Sample and study design
This study is a secondary analysis of data drawn from the nationally representative, cross-sectional 2011–2012 Australian National Nutrition and Physical Activity Survey (NNPAS 2011–12) and was registered at anzctr.org.au as ACTRN12617001029381. The NNPAS, conducted between May 2011 and June 2012, was administered by the Australian Bureau of Statistics (ABS); full details of the study design and methods have been described elsewhere . Briefly, 12,153 participants aged 2 or more years (including 9338 adults, aged ≥ 19 years; 77% response rate) were selected using a multistage, probability sampling design of private dwellings (Fig. 1). The Australian Government Census and Statistics Act 1905 provided the ABS with ethics approval to conduct the survey components of the NNPAS .
All BP measurements were voluntary, excluded pregnant women, and were taken by trained ABS staff during the home visit. Participants were asked to sit comfortably and relax their left arm. Two SBP and DBP measurements were taken on the left arm, with the palm facing upwards, using an automated BP monitor. The second reading was used, except where there was greater than 10 mmHg difference between the first and second readings, in which case, a third reading was taken and the second and third readings averaged. If all three readings differed from each other by > 20 mmHg, then these readings were considered invalid. Participants were classified as hypertensive (≥ 140/90 mmHg) or non-hypertensive (< 140/90 mmHg) . No data about use of BP medications were collected; however, participants’ self-reported whether they had current or previous hypertensive disease.
Dietary data were collected over two 24-h recalls, conducted approximately 9 days apart, on a different day of the week. During each dietary recall, based on the validated USDA automated multiple 5-pass method, participants were asked to identify the type of EO (e.g., breakfast, lunch, dinner, or snack) and the time when each EO commenced. The Australian Supplement and Nutrient Database 2011–2013 was used to determine energy and nutrient intakes from all foods and beverages and dietary information was averaged across the two recall days to obtain mean estimates of eating patterns, total energy intakes, and food intakes.
Frequencies of all EO, meals, and snacks
The methods used to calculate mean total frequencies of all EO, meals, and snacks have been described previously [18, 27]. Briefly, an EO constituted any eating event that provided a minimum energy content of 210 kJ (50 kcal) and was separated in time from the surrounding EO by 15 min. This approach was informed by the previous research and current recommendations for defining EO in eating patterns research [4, 27]. EO were then classified as meals and snacks according to participants’ self-report of EO. Meals included EO reported as breakfast, brunch, lunch, dinner, and supper, whereas snacks included EO reported as morning/afternoon tea and beverage break were classified as snack EO. Based on the sample distribution, frequencies of all EO, meals and snacks were divided into categories of 1–3, 4–5, or ≥ 6 EO, 1–2, 3, or > 3 meals, and 0–1, 2–3, or > 3 snacks.
Temporal eating patterns
Temporal eating patterns were determined using latent class analysis, as reported in detail previously [23, 28]. Briefly, three latent classes of temporal eating patterns were identified for men and women based on frequency and timing of EO across the day and labelled according to their defining characteristics. The optimal number of classes was selected based on model fit indices, likelihood ratio tests comparing k with k−1 class models, and pattern interpretability . The first pattern was labelled as “conventional” due to the probability of participants having three EO at “conventional” mealtimes in Australia (e.g., between 7 and 8 a.m., 12–1 p.m., and 6–7 p.m.). The second pattern was distinguished by a > 0.9 probability of a “lunch” meal approximately 1 h later than the “conventional pattern” and labelled the “later lunch” pattern. The probability of a “dinner” meal 1 h later than the conventional pattern was also higher (e.g., > 0.6) in participants with a “later lunch” pattern. The third pattern was labelled the “Grazing” pattern, because it was characterized by frequent but less distinct meal occasions and a higher probability of having an EO after 8 p.m. These temporal eating patterns have been associated with socio-demographic and eating pattern characteristics , diet quality, and, among women, adiposity outcomes .
The following socio-demographic, health behaviours, and anthropometric variables, collected during the household survey, were considered as potential covariates due to their previous reported relation with hypertension risk [1, 26].
Education level was categorised as: low (completed some high-school or less), medium (completed high-school or completed some high-school and/or certificate/diploma) or high (having a tertiary qualification). Country of birth was categorised by the ABS as: Australia, predominantly English-speaking countries (other than Australia) and all other countries.
