Background

Over the last 40 years, the worldwide prevalence of obesity has grown such that more than 600 million people now live with obesit y [1]. In Asia, where around 60% of the world’s population resides, the rise in obesity amongst Asian populations has been fueled by rapid changes in diet and lifestyle practices due to increased economic growth and urbanization [2]. Increased adiposity is a major risk factor for many adverse health conditions such as type 2 diabetes, cardiovascular disease, depression and certain cancers, resulting in decreased quality of life, increased healthcare costs and increased mortality [3,4,5,6]. In particular, cardiovascular disease and diabetes were the primary causes of mortality and disability-adjusted life-years related to high body mass index (BMI) in 2015 [7]. Obesity is thus a major global and regional public health problem with substantial economic impacts that will increase over time if current trends continue [8, 9].

Obesity is a complex multifactorial disease influenced by both genetic and environmental factors [10]. Observational and interventional studies suggest that short sleep duration may be a modifiable risk factor for obesity. Meta-analyses of prospective observational studies consistently show that short sleep duration is associated with a higher risk of developing obesity in both adults and children [11,12,13]. Interventional studies in adults and children show that improved sleep duration results in lower weight [12, 14]. In contrast, adults subjected to short-term sleep restriction show evidence of weight gain [15].

While sleep duration is important, optimal sleep is now recognised to consist of good sleep quality in addition to adequate duration. Sleep quality includes additional consideration of sleep latency, number of arousals at night, depth of sleep, and satisfaction with sleep [16]. A meta-analysis of nine studies, based on 25,082 young subjects (0–34 years old), suggests a higher odds of overweight and obesity in subjects with poor sleep quality, independent of sleep duration [17]. In addition, Mendelian randomisation studies, which use genetic variants as instrumental variables and can assess causality, have suggested a causal relationship between dozing, daytime napping, snoring, and insomnia as causal risk factors for adiposity [18, 19]. Additionally, sleep disturbances are strongly linked to mood disorders such as generalised anxiety disorder and depression [20, 21], which may also influence the development of obesity [22].

Asian populations reported shorter sleep duration than other groups [23,24,25,26], raising the possibility of short/poor sleep contributing to the rising prevalence of obesity in Asia. Hence, the purpose of our study was to confirm the relationships between sleep duration, sleep quality, dozing, daytime napping, snoring, insomnia and adiposity in a multi-ethnic Asian population, and investigate the potential contribution of disturbed sleep to the high risk of obesity amongst Asian populations.

Methods

Study design and participants

We used the cross-sectional baseline data from the first 9971 people recruited to the Health For Life in Singapore (HELIOS) study (April 2018 to October 2021). The HELIOS study is a population-based prospective cohort study, initiated in 2017, with approval from the Nanyang Technological University Institutional Review Board (reference: IRB-2016-11-030). Singaporean (citizens or permanent residents) men and women, aged 30–84 years, are recruited from the general population through multiple community engagement strategies to ensure participation from the ethnic minority, working-age, and lower socio-economic groups. Pregnant, breastfeeding, acutely ill individuals, or those unable to give informed consent are excluded. To assess the reproducibility of the study’s measures, we carried out a complete re-assessment of a randomly selected sample of 292 participants, 17.4 ± 9.9 months after their initial recruitment.

Demographics and clinical characteristics

Participants’ sex, age, and ethnicity were recorded according to national registry data. Educational level, household income, lifestyle, and medical history data were collected via self-administered questionnaires. Regular alcohol drinking was defined as once monthly or more. Height and weight were measured using the BSM 370 automatic stadiometer (Inbody, Seoul, South Korea). BMI was calculated by dividing weight in kilograms by the square of height in meters, with obesity defined as BMI ≥ 30 kg/m2 according to World Health Organisation (WHO) guidelines [4]. Waist circumference was measured as the horizontal distance at the mid-point between the iliac crest and the lowest rib as recommended by WHO [27] and the International Diabetes Federation [28]. Hip circumference was measured as the horizontal distance at the maximum circumference of the buttocks over the greater trochanters. The mean of three readings was recorded. Similarly, blood pressure readings from the right arm were taken thrice and averaged, using an automated blood pressure device (Omron HEM-9210T & GE Carescape V100 Vital Signs monitor). Fasting blood samples were collected after an 8-hour fast and sent to an internationally and locally accredited laboratory (Innoquest Diagnostics Pte Ltd), where fasting blood glucose (ADVIA 1800 Chemistry system, Siemens & Cobas 8000, Roche Diagnostics) and haemoglobin A1c (HbA1c) (Bio-Rad D-100 Haemoglobin Testing System & Cobas c513, Roche Diagnostics) were quantified. Diabetes was defined as either self-reported (‘yes’ or on medication for diabetes), fasting blood glucose ≥ 7.0 mmol/L, or HbA1c ≥ 6.5%. Hypertension was defined as either self-reported (‘yes’ or on medication for hypertension), measured systolic blood pressure ≥ 140 mmHg, or diastolic blood pressure ≥ 90 mmHg.

