Introduction

Obstructive sleep apnea (OSA) is a common and under-recognized sleep disorder, characterized by periodic reductions or cessations in ventilation caused by dependent complete or partial collapse of the upper airway, resulting in consequent hypoxia, hypercapnia, or arousals from sleep [1]. OSA has affected 9% to 38% of the general adult population in Europe and North America [2], 14.0% to 39.4% in Asia [3], and 8.8% to 24.2% in China [4]. It is estimated that only about 1 in 50 patients with symptoms suggestive of OSA syndrome is evaluated and treated [5], as quite a few OSA patients are under-diagnosed or asymptomatic [6]. When left untreated, individuals with OSA are at heightened risk of metabolic syndrome, cardiovascular diseases, reduced quality of life, premature death, etc. [1, 6].

Obesity is one of the most important risk factors for OSA [1], and weight change can influence OSA severity [7]. Body mass index (BMI) is a traditional indicator of general obesity and is widely used in predicting OSA [8]. However, BMI has been criticized for failing to distinguish the fat distribution [9], because OSA is mainly associated with the central distribution of body fat [10]. Waist-to-hip ratio (WHR), an indicator of abdominal obesity, has been demonstrated more strongly linked with OSA than BMI [11]. Most adult OSA patients have abdominal obesity and increased visceral fat, releasing more inflammatory cytokines than peripheral obesity with predominant subcutaneous fat accumulation [7, 11]. This could lead to neck adiposity, increased upper airway fat, and metabolic abnormalities, even in normal-weight subjects [7]. A cross-sectional study among 1,912 Turkish adults showed that neck circumference (NC) was significantly associated with OSA risk, and its ability to predict OSA was greater than that of waist circumference (WC) [12].

Visceral adiposity index (VAI) and lipid accumulation index (LAP) are newly proposed indicators combining anthropometric indicators with lipid levels. The former is a sensitive indicator to reflect visceral obesity, and the latter is derived from the combination of triglyceride level and waist circumference [13]. Zou and colleagues found that LAP and VAI were moderately correlated with OSA severity, and suggested that anthropometry combined with visceral fat markers could be a more effective diagnostic tool for OSA [13]. Besides, body fat percentage (BF%) is commonly used in obesity research, but there are few studies on its relationship with OSA. A study in Uppsala found that men with severe OSA had a higher BF% than those without OSA, even if the cases and controls were matched for age and BMI [14]. Also, considering that obesity is the result of energy imbalance and the resting metabolic rate (RMR) is correlated with daily energy expenditure, it would be more useful to combine RMR with adiposity indicators to explore the relationship between obesity and OSA.

However, the single utilization of the aforementioned indicators could not adequately reflect the effect of adiposity on OSA risk and current studies have not yielded consistent conclusions. Less is known about the association of novel indicators (such as VAI and LAP) with the risk of OSA. Therefore, this large-scale study was conducted by considering NC, WHR, VAI, LAP, BF%, and RMR to examine the association of adiposity with OSA risk based on Chinese adults.

Methods

Setting and subjects

This cross-sectional study was based on the Guangzhou Heart Study, an ongoing population-based prospective cohort. The baseline survey was accomplished from 2015 to 2017 in Guangzhou permanent residents by multistage sampling method. The details have been described elsewhere [15,16,17,18]. In brief, a total of 12,013 participants aged ≥ 35 years were recruited in the baseline survey, and 2,280 subjects were excluded due to the following exclusion criteria: age older than 74 years (n = 1,043), lack of OSA-related data (n = 5), suffering from the chronic obstructive pulmonary disease (COPD, n = 678) or cardiovascular disease (CVD, n = 554). Recent studies have demonstrated that COPD characterized by a chronic bronchitis phenotype could promote OSA, while lung hyperinflation could protect against OSA [19]. OSA patients tend to be comorbid with CVD [20], which may affect the reliability of our results. Therefore, participants with COPD or CVD were excluded to avoid potential bias. Ultimately, 9,733 participants were selected for further analyses. This study was approved by the Ethical Review Committee for Biomedical Research, School of Public Health, Sun Yat-sen University. The study was performed following the Declaration of Helsinki and written informed consent was obtained from each participant.

OSA ascertainment

OSA was determined by the Berlin Questionnaire (BQ), which was widely used to screen for OSA [21]. The Chinese versions of BQ have been proven to have superior predictive validity and reliability [22, 23]. BQ is a commonly used questionnaire in epidemiological and clinical settings and consists of ten questions in three categories: snoring and breathing cessation (Category 1), excessive daytime sleepiness (Category 2), and BMI and hypertension (Category 3). Category 1 and Category 2 are considered positive with a persistent report of corresponding symptoms (frequency more than three times per week), and Category 3 is considered positive with the report of a history of hypertension or with a BMI of more than 30 kg/m2. Positive scores in two or more categories suggest that the respondent is at high risk for OSA, otherwise at low risk [24]. Then the participants judged to be at high risk of OSA by BQ were assigned to the OSA group and those at low risk of OSA were assigned to the non-OSA group.

