygen saturation and MetS. When MeanSpO2 was below 94%, the risk of MetS increased significantly.

The mediating effects of inflammation and uric acid on the association between ODI and MetS are shown in Figure 3 . CRP, inflammation scores (INFLA), and uric acid played significant mediating roles in the association between ODI and MetS, with mediation ratios of 4.2%, 10.3%, and 12.7%, respectively ( Supplementary Table S8 ).

After controlling for covariates, regression analyses were conducted using the PSQI scale’s indicator of poor sleep quality, the overall scale score, seven component scores, sleep onset time, nighttime sleep duration, and nap duration as independent variables. Most of the associations between the variables and the research outcomes were not statistically significant, except for the use of hypnotic drugs ( Supplementary Table S9 ). Similar results were observed for daytime sleepiness and insomnia when considered separately ( Supplementary Tables S10, S11 ). Among symptoms related to SDB, loud snoring was associated with an OR for MetS of 2.02 (95% CI 1.57, 2.59), and observed apnea was associated with an OR for MetS of 1.62 (95% CI 1.09, 2.40) ( Supplementary Table S12 ). Worse overall sleep quality was associated with a higher likelihood of having metabolic syndrome (OR 1.14, 95% CI 1.05, 1.23, p=0.0013 Table 5 ).

Based on clinical interpretability, four subtypes were determined to be the most suitable clustering scheme. Supplementary Table S13 presents the clinicodemographic participant characteristics in each cluster. We categorized the patients into four distinct clusters: Cluster 1 includes the pure insomnia group with fewer daytime symptoms; Cluster 2 consists of the minimally symptomatic group; Cluster 3 comprises the insomnia group with multiple daytime symptoms; and Cluster 4 encompasses the group with upper airway symptoms and sleepiness. Figure 4 displays a heatmap of the frequency of specific symptoms and signs in each cluster, highlighting the major differences among the SDB subtypes. The proportion of comorbid hypertension was the highest in Cluster 1 (61.1%). There was no significant difference in the proportion of patients with comorbid MetS among the subtypes ( Table 6 ).