Self-reported sleep traits

The Pittsburgh Sleep Quality Index (PSQI) questionnaire [29], self-administered on an electronic platform, was used to assess sleep quantity and quality over the past month. This 19-item self-rated questionnaire is widely validated and the most commonly used standardised measure of sleep in clinical and research settings [30]. It produces scores for 7 components of sleep: subjective quality, latency, duration, efficiency, disturbances, medication use, and daytime dysfunction. These are summed to a global score ranging from 0 (no difficulty) to 21 (severe difficulties in all areas). A score of > 5 has been found to distinguish good and poor sleepers with both sensitivity and specificity close to 90% [29], and hence, was considered indicative of poor sleep quality. For sleep duration, <7 h of sleep was considered inadequate since 7 h of sleep is widely considered to be the lowest sleep duration optimal for health in adults [31], while <5 h was considered grossly inadequate (sleep deprivation), with more extreme outcomes [32, 33]. Participants who reported a sleep duration that exceeded the duration calculated from ‘usual bedtime’ and ‘usual getting up time’ were excluded as this was considered implausible. As a sensitivity analysis, we also calculated and used the ‘Perceived Sleep Quality’ score [34], a score based on 5 out of the 7 components (quality, latency, disturbances, medication use, and daytime dysfunction).

In this study, dozing, snoring and insomnia were ascertained from questions that are part of the PSQI questionnaire. Dozing was defined as having trouble staying awake while driving, eating meals, or engaging in social activity ≥ 3 times/week; snoring was defined as having trouble sleeping because of cough or loud snoring ≥ 3 times/week; and insomnia was defined as not being able to get to sleep within 30 min, or waking up in the middle of the night or early morning, ≥ 3 times/week. Participants were also asked ‘Do you have a nap during the day?’, with responses ‘frequently’ or ‘always’ being positive for daytime napping. As a sensitivity analysis, possible obstructive sleep apnoea (OSA) was considered in our analysis of snoring, with the possibility of OSA being measured by a modified STOP-BANG risk scale used by other studies [35]. This modified version excludes the question ‘Has anyone observed you stop breathing during sleep?’ and replaces neck circumference with waist circumference, dichotomised according to Asian cut-offs [36] for our study. We considered scores of ≥ 5 as high risk for OSA based on the original STOP-BANG criteria [37], and also scores of ≥ 4 to account for the modification, in our sensitivity analyses.

Covariates: anxiety and depression

The Generalised Anxiety Disorder (GAD)-7 questionnaire and the Patient Health Questionnaire (PHQ)-9, were used to assess the prevalence of anxiety and depressive symptoms, respectively. The GAD-7 comprises 7 items, while the PHQ-9 includes 9. Scores for each item range from ‘0’ (None at all for the past 2 weeks) to ‘3’ (Nearly every day for the past 2 weeks). The total score is computed by summing the individual scores across all the items. Scores of < 5 were considered normal, whereas scores of ≥ 5 represented possible anxiety [38] or depression [39]. Both instruments have been rigorously validated and are recognized as reliable and efficient tools for both clinical assessment and research applications [38, 39].

Data Analysis

All continuous variables were assessed for normality using histograms and skewness, with age, anthropometry measurements, sleep duration, and sleep quality (PSQI score) considered to be normal. Hence, these variables are presented as mean ± standard deviation. Differences among the three ethnic groups were analysed using the chi-square test for categorical data and one-way analysis of variance (ANOVA) with post-hoc Tukey’s tests for continuous data. Partial correlation was used to analyse the correlation between continuous sleep traits (sleep duration, and sleep quality) and adiposity (BMI and waist circumference), adjusted for age, sex, and ethnicity. Next, multivariable linear regression models were used to further investigate the relationship between sleep traits and adiposity (BMI and waist circumference), progressively adjusting for additional covariates across three models. Model 1 adjusted for age, sex, and ethnicity; Model 2 included education, household income, smoking and alcohol drinking status, and the presence of diabetes and hypertension; and Model 3 added possible anxiety, and depression to explore their impact.

