A total of 500 individuals aged 18 years and older who could understand and read Chinese were recruited. We excluded those who were deaf, needed hearing aids, had psychiatric illnesses, took pills or other medical treatments for sleep, were pregnant, had children under 2 years of age, were unwilling or unable to wear an ActiGraph GT9X Link, or were unwilling or unable to assess nocturnal noise exposure.
Sample size calculation was first based on the assessment of an estimated 30 factors associated with nonrestorative sleep or noise sensitivity and, using the usual rule of thumb of 10 participants per independent variable, a minimum of 300 participants were required. Second, to assess the moderating effect of noise sensitivity, we considered a conservative error of 0.1 on the standardized interaction effect, which, using a 95% confidence interval (CI), indicated a minimum sample size of 396. Hence, we planned to recruit 500 participants after considering the possibility of participants drops out. During recruitment, we approached 1625 individuals before 500 individuals consented to participate in this study. We excluded 584 individuals due to ineligibility. Moreover, 227 declined to participate and 314 were not at home more than 5 times during household visits.
Sampling design and survey methodology
This was a population-based cross-sectional household survey conducted in Hong Kong from February 2018 to September 2019. The household sampling began by first obtaining a sample list based on the frame of quarters maintained by the Hong Kong Census and Statistics Department, which is the most complete, up-to-date, and authoritative sampling frame available in Hong Kong. Records of household addresses were organized in quarters and stratified by geographical districts and types of quarters. A systematic sampling design with fixed sampling intervals and non-repetitive random numbers was then applied to obtain a random sample of quarters. For the quarters selected, all households residing in the quarters were included in the survey.
Before household visits took place, notification letters outlying the study details, planned visit times, and interviewer identities were mailed to all targeted households. During the household visits, the eligibility criteria were assessed and written informed consent was obtained before taking study measurements.
After providing consent, a research assistant calibrated a noise dosimeter (Spark 706RC, Larson Davis Inc., US) or a sound level meter (NSRT Mk2, Convergence Instruments, Canada) using a CAL150 (Larson Davis Inc., US) at 94dBSPL and helped to identify an appropriate location for positioning the device. Each participant was then shown how to wear an ActiGraph GT9X Link (ActiGraph, US) strapped securely to the non-dominant wrist for a week and completed a battery of self-report questionnaires. After a week, another household visit was made to collect the devices and to distribute shopping coupons valued at HK$300 (around US$40) to each participant.
Ethics approval of this study was obtained from the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (Ref no.: UW17–011).
Nocturnal noise exposure
Spark 706RC or the NSRT Mk2, both meeting American National Standards Institute (ANSI) S1.4, was used to record sound intensity levels for seven consecutive days to indicate nocturnal noise exposure . It was placed on a stable surface within 2 m from where a participant slept and close to participant ear level during sleep and recorded A-weighted energy equivalence levels, set at 1-min intervals. Nocturnal noise exposure was calculated as the average equivalent continuous sound pressure level from 00:00 to 8:00 (LAeq, 8h).
Objective physiological sleep parameters, including sleep latency, sleep efficiency, time in bed (TIB), total sleep time (TST), wake after sleep onset (WASO), and awakenings and average awakenings, were measured by ActiGraph GT9X Link, and calculated by ActiLife (ActiGraph, US) using the Cole-Kriple algorithm, which has been validated among adults [38, 39]. The results of actigraphy have been demonstrated to be consistent with those of polysomnography . Participants were required to wear for seven consecutive days, and a valid record required recording from at least four weekdays and one weekend day. Data obtained from the ActiGraph GT9X Link were corroborated by a sleep diary where participants recorded the time they went to bed and the time they woke up every day. Sleep latency was the duration between time in bed and the first minute that the algorithm scored “asleep” [38, 39]. TIB was calculated as the duration between time in bed and time out of bed, and TST refers to the total duration scored as “asleep” . Sleep efficiency was the TST divided by the TIB . WASO was the total time that the participants were awake after sleep onset and awakenings were the number of awakening episodes during the night, while average awakenings refers to the average duration of all awakening episodes .
Nonrestorative sleep scale (NRSS)
The Chinese version of the NRSS which includes four domains, namely, refreshment from sleep (e.g., Usually, do you think your sleep is restoring or refreshing?), the physical/medical symptoms of nonrestorative sleep (e.g., Do you feel that physical or medical problems are dragging you down), daytime functioning (e.g., What is your usual level of daytime energy?), and the affective symptoms of nonrestorative sleep (e.g., Do you feel depressed or down if you didn’t sleep well the night before?), has been demonstrated to be a valid and reliable instrument for measuring nonrestorative sleep . The coefficient omega of the global score of the Chinese scale was 0.92 . NRSS scores were standardized on a 0–100 scale, with higher scores indicating less nonrestorative sleep (i.e., better restorative sleep).
Weinstein noise sensitivity scale (WNSS)
The WNSS was developed to assess noise sensitivity. The traditional 18-item Chinese version of WNSS has been verified to be a reliable and valid scale to assess noise sensitivity . Each item (e.g., I am easily awakened by noise) was rated on a 6-point Likert scale with a total score ranging from 18 to 108. This score was then standardized to a 0–100 scale. A higher total score indicates more sensitivity toward noise. The Chinese WNSS had a Cronbach’s alpha of 0.83 .
ENRICHD social support instrument (ESSI)
Social support was assessed via the ESSI. The Chinese version of ESSI consists of 6 items (e.g., Is there someone available to help with daily chores?), with each item rated on a 1–5 scale. The global score ranges from 6 to 30, with higher scores indicating a higher level of social support . The internal consistency of the Chinese version scale is satisfactory (Cronbach’s alpha = 0.79) .
