This study used data from the Longitudinal Study of Australian Children (LSAC), which is a population-based study following the development of a nationally representative Australian sample of children and their families [20]. The LSAC commenced in 2004 and collects data every two years from two cohorts of children: the B (baby)-cohort (born March 2003–February 2004) and the K (kindergarten)-cohort (born March 1999–February 2000) [21]. Participants were sampled from the Medicare database, which is Australia’s universal healthcare programme and includes 98% of Australian children by the age of 12 months. Further details on the methodology, recruitment and response rates can be found online [21]. Families provided written informed consent to participate in the study, and the Australian Institute of Family Studies Ethics Committee approved each wave of the LSAC.

The present study includes data from waves 1 to 6 of the B-cohort and waves 1 to 4 of the K-cohort (ages 4/5 to 10/11 years), collected from 2004 to 2014 (Supplementary Table 1). A total of 5107 and 4983 children were recruited for the B- and K-cohort, of which 3764 and 4164 were retained at age 10/11 years, respectively [21]. Of the combined study sample with data at age 10/11 years (n = 7928), participants were excluded if parent 1 (i.e., the parent who knows the study child best and who completed the face-to-face interviews and questionnaires) was not the biological mother (n = 194) or if parent 2 (i.e., parent 1’s partner or another adult in the home with a parental relationship to the study child) was not the biological father (n = 686) to be able to account for any direct influences of parental characteristics (such as maternal hypertension in pregnancy, maternal and paternal BMI) on offspring blood pressure. Participants were also excluded if dietary intake data was not available at two or more time points (n = 6), if data on BP were missing at age 10/11 years (n = 831), or if data on relevant covariates were missing (n = 1851). This study therefore includes 4360 children. A comparison of characteristics of children included (n = 4360) and excluded (n = 5730) is shown in Supplementary Table 2.

LSAC’s data collection methods include face-to-face interviews with parents and children conducted by trained interviewers, and audio computer-assisted self-interviews completed by children.

Dietary intake was assessed at each wave. Parents (for children aged 4-9 years) or children (from age 10) were asked a standard set of 11 questions relating to the child’s consumption of individual or grouped food or drink items [22]. Questions assessed the frequency of consumption in the previous 24 h, with answers ranging from “Not at all”, “Once” and “More than once”.

Dietary scores were derived at each wave in line with the previously developed dietary scoring system by Gasser et al. [22]. In line with the 2013 Australian Dietary Guidelines [23], partcipants were assigned an individual score (ranging from 0 to 2) for each of seven food groups: fruit, vegetables, water, fatty foods, sugary foods, sweetened drinks, and milk products or alternatives. Vegetables, fruit, water and milk products were assigned higher scores for more frequent consumption, whereas fatty foods, sugary foods and sugary drinks were assigned higher scores for less frequent consumption. The continuous scores for each food group were summed to give an overall score between 0 (least healthy) and 14 (most healthy) for each participant at each wave [22].

The dietary questions, scoring system and dietary trajectories have not been formally validated, however, the dietary intake data collected in the LSAC have previously been used in other studies, including analysis of associations with longitudinal body composition measures [24], family socioeconomic position [25], parental health behaviours [26], parenting styles [27], and preclinical cardiovascular phenotypes [18]. Moreover, the dietary trajectories derived using the scoring system by Gasser et al. [22] have been used in multiple analyses based on LSAC dietary data [18, 24,25,26,27].

SBP and DBP measurements were taken by trained professionals in wave 6 of the B-cohort and wave 4 of the K-cohort when children were aged 10/11 using an A&D Digital Blood Pressure Monitor – Model UA-767 (A&D Co., LTD, Japan). This monitor has not been validated for use in children. Two measurements were taken by the interviewer with a one-minute interval between the measurements, according to standard protocol [28]. The mean of the two measurements was calculated and used for analysis.