To confirm the role of sleep quality, insomnia and snoring underlying adiposity, stratification by sleep duration (< 5 h, ≥ 5–7 hours, ≥ 7 h) was done with the three regression models. Stratification by ethnicity, with testing for heterogeneity of effect, was also done to investigate whether these relationships vary across the three ethnic groups. Additionally, the impact of sleep duration, sleep quality, insomnia and snoring on the relationship between ethnicity and adiposity was also assessed using multivariable linear regression models. Finally, based on the odds of obesity (BMI ≥ 30 kg/m2) associated with sleep deprivation (< 5 h)/snoring, and the proportion of those with sleep deprivation/snoring among people with obesity (total, and by ethnicity), the estimated population attributable fractions were calculated and compared between the ethnic groups.

To assess the reproducibility (test-retest reliability) of the PSQI questionnaire (and the additional question on daytime napping) used to measure the sleep traits, the intraclass correlation (two-way mixed effects model, absolute agreement) and Pearson correlation were calculated for sleep duration and sleep quality (continuous data), while the Cohen’s weighted Kappa was calculated for insomnia, dozing, napping, and snoring (categorical data).

Statistical analyses were performed using the software package IBM SPSS Statistics for Mac, version 27.0 (IBM Corp., Armonk, New York, USA), and Stata Statistical Software, version 14.0 MP (Stata Corp LLC, College Station, Texas, USA).

Power calculations

Based on previous studies on local cohorts [40, 41], a population of 10,000 people would provide 90% power to detect a difference of at least 0.5 kg/m2 in BMI, 0.13 h/day in sleep duration, and 0.3 in global PSQI score between the Chinese and the ethnic minorities (Malays and Indians) at P < 0.05, which are all within the range of expectations according to previous studies and national health survey [41,42,43] (Supplementary Table S1).

Results

Study participants

Of the 9971 study participants, we excluded 1095 participants with incomplete data, implausible PSQI data or who were not of Chinese, Malay or Indian ethnicity (Supplementary Fig. S1). The characteristics of the 8876 participants included in the study are shown in Table 1. There were 6151 (69.3%) Chinese, 1106 (12.5%) Malays, and 1619 (18.2%) Indians, with more females (59.6%) than males (40.4%), and a mean age of 51.7 ± 11.8 years. Compared to Malays and Indians, Chinese had a higher proportion of individuals with diploma-level education and above, as well as a higher household income exceeding SGD$10,000. Additionally, Chinese had a lower proportion of individuals with possible anxiety and depression (P < 0.001).

Table 1 Participant characteristics and sleep traits by ethnicity (n = 8876).
Full size table

Adiposity and sleep traits

Mean BMI and waist circumference were 24.8 ± 4.6 kg/m2 and 83.0 ± 11.7 cm, respectively, for the study population (Table 1). Chinese had the lowest mean weight, BMI, and waist circumference (WC) amongst the three ethnic groups (P < 0.001). Obesity prevalence (based on BMI ≥ 30 kg/m2) was 6.1%, 29.5%, and 23.6% for Chinese, Malays, and Indians respectively (P < 0.001). Both BMI and waist circumference were associated with sex, age, income, education, smoking, regular alcohol drinking (except BMI), diabetes, hypertension, possible anxiety and depression (P < 0.01, Supplementary Table S2).

Mean sleep duration and PSQI score were 6.27 ± 1.2 h and 5.53 ± 1.2 (Table 1), respectively, with 51.0% reporting symptoms of either insomnia (36.7%), dozing (1.4%), daytime napping (16.2%), snoring (11.6%) or a combination of these symptoms (12.9%) (Table 1, Supplementary Table S3). Test-retest reliability analyses showed that there was moderately good reliability for the measurement of sleep duration and sleep quality, and fair to moderate agreement for insomnia (Supplementary Table S4).

Malays had the shortest sleep duration (5.74 ± 1.3 h), compared with Chinese (6.37 ± 1.1 h) and Indians (6.27 ± 1.2 h) (P < 0.001), with only 24.8% of Malays achieving adequate sleep (≥ 7 h), and 40.1% of them having less than 5 h of sleep, when compared with Chinese and Indians (P < 0.001, Table 1). Chinese had the longest sleep duration and best sleep quality based on the lowest mean PSQI score (5.30 ± 2.7), compared with Malays (6.11 ± 3.0) and Indians (6.03 ± 3.0) (P < 0.001). However, there were still 38.7% of Chinese considered as poor sleepers (PSQI > 5), with Malays and Indians having higher proportions at 52.1% and 52.0%, respectively (P < 0.001). Chinese also had the lowest proportion with insomnia, dozing, daytime napping and snoring, when compared with the Malays and Indians (P ≤ 0.001).

