Associations Between Night Sleep Duration and Fasting Glucose and Triglyceride To High-Density Lipoprotein Cholesterol Ratio Among Adults Free of Type 2 Diabetes or Without Diagnosed Type 2 Diabetes

We aim to assess the associations between night sleep duration and fasting glucose (FG), triglyceride (TG) to high-density lipoprotein cholesterol (HDL-C) ratio, and body mass index (BMI) among adults free of type 2 diabetes (T2D) or without diagnosed T2D. Methods We analyzed the baseline data of a cohort. We included adults free of T2D or without diagnosed T2D who completed the validated questionnaire, biochemical and anthropometric measurements. Independent association between sleep duration and FG, TG/HDL-C, BMI was evaluated with multiple U-shaped or linear regressions.


Introduction
According to the International Diabetes Federation Diabetes ATLAS (9th edition) data released in 2019, approximately 463 million adults aged 20-79 were suffering from diabetes globally, with a crude prevalence rate of 9.3% [1]. Type 2 diabetes (T2D), accounting for around 90% of all diabetes worldwide, has shown an increased prevalence in China in recent decades. In parallel, there has been an increased sleep deprivation in China [2]. Sleep is a fundamental and complex physiological process. Long-term sleep deprivation will cause a series of physical and mental illnesses [3]. Optimal sleep duration is one of the critical factors to maintain metabolic balance [4]. Both short and long sleep duration reduce insulin sensitivity and induce the production of a variety of in ammatory cytokines, which will affect the body's

Sleep Assessments
The modi ed Pittsburgh Sleep Quality Index (PSQI) questionnaire, which has been proved to have high reliability and validity in Chinese population [14], was used to determine self-reported items. Participants were asked questions regarding sleep in the past month: "when have you usually gone to bed at night?", "when have you usually gotten up in the morning?" and "how long does it usually take you to fall asleep each night?". We calculated participants' average daily night sleep duration based on these questions.
The average self-reported sleep duration was reported by hours and was calculated by the time to fall asleep and the time to get up. If they did not indicate an exact period, we used the lled period's middle time point. For example, if participants report they usually go to bed at 22:00 ~ 23:00, then we used 22:30 as the point of time. According to the distribution of different night sleep duration and studies [15], the self-reported sleep duration was analyzed categorically (<7 hours, 7 to <8 hours, 8 to <9 hours, and ≥9 hours). Subjective sleep quality was asked by the following question: "What do you think of your sleep quality in the prior month?". Answer options included excellent, good, bad, and awful. Subjective sleep quality has been proved to have the highest correlation with the global score among the seven factors when researchers evaluated the Chinese version of PSQI[16], thus we used it to assess the participants' sleep quality. Besides, participants were also asked, "How often do you work at night shift in the past month?" and "how often do you use sleeping medication?". Response options included (1) not during the past month; (2) less than once a week; (3) once or twice a week; (4) three times or more a week.
Fasting glucose, Triglycerides, and High-Density Lipoprotein Cholesterol Participants were required to fast for at least 12 hours before the blood sample was obtained the following morning. TC and HDL-C were measured using the Hitachi 7600 automated biochemical analyzer (Hitachi, Inc., Tokyo, Japan). All procedures were performed by the trained professional medical assistant and staff at the medical examination centers.

Anthropometric measurement
All anthropometric measurements were assessed with participants wearing light clothing without shoes before breakfast. Height was measured to the nearest 0.1 (cm), and weight (kg) was measured to the nearest 0.1 (kg) by the professional medical staff of the medical examination centers using the same calibrated electronic scale (GL-310, Seoul, Korea), zeroed before each measurement. BMI was expressed as body weight (kg) divided by squared body height (m 2 ).

Demographic Characteristics
Participants were asked to report their age, gender, ethnic group, highest education level, marital status, and employment status.

Other Covariates
Alcohol drinking, smoking and physical exercise were also evaluated in the questionnaire. Participants who reported had been smoking for more than half a year were de ned as smokers. In contrast, those who reported having smoked but later quit for a sustained period of half a year or longer by the interview time were former smokers. Those who indicated consuming alcohol at least once a week were de ned as alcohol drinkers. Those that had been away from drinking alcohol for half a year or longer by the time of the interview were drinking abstainers. Physical exercise was described as an exercise once a week or more for at least half an hour each time.