Patient health questionnaire (PHQ-15)
Psychosomatic symptoms were assessed using the Chinese version of the PHQ-15, which has been validated in a Chinese population . It addresses 15 somatic symptoms (e.g., stomach pain), each assigned a score ranging from 0 (not bothered at all) to 2 (bothered a lot). The items cover the most prevalent DSM-IV somatization disorder somatic symptoms. The Chinese PHQ-15 had a Cronbach’s alpha of 0.79 .
Perceived stress scale (PSS)
The Chinese 10-item PSS was adopted to assess the level of stress in participants. The 10-item version has been demonstrated to have better validity and reliability than the 14-item and 4-item scales. Six of the ten items were positively worded (e.g., How often have you been upset because of something that happened unexpectedly”) and the other four items were negatively worded (e.g., How often have you felt confident about your ability to handle your personal problems?). Positive items were reversed-scored before computing the total score. Higher total scores indicate higher stress levels . The Cronbach’s alpha of the Chinese PSS was reported as 0.85 .
Hospital anxiety and depression scale (HADS)
The Chinese version of the HADS comprises 14 items, seven of which measure anxiety (e.g., I feel tense or wound up) and seven measure depression (e.g., I still enjoy the things I used to enjoy). The subscales were scored independently, with each subscale score ranging between 0 and 21, with higher total scores indicating more severe anxiety/depressive symptoms . Reliability, tested by Cronbach’s alpha, was 0.80 and 0.63 for the anxiety and depression subscales, respectively .
The Chinese version of STOP-BANG questionnaire, which was modified from the STOP questionnaire, has been validated in Chinese adults in Hong Kong with high sensitivity, and is an appropriate screening instrument for obstructive sleep apnea (OSA) . This scale addresses risk factors such as “BMI > 30kg/m2” and “Age > 50 years old”. Individuals are categorized as having a high risk of OSA when they score “Yes” on three or more items. It is a satisfactory questionnaire for OSA screening with a sensitivity up to 86% .
Sociodemographic and lifestyle characteristics
The sociodemographic and lifestyle data were self-reported in front of an interviewer. An investigator-developed self-reported information sheet which aimed to collect sociodemographic and lifestyle data from participants included age, sex (Male; Female), marital status (Single; Married/Cohabiting; Separated/Divorced/Widowed), education level (Primary school or below; Secondary; Bachelor or above), occupation (Working; Not working; Retired; Students), family income (from < HK$5000 to > HK$50000), exercise (from no to 5 h or more per week), smoking (Never; Quit; Yes), as well as consumption of alcohol (Never; Quit; Yes), cola, soda, coffee, and tea (Every day; Every week; Every month; Every year; Never). Season of conduction was summarized according to the recording time (Spring: Month 3–5; Summer: Month 6–8; Autumn: Month 9–11; Winter: Month 12–2). If the recording was conducted across different seasons, it was categorized according to the season with records for 4 days or more that were retained.
First, we regressed nonrestorative sleep on each independent variable in a bivariate regression model. Then, a structured multiphase regression analysis was conducted to assess the association of nocturnal noise, noise sensitivity, and other variables with nonrestorative sleep and physiological sleep parameters. The structured multiphase regression model accounts for the possible causal relationship among the variables by first grouping them into sequential clusters . Specifically, we defined five clusters such that variables in Cluster 1 could affect variables in Clusters 2, 3, 4 and 5, but not vice versa. Similarly, Cluster 2 variables may affect variables in Cluster 3, 4 and 5, but not vice versa, and so on. Cluster 1 included sociodemographic and time variables, i.e., age, sex, marital status, educational level, occupation, family income and season. Cluster 2 included lifestyle and living environment factors, i.e., exercise, smoking, alcohol, cola, soda, coffee, tea, social support, and nocturnal noise level. Cluster 3 included somatic symptoms, stress, anxiety, and depression levels. Cluster 4 included noise sensitivity. The physiological sleep parameters were included in Cluster 5. The structured multiphase regression analysis was then performed in five phases (i.e., enter regression). In Phase 1, a regression was conducted on all variables in Cluster 1. In Phase 2, a regression was conducted on all Cluster 2 variables adjusting for variables in Cluster 1. In Phase 3, variables in both Clusters 1 and 2 were considered as potential confounders for variables in Cluster 3, and so on. The effect of a variable in Cluster 1 was taken as that estimated in Phase 1, while the effect of a variable in Cluster 2 was taken as that estimated in Phase 2, and so on. Therefore, variables were not adjusted for the variables in the next cluster, which theoretically avoided potential over adjustment.
Moreover, all physiological sleep parameters were regressed on Clusters 1–4 separately. Furthermore, an interaction between noise sensitivity and nocturnal noise was also assessed based on each corresponding Phase 4 model.
All collected data were entered into SPSS version 23 (Armonk, NY: IBM Corp.) and cleaned by cross-checking with the original paper records. The significance level was set at 5%, and all estimates are accompanied by 95% CIs where appropriate. Regression was conducted with RStudio-1.2.1335. The adequacy of the regression models was assessed by examining the studentized residuals. The presence of multicollinearity was assessed by the variance inflation factor (VIF) with the “car” package under RStudio-1.2.1335 . The VIF quantifies how much the variance is inflated by the existence of correlation among the independent variables in the model. A value of 1 indicated no correlation between independent variables while 1 < VIF ≤ 5 and > 5 indicated moderate correlation and high correlation, respectively .