In addition to ethnicity, sleep traits were closely associated with multiple socio-demographic and measured biological phenotypes (Supplementary Fig. S2, Supplementary Table S5). Sleep duration and sleep quality were closely and inversely related to symptoms of anxiety and depression, alcohol intake, presence of diabetes and hypertension, and positively related with education status. Insomnia, snoring, daytime napping and dozing were similarly strongly associated with symptoms of mood disorders, and were typically also associated with alcohol intake, education status, and presence of cardiometabolic diseases (Supplementary Fig. S2, Supplementary Table S5).

Univariate relationship of sleep traits with adiposity

In univariate analysis, there was a strong association of all sleep traits with BMI and waist circumference (Fig. 1, Supplementary Table S6). BMI and waist circumference were higher amongst people with short sleep, poor sleep quality, insomnia, dozing, daytime napping and snoring compared to people without these sleep traits (P < 0.01, Supplementary Table S6). For example, mean BMI was 26.3 ± 5.4 kg/m2 amongst people sleeping <5 h per night, compared to 24.4 ± 4.8 kg/m2 amongst people with normal sleep duration (P < 0.001). The strongest association between sleep traits and adiposity was seen for snoring. People reporting snoring had 2.9 ± 0.2 kg/m2 higher BMI and 7.8 ± 0.4 cm higher waist circumference compared to those without snoring.

Fig. 1: BMI and waist circumference by sleep duration and sleep quality (PSQI score) (n = 8876).
figure 1

BMI by (a) sleep duration and (c) by sleep quality (based on PSQI score). Waist circumference by (b) sleep duration and (d) by sleep quality (based on PSQI score) (n = 8876).

Full size image

Multivariate relationship of sleep traits with adiposity

The relationships of sleep duration, sleep quality, and snoring with both BMI and waist circumference were independent of age, sex, and ethnicity, and remained significant after additional adjustment for education, household income, current smoking and regular alcohol drinking status, presence of diabetes and hypertension, and markers for anxiety and depression (P < 0.005, Table 2). The relationships of insomnia, dozing, and daytime napping with adiposity were weak and of marginal statistical significance after adjusting for age, sex, and ethnicity, and close to null in fully saturated statistical models.

Table 2 Multivariable linear regression models between sleep traits and BMI, Waist Circumference (WC) (n = 8876).
Full size table

Sleep duration and sleep quality were highly correlated (0.583, P < 0.001, Supplementary Table S7). Nevertheless, stratified analysis suggested an independent effect of sleep quality on adiposity amongst people with short sleep (P < 0.05, Table 3), and in sensitivity analysis that excluded sleep duration from the assessment of sleep quality (Perceived Sleep Quality) (Supplementary Tables S8, S9). Snoring was found to be strongly associated with BMI and waist circumference independent of sleep duration (Table 3) and sleep quality (Supplementary Table S10), and even in people with normal adiposity (Supplementary Table S11), or low probability of obstructive sleep apnea (Supplementary Table S12).

Table 3 Multivariable linear regression models between sleep quality (PSQI score), snoring and BMI, Waist Circumference (WC) by sleep duration (n = 8876).
Full size table

Stratification by ethnicity confirmed the relationship of both sleep duration and sleep quality with BMI and waist circumference in Chinese people (n = 6151) (P < 0.001, Supplementary Table S13). The relationships were statistically weaker in the smaller Malay (n = 1106) and Indian (n = 1619) subgroups (Supplementary Table S13). There was no statistical evidence for heterogeneity of effect between the ethnic groups, in the relationships of adiposity with sleep, except for sleep quality and BMI in Model 1 (Supplementary Table S14). Snoring was strongly associated with BMI and waist circumference in all three ethnic groups (P < 0.001, Supplementary Table S13).

Ethnicity as a predictor of adiposity-related sleep traits

In multivariable regression analyses, Malay ethnicity remained strongly associated with shorter sleep duration, lower sleep quality and increased snoring, compared to Chinese people, after adjustment for age, sex, education, household income, current smoking and regular alcohol drinking status, presence of diabetes and hypertension, and markers for anxiety and depression (P < 0.01, Supplementary Table S15). In contrast, the relationships between Indian ethnicity and adiposity-related sleep traits were weaker and generally no longer statistically significant after adjustment for socio-demographics and co-morbidities (Supplementary Tables S15, S16).