Statistical Analysis
Data were input, stored, and managed centrally. Sleep duration was categorized into 4 groups (<7h, 7 to <8h, 8 to <9h, and ≥9h) in the statistical analyses. Outcomes in this study were FG, TG/HDL-C, and BMI. Continuous variables tested normally distributed were expressed as mean ± SD. TG/HDL-C, which conformed to a right-skewed distribution, was logarithmically converted to lg (TG/HDL-C). Oneway analysis of variance (ANOVA) was used to perform the unadjusted group comparisons of normally distributed continuous variables. The other continuous variables not normally distributed were summarized as median with interquartile range (IQR). Categorical variables were described as percentages and compared using Pearson's test.
Based on the scatter plot and the mean graph of ANOVA, we assumed that night sleep duration has curvilinear (U-shaped) correlations with FG and TG/HDL-C and linear correlation with BMI. Multiple linear and curvilinear regression models were performed to identify the assumed independent linear or Ushaped associations of FG, TG/HDL-C, and BMI with sleep duration using FG and TG/HDL-C as dependent variables. In Model 1, FG and TG/HDL-C were regressed on night sleep duration after adjusting for sex, age, occupation, highest education, marriage status, physical exercise, BMI, subjective sleep quality, and sleeping drug use. In Model 2, FG and TG/HDL-C were regressed on night sleep duration using both a linear and a quadratic term to identify the assumed U-shaped relation after adjusting for sex and age. In Model 3, BMI was additionally included as a possible explaining factor to the relation tested in Model 2. In Model 4, occupation, highest education, marriage status, physical exercise, subjective sleep quality, and sleeping drug use were additionally included as possible confounders.
In Model 5, a multiple linear regression model was performed to explore the independent association of BMI with sleep duration after adjustment of sex, age. In Model 6, occupation, highest education, marriage status, physical exercise, subjective sleep quality, and sleeping drug use were included as potential confounding factors. It was tested whether the linear associations of BMI with sleep duration in Model 5 remain present. Two-side P < 0.05 was considered statistically signi cant. We used Epidata 3.0 to input and manage the data and Stata 15.1 (College Station, TX, USA) to analyze the data.

Participants' Characteristics
Overall, 40,525 participants completed the baseline questionnaire. After excluding those who were currently taking hypoglycemic agents (n=2,322), or lipid-lowering agents (n=780), or missing metabolic indicators (FG, TG, HDL-C), or height or weight (n=4,944), a total of 32,497 individuals were included in the nal analysis. The study population selection process is shown in Figure 1. Key participant's characteristics are summarized in Table 1. Participants' age ranged from 18 to 99 (median = 44, IQR: 35-56) years, and over half (52.5%) of the participants were men. Self-reported sleep duration and sleep quality were also presented in Table 1. Speci cally, characteristics were strati ed by sleep duration. In short, the proportions of participants who reported different sleep durations were 12.80% (<7 hours), 38.06% (7 to <8 hours), 39.47% (8 to <9 hours), and 9.67% (≥9 hours). More than forty percent (42.54%) of the participants reported excellent sleep quality. None of the participants work on the night shift in our cohort.

Associations of Night Sleep Duration with FG, TG/HDL-C, and BMI
Results of one-way ANOVA were presented in Table 2. We found statistically signi cant differences in FG, TG/HDL-C and BMI at different night sleep durations (all P<0.001, shown in Figure 2). Compared with those reported sleep durations was 7 to< 8 hours, FG was higher among sleep duration was <7 hours and ≥9 hours (P= 0.018 and P=0.008, respectively). Similarly, TG/HDL-C was signi cantly higher among sleep duration <7 hours and 8 to<9 hours than the reference group (P= 0.001 and P=0.002, respectively). In terms of BMI, the average BMI among those who reported less than 7 hours of sleep duration was signi cantly higher than the other three groups (P≤0.001 for all).
We observed a U-Shaped relationship for sleep duration and FG and therefore tested whether the Ushaped relationship was statistically signi cant. Results were shown in

