Participants
The participants, recruited in this cross-sectional study through flyers and bulletin board poster, were 44 healthy Japanese university students (24 males and 20 females), aged 20–24 years. The sample size required for the multivariate regression analyses was calculated using G* Power software [18]. The calculation, based on an assumption of 2 independent variables, a priori large effect size (0.35) at an α error level of 0.05, and a power (1 − β error probability) of 0.90, resulted in a total sample size of 40. None of the participants took any medications for more than a month before the 7-day recording of wrist activity. Inclusion criteria for participation in the study were: free of unusual events during the 7-day recording period, no sleep disturbance symptoms, and no hospitalization in the past year for eating disorders or serious medical illness. Each participant gave his/her written informed consent to participate, after having been provided with the full description of the test protocol, aim, and prospective effect on their health. All the study procedures were reviewed and approved by the Institutional Review Board of Toyo University (Approval No. TU2016-001).
Study protocol
All participants completed a self-administered questionnaire during the recording period that included demographic characteristics, morningness-eveningness preference, and habitual food intakes. The questionnaires submitted by the participants were checked for missing data and inconsistencies by a well-trained dietitian. Wrist activity was continuously measured using Actigraph (AMI Ltd., Ardsley, NY), over the 7-day period, including weekdays and weekends under the free-living condition. The Actigraph recording, which began on Wednesday 21:00, was completed on the next Wednesday 21:00. Each participant was asked to wear Actigraph on the wrist of the non-dominant arm and not to remove the device during the recording period except for when taking a bath. To confirm their routine sleep-wake cycle in the free-living condition, we also asked the participants to record their self-rated bedtimes and wake-up times during the recording period in a sleep diary.
Data processing
Actigraph was programmed to continuously collect activity data in a minute epoch with the zero crossing method, which counted the number of times of zero level crossing in the synthetic acceleration signal during each epoch [19]. When the participants had a non-daily routine during the recording period or had missed the recording because they forgot to wear the device or due to device failure, the participants’ data were excluded from the statistical analysis.
Cole’s algorithm [20] was used to score the sleep or wake in each 1-min epoch over the 7-day activity data. To identify the nocturnal sleep duration per day, bedtimes and wake-up times given by the participants in their daily sleep diaries were used to set the interval for calculating the total minutes for sleep epoch per night. Sleep durations on weekdays and weekends were calculated as the mean sleep durations from Sunday to Thursday nights and for Friday and Saturday nights, respectively.
To estimate the nocturnal sleep timing per night, the sleep/wake (0/1) scored data over the 7-day period were firstly processed using a band-pass filtering from 0.9 to 1.1 Hz for a 24-hour period. Then, the filtered signals were applied to the Hilbert transform to obtain the phase angles of the filtered signal. The timing of sleep (Figure 1) was estimated using the times at ±pi radians in the transformed data. The sleep timing on weekend nights was calculated as the mean of sleep timing on Friday and Saturday nights, while the sleep timing on weekday nights during the 7-day period was calculated as the mean sleep timing on Sunday, Monday, Tuesday, Wednesday, and Thursday nights. The relative sleep timing on weekday nights during the 7-day period was calculated by subtracting the sleep timing on weekends from the sleep timing on weekdays. A negative value means that the sleep timing on weekday nights is earlier than that on weekend nights.
Self-administered questionnaire
The questionnaire included demographic characteristics of the participants: age, height, weight, residential status, drinking habits, and smoking habits. Drinking and smoking habits were evaluated on the basis of the National Health and Nutrition Survey definitions [21]: drinking of more than 22 g of alcohol per day for more than three times per week, and currently smoking and having smoked for more than 6 months, or currently smoking and having smoked a total of 100 cigarettes, respectively. Body mass index (BMI) was calculated on the basis of the self-reported height and weight [weight/height2 (kg/m2)]. The Japanese version of the Horne-Östberg Morningness-Eveningness Questionnaire (MEQ) [22] was used to measure the self-rated diurnal preference, in which lower values indicate greater eveningness. Habitual dietary intakes, except for dietary supplement over the past month were evaluated using a self-reported dietary history questionnaire (DHQ), for which the validity and reproducibility have been reported in a previous study already [23]. Intakes of total energy, protein, fat, carbohydrate, and each food groups were assessed. To examine dietary compositions, each food group intake was adjusted by the total energy intake and sex using the residual method [24-26].
The score for adherence to the Japanese Food Guide Spinning Top (food guide score) was calculated from the results of the DHQ to assess diet quality [27]. Details of the score has been previously shown [27]. In brief, the scores (ranging from 0–10 points) for the servings of 7 categories of the Japanese Food Guide Spinning Top, such as grain dishes, vegetable dishes, fish and meat dishes, milk and milk products, fruits, energy intake from confectioneries, and total energy intake were calculated based on the reference amounts. Finally, the scores of all categories were summed to provide the food guide score (ranging from 0–70 points). A higher score represents a greater adherence to the dietary recommendations of the Japanese dietary guidelines.
Statistical analysis
Of the 44 participants who undertook the 7-day recording, 8 that did not sleep for more than one night for any reasons (e.g., participating in non-habitual events during the nights) during the recording period were excluded from the statistical analyses. Furthermore, a participant with a BMI outlier (43.0 kg/m2), which was detected by the Smirnov-Grubbs test at a significance level of 0.05, was excluded. Therefore, 35 participants (21 males and 14 females) were included in our statistical analyses. Pearson’s correlation coefficients were used to assess the associations of the sleep timing on weekends and the relative sleep timing on weekdays with demographic characteristics and habitual dietary intakes, and the associations of the sleep durations on weekdays and weekends with habitual dietary intakes. We dichotomized the participants based on the sleep timing on weekends at its median value into early (n = 17) and late (n = 18) sleepers, to determine the association between the sleep timing on weekends and the changes in sleep timing over the recording period. Linear mixed-effect model was used to analyze the effects of the group of weekend sleep timing (dummy: early sleepers = 0, late sleepers = 1), the day of the week (dummy: Wednesday nights = 0, Thursday/Friday/Saturday/Sunday/Monday/Tuesday nights = 1), and the interaction terms between them on the daily sleep timings. Multivariate linear regressions were performed to explore the independent associations of the sleep timing on weekends and the relative sleep timing on weekdays with habitual food group intakes. Residential status was included as a covariate in the multivariate linear regression analysis (dummy: living alone = 0, other = 1). To confirm the effects of the excluded participants, we used unpaired t-tests and χ2 tests for continuous and categorical variables, respectively, to examine the differences in characteristics between the excluded participants and those included in the main analysis. All statistical analyses were performed using Stata 16 (Stata Corporation, College Station, TX, USA). P values less than 0.05 were considered statistically significant using two-tailed tests.