Contribution of sleep to adiposity in Asian populations

We estimated the population attributable fraction (PAF) for short sleep and snoring as a contributor to obesity to be 6.6% and 18.6% respectively (Table 4). The PAF for short sleep was higher amongst Malays (9.5%, 95% CI: 3.5–15.1) than the Chinese (5.0%, 95% CI: 1.9–8.0) and Indians (5.8%, 95% CI: 2.2–9.2) (P = 0.005, P = 0.009, respectively) (Table 4), while the PAF for snoring was higher amongst Chinese (20.6%, 95% CI: 15.5–25.7) compared to the Indians (10.7%, 95% CI: 5.7–15.7) (P = 0.006). Finally, we assessed whether sleep duration, sleep quality, and snoring account for the difference in BMI and waist circumference between ethnic groups. In multivariable regression analyses, Malay and Indian ethnicity remained strongly associated with BMI and waist circumference, after adjustment for sleep duration, sleep quality, insomnia, and snoring (P < 0.001, Fig. 2). Sensitivity analysis based on perceived sleep quality confirmed these findings (Supplementary Fig. S3).

Table 4 Population Attributable Fractions (PAF), total and by ethnicity, for sleep deprivation (<5 h) and snoring as a contributor to obesity (BMI ≥ 30 kg/m2) (n = 8876).
Full size table
Fig. 2: Forrest plot of unstandardised beta of Malay and Indian ethnicity in multivariable regression models predicting BMI and waist circumference (n = 8876).
figure 2

Model 1: age, sex. Model 2: 1 + education, household income, current smoking, regular alcohol drinking, diabetes and hypertension. Model 3: 2 + sleep duration and sleep quality. Model 4: 2+ snoring. Model 5: 2 + sleep duration, sleep quality and snoring.

Full size image

Discussion

The rising prevalence of obesity in Asian populations is a major public health problem [8, 9], and a key determinant of the high burden of diabetes and cardiovascular disease in the region [7]. In the present study, we show an independent relationship between adiposity and key indicators of sleep duration and quality. Our results suggest that sleep disturbances may make an important contribution to the increased risk of obesity in Asian populations.

We investigated 8876 people participating in a community-based, multi-ethnic population study conducted in Singapore. We quantified obesity and sleep using validated and reproducible tools. Our results demonstrate substantial differences in adiposity between Asian sub-groups, with raised BMI, waist circumference, and obesity prevalence amongst Malays and Indians compared to Chinese. We also show a high prevalence of reduced sleep duration and sleep quality in our participants, and that both sleep duration and sleep quality are strongly and inversely associated with measures of adiposity. Notably, sleep duration and quality are reduced in both Malay and Indians compared to Chinese. Our results suggest that poor sleep may be an important contributor to the rising burden of adiposity in Asian populations, and further suggest that improving sleep duration and quality could mitigate obesity rates through targeted public health interventions.

Rapid economic development and technological advancements leading to shifts in lifestyle patterns towards positive energy balance [2, 44], and shorter and poorer sleep [45], have increased obesity rates in many developing Asian countries [2, 44]. However, there is considerable heterogeneity in obesity prevalence across these countries. In 2016, age-standardised prevalence of obesity ranged from 6.2% in China (East Asia), to 3.9% and 8.6% in India and Pakistan respectively (South Asia), to 6.9% and 15.6% in Indonesia and Malaysia respectively (Southeast Asia) [46]. Even within the same country, obesity rates can vary significantly between ethnic groups. For example, in Singapore, a multi-ethnic Asian city-state, the rates were 24% for Malays, 16.9% for Indians, and 7.9% for Chinese in 2010 [42]. Understanding these patterns and the contributing factors to obesity is crucial for promoting health and human potential.