Discussions
Using a large, FG diverse sample free of T2D or without diagnosed and untreated T2D, our current study demonstrated an independent U-shaped relationship between sleep duration and FG with the optimal sleep duration 6.9 hours, indicating that both short and long night sleep were associated with higher measures of glycemia after controlling for demographic characteristics, lifestyles and BMI. We also found a BMI mediated association between sleep duration and TG/HDL-C after adjusting for sex and age, and an inverse linear association of sleep duration with BMI after control of demographic characteristics and lifestyles.
The ndings of the U-shaped relationship between sleep duration and FG are consistent with the research among overweight or obese adults with prediabetes or recently diagnosed untreated T2D in the United States [12]. Similar results were also found in a cross-sectional study among adults aged over 40 in China [17], but the association was only found among female participants, where short (< 6 hours) and long (≥8 hours) night sleep duration increased the risk of higher measures of glycemia (OR=1.12, OR=1.14). However, ndings on the relationship between sleep duration and glycemia or diabetes were equivocal. A cross-sectional study conducted in South Korea only observed the association between short sleep duration and impaired fasting glucose (IFG) among male participants[18], whereas long sleep duration was not associated with IFG. However, a cohort study conducted in China among middle-aged and older adults revealed that long sleep duration was an independent predictor of incident diabetes[8], and similar results were con rmed among a group of retired Chinese adults [9]. We speculate there are several potential reasons for the inconsistency of the results. First, the above studies were conducted in different settings and among diverse populations, where sleep duration and subjective sleep quality vary across cultures and countries [19,20], thus may alter the associations between sleep pro les and diabetes risks [21,22]. Second, differences in participants' demographic characteristics may also in uence the correlation between sleep duration and blood glucose.
Multiple experimental studies have assessed the role of sleep in controlling energy balance and glucose metabolism and revealed that sleep restrictions could lead to glucose dysregulation through several biological pathways. One is by altering levels of neurohormones that regulate eating behaviors. It is suggested that acute sleep deprivation leads to a 28% increase in the average level of appetitestimulating hormones, while the average level of leptin decreases by 18%. Besides, lack of sleep is associated with increases in hunger, appetite, and hedonic food intake, especially for high-calorie food [23]. Another possible way for insu cient sleep to induce T2D is through activation of the sympathetic nervous system. Insu cient sleep can reduce glucose tolerance, increase cortisol levels and heart rate variability. These neuroendocrine changes interfere with the mechanism of regulating blood glucose, resulting in elevated steady-state blood glucose concentrations [24]. Additionally, activation of in ammatory pathways may also play a role in the link between short sleep duration and impaired glucose metabolism. Studies have con rmed that insu cient sleep is associated with increases in tumor necrosis factor, interleukin 6 and C-reactive protein [25,26]. Long sleep duration has many similar potential metabolic problems, but it may differ from short sleep duration [6,27]. Several studies have demonstrated that prolonged sleep duration can affect energy metabolism throughout the human body and thereby increase the risk of T2D through various possible complex mechanisms, including poor sleep quality, sedentary lifestyle, unhealthy dietary habits, and circadian rhythm [28]. These results suggest that interventions in behavioral patterns such as a sedentary lifestyle may promote healthy sleep habits, balance fasting blood glucose, thus prevent T2D.
In the current study, we only observed the U-shape relationship between night sleep duration and TG/HDL-C after adjusting for age and gender. After adjusting for BMI, this association was no longer signi cant, suggesting BMI may mediate the association between sleep duration and TG/HDL-C. At present, the association between sleep duration and TG/HDL-C is not consistent in the literature. A cross-sectional study of Danish adults did not nd a signi cant correlation between sleep duration and TG/HDL-C [29], while a cohort study among Mexican adolescents revealed that the two variables were independently related [30]. Considering the importance of adjusting BMI on the results, there are at least two explanations for the BMI adjusted model. First, insu cient sleep and sleep disturbance may increase BMI, leading to impaired glucose homeostasis. There is evidence that BMI may mediate the association between insulin and glycosylated hemoglobin concentrations and short sleep duration [31]. Secondly, our research found that BMI decreases with the increase of sleep duration, and thereby BMI may be a confounding variable between sleep duration and TG/HDL-C [31]. Future studies should research the effect of sleep duration on insulin resistance.
Lastly, our ndings revealed an inverse association between night sleep duration and BMI. This nding is consistent with previous studies among adolescents and adults [32,33], and the association was generally consistent across different categories of age, educational level, smoking status, baseline body mass index, and physical activity level [34]. Although some studies suggested that long sleep duration is also a risk factor for obesity and BMI increase [35,36], the evidence is insu cient among different races and populations [37]. Many factors such as race/ethnicity, tobacco use, sedentary behaviors may confound the proposed association [34]. Short sleep duration has been proved to be a potential risk factor for obesity and insulin resistance [5]. Insulin resistance is considered a common pathological basis of metabolic syndrome and a risk factor for T2D [38]. This study found that sleep duration is U-shaped with blood glucose, linearly related to BMI, and the U-shaped relationship with TG/HDL-C disappeared after adjusting BMI, which suggested that sleep duration has more complex effects on blood glucose and insulin resistance. These associations may also indicate that short sleep duration can either directly increase blood glucose or affect insulin resistance through different pathways, including obesity, which indirectly leads to glucose dysregulation and breaks blood glucose homeostasis. It can also directly increase insulin resistance by changing glucose metabolism [39]. Our ndings suggest that an increase within a reasonable range in sleep duration can reduce excessive fasting blood glucose and reduce BMI simultaneously, thereby reducing the risk of T2D.
Our study has several strengths. First, our cohort has a large number of adults as study samples. The age and gender composition are close to that of the total Chinese population [40], the age span is large (18-99 years), and participants have diverse occupational and educational backgrounds. All these increased the generalizability of our ndings. Secondly, we studied adults with various diabetes status, including adults with normal blood glucose, prediabetes (impaired blood glucose), and adults with T2D but were not diagnosed or receiving treatment. In addition, we excluded adults taking lipid-lowering drugs. Therefore, the interference of treatment on blood glucose and blood lipids was well controlled when analyzing the effect of sleep on FG and TG/HDL-C. Lastly, this is a multi-site study. Participants were recruited from 7 different centers in the Beijing-Tianjin-Hebei Region, and a standardized sleep questionnaire was used across all centers. All metabolic markers' measurements adopted the same methods and standards, which substantially reduced the measurement errors and improved the results' reliability.
The strengths should be interpreted with the limitations. Due to this study's cross-sectional nature, we cannot examine the temporality of the associations between sleep duration and FG, TG/HDL-C, and BMI.
Meanwhile, sleep duration and sleep quality were self-reported and subject to social desirability bias and recall bias. Future research should implement objective measurements to obtain more accurate sleep durations. However, a study showed good correlations between actigraphy and sleep logs when measuring sleep duration, and the self-reported data were reliable [41]. It is not feasible to obtain sleep duration through actigraphy due to the large sample size of this study. Therefore, we believe self-reported data collection was the most appropriate approach.