Adiposity indicators and anthropometric measurements

Six adiposity indicators were assessed, including NC, BF%, WHR, RMR, VAI, and LAP. Each participant was asked to wear light clothes and step barefoot on the uniformed device to undergo a physical measurement by trained staff. Height and weight were measured to the nearest 0.1 cm (cm) and 0.1 kg (kg), respectively. NC, WC, and hip circumference (HC) were measured to the nearest 0.1 cm through a portable measuring tape. Subjects were asked to stand upright and look straight ahead with shoulders down, and NC was measured by putting the measuring tape midway around the neck, at the level of the laryngeal prominence. WC was gauged at the midpoint between the iliac crest and the lower end of the rib cage, and HC was measured at the maximum extension of the buttocks. Height, weight, NC, WC, and HC were all measured three consecutive times and the mean of each parameter was calculated. BMI was calculated as the mean of body weight in kilograms divided by the mean of height in meters squared (kg/m2) and WHR was calculated by dividing the mean measurement of WC by that of HC.

BF%, VAI, and RMR were measured by the bioelectrical impedance device (OMRON-HBF-371-SH: OMRON Corporation, Yangzhou, China) [25]. BF% was calculated by dividing total fat mass by total mass (including fat mass and fat-free mass) and then multiplying by 100. LAP is based on a combination of waist circumference and the fasting concentration of circulating triglycerides and is defined to describe the extent to which an individual has traveled the route of both increasing waist and increasing triglycerides [26]. LAP is calculated depending on gender: LAP for men = (WC [cm]—65) × (triglycerides concentration [mmol/L]), LAP for women = (WC [cm]—58) × (triglycerides concentration [mmol/L]). To avoid having nonpositive values for LAP, any waist values for men that were 65 cm or less were revised upward to 66.0 cm and for women that were 58 cm or less were revised upward to 59.0 cm [26].

Potential confounding factors

Structured questionnaires were applied to acquire information on demographic characteristics, lifestyle factors, and history of diseases at the face-to-face interview. The modified Global Physical Activity Questionnaire was used to assess leisure-time physical activity (LTPA, MET-h/week) for each participant as we reported previously [15]. Blood pressure was measured and serum cholesterol, low-density lipoprotein cholesterol, and triglyceride were detected. The participant who self-reported physician-diagnosed dyslipidemia or with serum cholesterol of ≥ 5.2 mmol/L, or low-density lipoprotein cholesterol of ≥ 3.4 mmol/L or triglyceride of ≥ 1.7 mmol/L was defined as having dyslipidemia [27]. The subject who self-reported physician-diagnosed hypertension or whose systolic blood pressure was ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg was considered as having hypertension. The confounders included age (years), sex (male, female), marital status (married, others), educational status (primary school and lower, junior high school, senior high school, and college or above), work intensity (light, moderate, vigorous, and retirement), smoking (never, occasion or frequent smoking), alcohol drinking (never, occasion or frequent drinking), vegetable intake (< once/day, ≥ once/day), fruit intake (< once/day, ≥ once/day), hypertension (yes, no), and dyslipidemia (yes, no).

Statistical analysis

All statistical analyses were performed using R software (version 3.6.3). Data were expressed as mean (standard deviation, SD), median (interquartile range, IQR), or frequency (proportion, %), in accordance with the normal, skewed, or categorical distribution. Differences in the baseline characteristics among the non-OSA and OSA groups were computed by t-test, Wilcoxon rank-sum test, or chi-square test. The Pearson correlation test was used for the normally distributed data, and the Spearman correlation test was used for the non-normally distributed data. Each adiposity indicator was converted to a categorical variable based on the quartiles.

The odds ratio (OR) with a 95% confidence interval (CI) was calculated using logistic regression models to demonstrate the association between each indicator and OSA risk. Three models were considered: model 1 was without any adjustment; model 2 was adjusted for age, sex, marital status, education, smoking, alcohol drinking, fruit intake, vegetable intake, work intensity, and LTPA; model 3 was further mutually adjusted for adiposity indicators, aiming to examine the independent association of each indicator with OSA. The multicollinearity was also considered among all variables in the models and variance inflation factors (VIFs) were calculated. The results showed that BMI was not suitable for the adjusted models (VIF > 10) because BMI was closely correlated with adiposity indicators.