We identified potential covariates for the analysis based on published literature that examined the relationship between dietary intake and BP in children [7,8,9,10, 12, 29]. Questions and response options for all covariates are described in Supplementary Table 3. Variables describing age, sex, country of birth, indigenous status, socio-economic status, diabetes and hypertension in pregnancy, and maternal education were obtained from interview data collected at wave 1.The child’s birthweight was recorded by the parents at wave 1 and transformed to birthweight z-scores using the US Centres for Disease Control (CDC) growth charts [30]. Mothers were asked about breastfeeding at wave 1 (B- and K-cohort) and wave 2 (B-cohort): “Was the child ever breastfed?” (yes or no), “How old was the child when he/she had any milk or food other than breast milk (months), and “How old was the child when he/she completely stopped being breastfed (including expressed breast milk)?” (months). Based on these questions, we examined breastfeeding as never or briefly (< 1 month), 1-3 months, 4-5 months, and ≥ 6 months. Maternal and paternal weight (kg) and height (m2) were self-reported at wave 1 and BMI was calculated. In addition to these socio-demographic and early life factors, potential covariates at age 10/11 were considered. This included the child’s pubertal status which was categorised as “has not yet or barely started”, “definitely started”, or “seems complete” based on a series of questions asked of the child’s mother on body hair growth, voice deepening (boys only), and breast growth and first menstruation (girls only). Child’s physical activity was categorised based on questions asked of the parents on the child’s choice of how they spend their free time (active or inactive). The child’s weight was measured using Tanita body fat scales (UM-051) (Tanita Australia) [31] and height using laser stadiometers. BMI was calculated and transformed to age- and sex-adjusted BMI z-scores [32], which were then used to categorise children as underweight, normal weight, overweight or obesity according to international BMI cut-offs [33]. Percentage body fat was measured using Tanita body fat scales (UM-051) [28, 31]. The fat mass index (FMI) was calculated as fat mass (kg) divided by the square of height (m2), to provide a height-adjusted measures of fat mass. These anthropometric measurements were taken by trained staff.

Group-based trajectory modelling was used to derive trajectories based on continuous dietary scores across waves 3–6 of the B-cohort and waves 1–4 of the K-cohort (ages 4/5 to 10/11 years) separately [34]. Overall dietary scores at each wave were normally distributed and used as continuous dependent variables, and the child’s age at each wave as independent variables. For each cohort, we compared models for three, four or five trajectories, and determined if an intercept, linear, quadratic or cubic model fitted the data best. Based on Bayesian Information Criterion (BIC) values [35], four trajectories were retained in both cohorts. This is in line with the previous analysis of dietary scores and trajectories in LSAC by Gasser et al. [22].

Participant characteristics were described across the four trajectories and compared using chi-squared tests for categorical variables and one-way analysis of variance (ANOVA) for continuous variables.

Linear regression models were used to derive coefficients (non-standardised) for the association between dietary trajectories and SBP and DBP at age 10/11. To identify relevant covariates, relationships of potential covariates with dietary trajectories and SBP and DBP were examined. Chi-squared tests were used to assess differences in categorical characteristics by dietary trajectories, and ANOVA was used to examine differences in continuous characteristics by dietary trajectories. Linear regression was used to determine associations between potential covariates and SBP and DBP. Based on these preliminary analyses, covariates that were significantly associated with dietary trajectories and with SBP and/or DBP were included in the linear regression models, as well as covariates commonly used in previously published studies examining the association between diet and BP [7,8,9,10, 12, 29]. Model 1 included socio-demographic covariates (child’s age, sex, indigenous status and maternal education, country of birth and socio-economic status). Model 2 additionally adjusted for parental covariates (maternal and paternal BMI and maternal hypertension in pregnancy) and model 3 additionally adjusted for child-related covariates (breastfeeding, pubertal status and physical activity). Model 4 included all covariates included in model 3 and BMI, and model 5 included all covariates included in model 3 and FMI. Covariates that were considered but not included were: maternal diabetes in pregnancy, maternal age at delivery, child birthweight, singleton or multiple birth, sleep, and diagnosis of diabetes in the child.

A cross-product term between dietary trajectories and child sex was included in the final model to test for differences in associations between boys and girls. A cross-product term was also included between dietary trajectories and cohort to test for differences in associations between the B- and K-cohort. All statistical analyses were conducted using STATA 16.0 software [36].