Conclusion
There is an independent U-shaped association between night sleep duration and FG with the optimal sleep duration 6.9 hours, indicating that both short and long sleep are associated with elevated blood glucose levels. The inverse linear association of sleep duration and BMI indicates short sleep is associated with elevated BMI while long sleep is associated with decreased BMI. In addition, the Ushaped relationship between sleep duration and TG/HDL-C is mediated by BMI. It is suggested that sleep affects blood glucose and insulin resistance through various mechanisms, and one of them may be through BMI. Prospective research is warranted to better delineate the relationship between night sleep duration and glycemia. Our ndings suggest that optimal sleep duration is important for maintaining glucose homeostasis, individuals with impaired FG or undiagnosed T2D may require more clinical attention targeting sleep duration to decrease the conversion from prediabetes to diabetes or delay the progression of diabetes.
Abbreviations Figure 1 Flowchart of study population selection.

Figure 2
Mean values and 95% con dence intervals for fasting glucose, TG/HDL-C and BMI levels according to self-reported night sleep duration. TG/HDL-C is expressed as the geometric mean. * Multiple comparison results of ANOVA P 0.05. (a) Fasting glucose concentrations in the night sleep duration 7 h and ≥9 h group were signi cantly higher than those in the control group with night sleep duration of 7 to < 8 h (P=0.018, P=0.008). (b) TG/HDL-C in the night sleep duration 7 h and 8 to < 9 h group were signi cantly higher than that of the control group with night sleep duration of 7 to < 8 h (P=0.001, P =0.002). (c) There were signi cant differences in Body mass index between the sleep duration 7h group and the 7 to <8h, 8 to <9h and ≥9h groups (P=0.001, P 0.001, P 0.001). There were signi cant differences in Body mass index between 7 to <8h group and 8 to <9h group and ≥9h group (P=0.001, P 0.001). There was no signi cant difference in Body mass index between 8 to <9h group and ≥9h group (P=0.834).