Study population
The participants of the current study were included from the Henan Rural Cohort, which has been previously described in detail [17, 18]. Briefly, a multistage cluster sampling method was utilized to select samples from permanent residents. Target population aged 18–79 years was recruited in Suiping, Yuzhou, Yima, Tongxu and Xinxiang counties of Henan province from July 2015 to September 2017 and registered in Chinese Clinical Trial Register (Registration number: ChiCTR-OOC-15006699). Consequently, a total of 39259 adults (15490 men and 23769 women) were obtained in the baseline of Henan rural cohort.
For the current analysis, a total of 29 995 completed the evaluation of anxiety symptoms. Furthermore, participants were excluded if they had missing data on PSQI score (n = 269), self-reported experience of night shift work (n = 1530), or had a history of cancer (n = 285). The final samples included 27 911 subjects aged 18–79 years in the current study.
Ethics approval was provided by the Zhengzhou University Life Science Ethics Committee. Signed informed consent was obtained for each participant.
Covariates
Data collection was performed by well-trained investigators in a face-to-face interview using a structured questionnaire. Demographic variables of participants included gender, age (continuous variable), marital status (married/cohabitation, other), educational levels (primary school or below, junior high school and senior high school or above), smoking status (non-smoker, or current smoker), alcohol consumption (non-drinker, or current drinker), high vegetables and fruits intake (defined as more than 500 g per day), and personal and family history of diseases.
High fat diet was determined according to reported intake of meat of live stocks and poultry of 75 g or above per day. Physical activity levels were classified into three categories; light, moderate and vigorous referenced to the criterion in the International Physical Activity Questionnaire [19]. Additionally, the physical measurement was conducted on the basis of a standard protocol [20]. Height and weight were measured with individuals wearing light clothes and barefoot to the nearest 0.1 kg and 0.1 cm. Body mass index (BMI) was computed by body weight in kilograms divided by square of height in meters.
Evaluation Of Sleep Quality
Information on sleep was collected by PSQI [21], which consisted of 19 items. The scale which scores 0 to 21 has been widely used to evaluate sleep quality and well validated and readily completed by most participants. A previous study reported that at least acuttoff score of 5 PSQI yields a sensitivity of 89.6% and a specificity of 86.5%[21]. Thus, a participant with a more than 5 PSQI score is considered as having a poor sleep quality in this study. Self-reported night sleep duration was obtained by asking the following question of the PSQI, “What time did you usually go to bed and wake up during the past month?” And the sleep onset latency was collected by the following question: “How long (in minutes) has it taken you to fall asleep each night during the past month?” The fall asleep time was calculated as bed time plus sleep latency. The night sleep duration was computed on the basis of wake-up time and fall asleep time[22].
Definition Of Anxiety Symptoms
The anxiety symptoms of participants were collected using the two-item generalized anxiety disorder scale (GAD-2)[23] which included two items (feeling nervous, anxious, or on edge and not being able to stop or control worrying) yielding a sensitivity of 85% [24]. The scores of this scale ranged from 0 to 6. At least a score of three was viewed as the occurrence of anxiety symptoms, which was also utilized in the current study [25].
Meta-analysis
Based on the results of previous studies and current studies, a meta-analysis was conducted on the relationship between poor sleep quality and anxiety symptoms. A systematic electronic literature search was conducted in PubMed, Web of Science, CNKI (China National Knowledge Infrastructure) and Wanfang databases: (sleep quality OR PSQI score) AND anxiety symptoms. The included studies had at least two groups (good sleep quality, poor sleep quality) or three groups (good sleep quality, intermediate sleep quality, poor sleep quality), and anxiety symptoms. The exclusion criteria were :(1) studies in aged 17 or younger, (2) editorials, (3) reviews, (4) studies no OR、RR、or HR effect. Information extracted from all relevant articles included title, first author name, year of publication, study design, sample size, the age range of participants, adjustment factors, sleep quality criteria, definition of anxiety symptoms, and OR (95% CI) for poor sleep quality. Data were extracted from the original literature, and ambiguity was resolved through discussion if there was any inconsistency.
Statistical analysis
Mean ± standard deviation (SD) and frequencies (percentages) was presented for continuous and categorical variables, respectively. Multivariable restricted cubic regression spline curves [28] with 3 knots (5th, 50th, and 95th) were fitted to observe the shape of the association between continuous PSQI score and anxiety symptoms. Furthermore, the PSQI score was dichotomized to examine the association between poor sleep quality (≥ 6) and anxiety symptoms with good sleep quality (< 6) as reference group by performing logistic regression models. In the fully adjusted model, age, gender (only in total population), high vegetables and fruits intake, high fat diet, physical activity, marital status, smoking status, drinking status, educational levels, average monthly income and BMI, night sleep duration and napping duration were included in the model as the underlying confounder.. Additionally stratified analyses were conducted by each potential modifier to examined whether poor sleep quality and anxiety symptoms were potentially changed by age, sex, marital status, smoking, drinking, income, physical activity, BMI, snoring, hypertension, ( T2MD) type 2 diabetes mellitus. Finally, a meta-analysis was conducted to validate the current result. The difference in effect size caused by heterogeneity in studies can be quantified by the I2 statistics [26]. The random effects model was used for substantial heterogeneity, and a fixed effects model was used for homogeneity (or low heterogeneity) (I 2 > 50%). Egger's test was used to evaluate publication bias in the process of meta-analysis. A two-tailed P value of less than 0.05 was determined the statistical significance in the current study. All analyses were run on SAS version 9.1 (SAS Institute) and R version 3.5.1.