Stratified analysis was conducted by age (< 65 years, ≥ 65 years), sex (male or female), and dyslipidemia (yes, no). The multiplicative interaction of adiposity indicators with age, sex, and dyslipidemia was calculated, with the likelihood ratio test by comparing the likelihood scores of the two models with or without the interaction items. A sensitivity analysis was conducted by adjusting the upper and lower 2.5% of the adiposity indicators to the means of which, aiming to exclude the influence of possible outliers. Besides, considering that women's menopausal status plays an important role in OSA occurrence [28], we divided all women into premenopausal group and postmenopausal group based on their self-reported information. Then, we repeated analyses to estimate whether there were differences in the associations between adiposity indicators and OSA risk in women with different menopausal status. All P values were two-tailed and a P value < 0.05 was considered statistically significant.

Results

A total of 9,733 participants were enrolled in this study and 1626 participants (16.71%) were classified into the OSA group. Relative to the participants in the non-OSA group, subjects in the OSA group were more likely to be older, male, and married, to smoke and drink alcohol, to be retirees or take up a vigorous occupation, to have a higher level of education, to eat vegetables or fruit at least once per day, to actively take up LTPA, to have hypertension or dyslipidemia, to have a higher value of BMI, NC, BF%, WHR, VAI, LAP, and RMR (Table 1).

Table 1 Basic characteristics of the study participants
Full size table

Regarding subjects in the lowest quartile of each indicator, ORs (95%CIs) for those in the highest quartile were 2.29 (1.78, 2.97), 2.65 (2.01, 3.48), 2.15 (1.73, 2.71), 4.58 (3.49, 6.02), 2.24 (1.81, 2.77) and 7.43 (5.75, 9.64) for NC, BF%, WHR, VAI, LAP, and RMR respectively after adjusting for all covariates (Table 2). The exposure–response trend of OSA with six indicators was observed (all P -trend < 0.05). Every 1-unit increment of NC, BF%, and VAI was associated with a 13%, 9%, and 14% increased risk of OSA, respectively; every 0.01-unit increment of WHR was associated with a 3% increased risk of OSA; every 10-unit increment of LAP and RMR was associated with a 2% and 4% increased risk of OSA, respectively. The sensitivity analysis yielded consistent results that six indicators were positively associated with an increased OSA risk, and the positive association was independent of a woman's menopausal status (supplementary table S 1 and S2).

Table 2 Association between adiposity indicators and obstructive sleep apnea
Full size table

In stratified analyses by age, sex, and dyslipidemia, the associations of NC, BF%, WHR, VAI, LAP, and RMR with OSA were not significantly changed (Tables 3, 4 and 5). The associations of NC, BF%, WHR, VAI, and RMR with OSA risk were stronger in the middle-aged than in the elderly (P -interaction < 0.001); the associations of VAI and RMR with OSA were slightly stronger in women than in men (P -interaction = 0.020 and 0.024, respectively); and the associations of BF%, WHR, and LAP with OSA were stronger in the non-dyslipidemia group than in the dyslipidemia group (all P -interaction < 0.05).

Table 3 Association between adiposity indicators and obstructive sleep apnea by age
Full size table
Table 4 Association between adiposity indicators and obstructive sleep apnea by sex
Full size table
Table 5 Association between adiposity indicators and obstructive sleep apnea by history of dyslipidemia
Full size table

Discussion

To our knowledge, this is the first study to comprehensively examine the effects of common and novel adiposity indicators on the risk of OSA. This large population-based study found that NC, BF%, WHR, VAI, LAP, and RMR were all independently and positively associated with the OSA risk. The stratified and sensitivity analysis yielded similar results, indicating the robustness of the results.

This study found that the OSA risk increased with NC increment, which was consistent with previous studies [13, 29]. Increased NC implies more adipose tissue adjacent to the upper airway, with consequent reduced upper airway caliber and predisposes to OSA [29]. By contrast, BF% has received little attention in the etiology of OSA. We found that every 1-unit increment of BF% was associated with a 9% increased risk of OSA, indicating excessive fat accumulation was a risk factor for OSA regardless of fat distribution.-T-he risk of tissue hypoxia develops as adipocyte hypertrophy continues, with subsequent inflammatory activation, oxidative stress, and increased sympathetic activity, which eventually leads to the occurrence of OSA [7].

Indicators of abdominal adiposity including WHR, VAI, and LAP were all found to be independent risk factors for OSA, which was consistent with previous studies [13, 30]. Two separate observational and longitudinal studies concluded that abdominal obesity characterized by WC and HC was more strongly correlated with OSA than general obesity in China [11]. A cross-sectional study suggested that VAI was significantly associated with OSA risk, with all significantly correlated with an apnea–hypopnea index (AHI), and mean and lowest oxygen saturation [31]. LAP was initially developed for recognizing cardiovascular risk and then applied in the identification of metabolic diseases and OSA. Zou et al. suggested that LAP might be one key exponent in screening for OSA [13]. Abdominal adiposity accumulation may reduce pharyngeal lumen size, decrease upper airway muscle protective force and size, and affect restrictive respiratory dysfunction, finally leading to daytime hypoxemia and the development of OSA [30, 32]. RMR was positively associated with the OSA risk in this study. A university-based cross-sectional study showed that increased resting energy expenditure was independently associated with AHI, resulting in greater severity of sleep-disordered breathing [33]. Another study conducted a three-month continuous positive airway pressure therapy for OSA patients and found that the basal metabolic rate (equal to the RMR) was reduced in the absence of changes in physical activity, thus favoring a positive energy balance in terms of energy expenditure [34].