Smoking status was self-reported and categorised as current smoker, ex-smoker, and never smoked. Participants were categorised as meeting or not meeting current Australian physical activity guidelines (150 min and 5 sessions), based on self-reported frequency and duration of walking for recreation or transport and moderate or vigorous leisure-time physical activity [30, 31]. Self-reported information on sleep duration the night before the survey and how much time participants spent sitting or lying down at work, during transport and leisure activities in the past week, were used to calculate (per day) total minutes spent sleeping per day and in sedentary behaviour, respectively. Participants reported whether they were currently on a weight-loss diet for health reasons (yes/no). Average daily total energy intake and diet quality scores were calculated from the 2 days of recall. The established food-based Dietary Guidelines Index (DGI) was used as a measure of overall diet quality [32,33,34]. The DGI assesses compliance with recommendations outlined in the Australian Dietary Guidelines, and is the sum of 13 components (score range of 0–130), each corresponding to an Australian Dietary guideline and scored proportionally out of 10. The components include meeting recommendations for food variety, intakes of fruits, vegetables (including legumes), grain foods, dairy and alternatives, meat and alternatives, unsaturated fat, fluids, discretionary foods, saturated fat, salt, added sugar, and alcohol. Higher scores indicate a better diet quality. Measurement of height (cm) and weight (kg) were taken to one decimal point by trained ABS staff using a portable stadiometer and digital scales. BMI (weight [kg]/height [m]2) was calculated.
The analytic sample included the 65% of adult participants who completed both dietary recalls (n = 6053; Fig. 1). Participants were eligible for this analysis if they were not pregnant, breastfeeding, or undertaking shift-work in the past 4 weeks (n = 5366) and were excluded if they reported no energy intake during either dietary recall (n = 8 excluded) or did not report the time at which an EO commenced or the type of EO (n = 116 excluded). Of the remaining 5242 participants, 578 (11%) had missing data for BP and a further 182 (3.9%) were missing data for BMI and covariates: BMI (n = 149), physical activity (n = 28), and sedentary time (n = 5). The final analytic sample was 2099 men and 2383 women.
All statistical analyses were stratified by sex and used Stata statistical software, Version 14.2 (Stata Inc., College Station, TX, USA). Point estimates and standard errors were determined by applying person and replicate weights that accounted for the probability of participant selection and the clustered survey design, respectively. Descriptive statistics for sample characteristics are presented as weighted means (95% CI) or weighted percentages. After examining the distribution of the data, the following variables were log-transformed to improve normality: BMI, daily total sedentary time, and total energy intake. Weighted geometric means (95% CI) were used for all log-transformed variables.
The F test (for continuous data) and adjusted Pearson χ2 test (for categorical data) were used to determine sex-specific differences in sample characteristics by hypertension status. Multiple linear regression (for continuous outcomes) and logistic regression (for binary outcomes) were used to test for associations of frequencies of all EO, meal and snacks (continuous), and temporal eating patterns, with SBP and DBP (continuous) and hypertension prevalence (binary). Four models were tested: model 1 was an unadjusted model; model 2 adjusted for age (years, continuous), education level (low, medium or high), country of birth (Australia, other predominantly English-speaking countries, all other countries), smoking status (never, former or current), daily, meeting physical activity guidelines (yes/no), daily sedentary time (min; continuous), sleep duration (h, continuous), dieting for health reasons (yes/no); model 3 further adjusted for BMI, and model 4 further adjusted for total energy intake and DGI scores (both continuous). In light of the previous research that reported a positive association among participants with the highest EO frequencies (i.e., > 5 EO) , in the present study, any observed statistically significant (P < 0.05) adjusted association between the continuous measures of frequencies of EO, meals, or snacks, and the outcome variables were further explored by examining associations for eating pattern frequency categories (e.g., 1–3 [reference], 4–5 or ≥ 6 EO; 1–2 [reference], 3 or > 3 meals and 0–1 [reference], and 2–3 or > 3 snacks). Finally, the effect of energy misreporting, defined as the ratio of total energy intake to total energy expenditure was considered ; however, its inclusion did not improve its predictive power when BMI was already in the model. A previous study has also shown that energy misreporting bias can be statistically corrected using predictors of energy misreporting (i.e., dieting behaviours and BMI) .
As data on BP medications use were not collected in the NNPAS , a sensitivity analysis was conducted that included only participants who self-reported no current or previous hypertensive disease (n = 1612 men and n = 1853 women).