Sleep is an active physiological process necessary for physical, mental, and emotional health [47]. Multiple epidemiological and interventional studies have demonstrated the detrimental effects of sleep deprivation and poor sleep on health, including the development of obesity [11,12,13, 48, 49]. Meta-analyses of prospective observation studies show that short sleep duration is consistently associated with an increased risk of obesity [11,12,13], and interventional studies have demonstrated that sleep restriction and sleep extension, increases [15] and decreases adiposity [14], respectively. Various mechanisms such as metabolic dysregulation, disruptions in appetite-regulating hormones and neurotransmitters, activation of inflammatory pathways, gut dysbiosis, and reduced physical activity have been found to link short/poor sleep and obesity [50]. Recent Mendelian randomisation studies have also supported a causal role for dozing, daytime napping, snoring, and insomnia on adiposity [18, 19]. Furthermore, sleep abnormalities such as short sleep appear to be more common amongst ethnic minorities [51]. Possible reasons include a higher prevalence of health-related issues, work-related stress, long working hours, and cultural factors such as prioritizing family and social obligations over rest [25]. In Singapore, people of Malay background are reported to have shorter sleep and poorer sleep quality [16, 41, 43, 52], raising the possibility that differences in sleep traits such as duration and/or sleep quality could indeed contribute to increased adiposity in Malays.

We found that sleep duration, sleep quality, and snoring were associated with BMI (overall adiposity) and waist circumference (central adiposity) after accounting for age, sex, ethnicity, education, household income, current smoking, and regular alcohol drinking status, presence of diabetes and hypertension and even after considering possible anxiety and depression. Stratification by sleep duration showed that sleep quality was associated with adiposity only when sleep duration was <5 h, suggesting that while both sleep duration and quality are important in predicting adiposity in our multi-ethnic Asian population, sleep quality is less important when there is adequate sleep duration. This may be because poor sleep quality exacerbates adiposity when sleep is severely inadequate, while adequate sleep mitigates these negative effects. Snoring remained strongly associated with BMI and waist circumference after considering sleep duration, sleep quality, probability of obstructive sleep apnoea, and even in people with normal adiposity. This is consistent with literature suggesting that snoring may not be benign and could precede the development of obstructive sleep apnoea and metabolic syndrome [53].

However, we found that even after adjusting for sleep duration, sleep quality, insomnia, and snoring, ethnicity remained a strong predictor of adiposity. This suggests that the disparities in sleep duration and sleep quality observed in Malay and Indian individuals in Singapore do not fully explain the high burden of obesity in these populations. Consequently, this raises the possibility of additional mechanisms that contribute to obesity amongst these Asian ethnic groups. Notably, risk allele frequencies and effect sizes of BMI-associated genetic variants can vary across ethnic groups [54], while other lifestyle factors such as diet and physical activity can also differ substantially between ethnicities [55, 56]. For example, South Asian diets tend to be high in carbohydrates with lower protein intake, whereas East Asian diets are typically higher in protein due to a greater emphasis on meats and seafood [55]. Nonetheless, our results suggest that if the relationship of sleep on adiposity is causal as discussed, approximately 6.6% of obesity could be prevented through improved sleep duration. The potential impact would be even higher amongst Malay and Indian individuals who have shorter sleep duration. Similarly, our results for snoring suggest that approximately 18.6% of obesity might be avoided through action to prevent snoring if snoring is indeed a causal factor, as supported by Mendelian randomisation studies [18, 19] and mechanistic studies [53]. Our findings support the development of public health interventions to promote sleep duration and quality, as well as further research to understand the mechanisms underlying the relationships between short sleep, snoring and adiposity in Asian populations.

Our study has strengths and limitations. We studied a large community-based sample of multi-ethnic Asian individuals using validated and reproducible tools including PSQI, GAD-7, and PHQ-9. We included multiple socio-demographic factors as well as physical and psychological co-morbidities in our statistical models to account for potential confounding effects. We carried out sub-group and sensitivity analyses to further demonstrate the robustness of our findings. Our study does have some limitations. First, the study population was predominantly Chinese, with a lower proportion of Malay and Indian individuals. Nevertheless, we show that the characteristics of our study population closely mirror those of the Singapore population supporting the view that our participants are representative of the respective Asian communities. Second, the measurements for the sleep traits were questionnaire-based, and self-reported which is susceptible to recall bias and response bias, and were not clinically or objectively confirmed. However, PSQI has been extensively validated in a variety of populations and is a recognised and acceptable measurement tool [30]. Our analysis of 292 participants with repeat PSQI data also suggests that there is moderately good test-retest reliability. In addition, the findings of this cross-sectional study are not conclusive evidence of a causal relationship between sleep and obesity amongst the ethnic groups in Singapore, although such a causal relationship is supported by existing interventional and Mendelian randomisation studies.

Conclusion

In conclusion, we show that sleep duration, sleep quality, and snoring are associated with adiposity in a multi-ethnic Asian population of Chinese, Malays, and Indians. Our results raise the possibility that an important fraction of obesity in Asian populations could be averted through public health interventions aimed at improving sleep duration and quality.