The stratified analysis by age showed that the associations of NC, BF%, WHR, VAI, LAP, and RMR with OSA risk were stronger in the middle-aged than in the elderly. This disparity could be explained by the contradictory effect of adipose tissue distribution on the elderly. Many elderly obese may exhibit late-onset obesity, health risks, and comorbidities not manifest due to its short duration [35]. Besides, Tung and colleagues followed 4,000 older adults for 5 years and found that older men were resistant to hazards of overweight and adiposity; mild-grade overweight or obesity might be protective [36]. The aging process is indeed characterized by an increase in total body fat mass and a concomitant decrease in lean mass and bone density, independent of general and physiological fluctuations in weight and BMI [37]. A systematic review concluded that five-year increases in the visceral adipose tissue (VAT) area declined with the advanced age group in both men and women, regardless of race [38].

In the stratified analysis by sex, the negative associations of VAI and RMR with OSA were stronger in women than in men. Studies have shown that women tend to have higher percent body fat throughout the entire life span with relatively more adipose tissue deposited in the hips and thighs, while men tend to have a greater degree of visceral obesity with excess fat more concentrated in the abdomen and neck [32]. These yielded consistent results that women had higher BF% (33.6% vs. 24.7%), lower VAI (6 vs. 11), and lower WHR (0.86 vs. 0.91) than men. However, it is reported that menopause is followed by redistribution of adipose tissue towards a more central phenotype and raised visceral adiposity in women during the peri-menopausal transition presumably due to the fall in estrogen levels [32, 39]. 67.3% of the women in this study were menopausal. Sensitivity analysis showed that the association between adiposity indicators and OSA was independent of menopausal status, which indicated that even premenopausal women should pay more attention to OSA prevention. Moreover, the energy expenditure in women was lower than in men, and women were more susceptible to accumulating fat tissue, especially old-age women.

In addition, among non-dyslipidemia subjects, BF%, WHR, LAP, and RMR were more strongly associated with OSA risk than those with dyslipidemia. There are complex interactions between obesity, dyslipidemia, and OSA, and in many cases, they coexist. Studies have reported that dyslipidemia predisposes to excess fatty deposition in the neck, thorax, and abdomen, impacts the pulmonary system and thereby increases OSA susceptibility [40]. Participants not suffering from dyslipidemia may be more sensitive to visceral fat accumulation, leading to a higher risk of OSA, compared to those with dyslipidemia.

Study strengths and limitations

There are some strengths. First, the multi-stage sampling method was applied to recruit participants from the general population in Guangzhou communities, which greatly reduced the selection bias and enhanced the representativeness of the sample. Second, the large sample size improved the statistical power and allowed for comparisons by age, sex, and history of dyslipidemia. Third, this study considered the effect of regional fat distribution on OSA and combined traditional and novel parameters of adiposity. Finally, we performed several stratified and sensitivity analyses and the results of which showed consistent associations, indicating the robustness of our results to a certain degree.

Some limitations also exist. First, the cross-sectional design could not provide causal inference according to our report. However, the dose–response relationship between adiposity indicators and OSA enhanced the existence of causation. Second, OSA was determined by the Berlin Questionnaire due to the lack of polysomnography during data collection, which is a commonly used validated tool in epidemiological and clinical research [24]. Compared with many other screening questionnaires that are lengthy and complicated, the Berlin questionnaire has been widely adopted and validated in various populations because of its ease of use, efficiency, and good sensitivity. Third, adiposity indicators were measured by Omron body composition monitor, which may not provide measurements as accurate as other advanced methods, such as Dual-energy X-ray absorptiometry [41]. The accuracy of the measurements was susceptible to being affected by body temperature, food ingestion, ambient temperature, and humidity. However, the portable Omron device has been applied in several studies and could provide a rapid, non-invasive, and reasonably accurate measurement of body composition [42,43,44]. But considering the cost and convenience, it was more practical to use portable protocols in this large-scale population study.

Conclusion

NC, BF%, WHR, VAI, LAP, and RMR were all independently and positively associated with OSA risk, regardless of age, sex, history of dyslipidemia, and menopausal status. Application of these new indicators could help to more comprehensively reflect and predict the risk of OSA in the general population. More attention should be paid to the middle-aged, women, or non-dyslipidemia population.