Children were recruited for Growing up in Australia: the Longitudinal Study of Australian Children (LSAC) during 2004, aged 4-5 years (Wave 1, N = 4983). The sampling frame was extracted from the enrolment database of Medicare, the national health care scheme in Australia that enrolls 98% of Australian residents by 1 year of age [11]. The children were selected using a two-stage design [11]. In the first stage, postcodes were stratified by state/territory and urban/rural location to ensure geographical representation. Very remote postcodes were excluded. Postcodes were then randomly selected. In the second stage, a 10% sample of children born between March 1999 and February 2000 were randomly selected. For each postcode, children were listed according to date of birth, and a systematic random sample was taken from this list to ensure a representative range of birth dates.
The response rate at baseline was 54% (N = 4983). Data are collected biennially, with a core face-to-face home visit supplemented by a range of additional measures and data linkages. Data used in this analysis were collected via three mechanisms: home-based face-to-face interview and two subsequent written questionnaires completed by a parent/caregiver at wave 3 (2008); questionnaires completed by teachers at wave 3 (2008) and wave 4 (2010); and linked data from the Australian National Assessment Program - Literacy And Numeracy (NAPLAN), the national standardized school assessment program (Additional file 1: Figure S1). The children were 8-9 years old at wave 3 (participation rate 87% of wave 1) and 10-11 years at wave 4 (participation rate 84% of wave 1, Fig. 1).
The analysis of confidentialized LSAC data had the approval of the Australian Institute of Family Studies ethic committee. Consent to participate in the study was given by a parent at wave 1 and permission to link with the child’s NAPLAN results at wave 3 (or wave 4, if the child did not participate in wave 3).
Breakfast (wave 3)
Breakfast consumption was assessed on three occasions, within 4 weeks, at wave 3 (baseline for this analysis). At the face-to-face interview, the parent/caregiver was asked “Did <study child> eat breakfast today?” After the interview, parents were asked to complete two time-use diaries on specified days the following week (one weekday and one weekend day), to which was appended a set of short dietary questions including whether the child had eaten breakfast that day. If the diary was not completed on the allocated date, the parent was asked to wait until the same day the following week. Parents reported whether the child was ill the day the diary was completed. The diaries were collected by the interviewer in person or returned by post. No data on breakfast consumption were collected at wave 4.
Children who skipped breakfast on at least one of the 3 days were classified as breakfast skippers and compared with those who ate breakfast on all three occasions (non-skippers). Further categorization was not possible because so few children skipped breakfast more than once (n = 17).
Academic performance (wave 4 teacher report, year 5 academic testing)
A questionnaire was sent to the child’s teacher as soon as feasible after the wave 4 home interview [12], and on average, was completed 2 months after the interview. Questionnaires were completed for 3269 children (response rate 75.6%). The teacher was asked to compare the child’s reading, mathematics and overall progress to other children of the same level, on a 5-point scale. Response options were collapsed into three categories for the analysis: “far below/below average”, “average”, “above/far above average”.
NAPLAN assesses all Australian students in Years 3, 5, 7 and 9 (aged 8-9, 10-11, 12-13, and 14-15 years, respectively) across four domains: reading, writing, language conventions (spelling, grammar and punctuation), and numeracy, using national tests held on the same day in May across Australia each year. Scores are standardized to range from 0 to 1000 for each test, enabling comparisons within and across school year levels [13]. NAPLAN data for 4159 children were linked to the LSAC dataset (98.4% of those who consented to NAPLAN access, 83.5% of the total sample at wave 1). Of those who were not matched, 552 were not asked for consent as they did not participate in wave 3 or wave 4, 117 forms were completed incorrectly, 68 could not be linked and 48 refused permission [13]. Because the children did not all begin school in the same year, the Year 5 NAPLAN tests were completed in 2009, 2010 or 2011. For children who repeated a grade and sat the same NAPLAN test more than once, the latest score was used. NAPLAN data were available for 2158 children included in this analysis.
Classroom behavior (wave 4)
The child’s teacher was asked to complete the Strengths and Difficulties Questionnaire (SDQ, Robert Goodman 1999, UK) in relation to the child’s classroom behavior. The SDQ includes 25 attributes, with the response options ‘not true’, ‘somewhat true’ and ‘certainly true’. The three scales recommended for low-risk or general population samples were used: internalizing problems (emotional + peer symptoms, 10 items, range 0-20), externalizing problems (conduct + hyperactivity symptoms, 10 items, range 0-20) and prosocial behavior (5 items, range 0-10) [14]. Lower scores for internalizing and externalizing problems and a higher score for prosocial behavior indicate better behavior. Children missing any of the subscale scores were excluded from the analysis.
Covariates (wave 3)
Covariates considered for inclusion in the adjusted models included sociodemographic variables associated with skipping breakfast (described below) and the other outcome variables (teacher-reported performance, classroom behavior, and the standardized tests; for example the behavior variables were considered as covariates in the analysis examining the association between skipping breakfast and academic performance). The child’s age in months was recorded at the face-to-face interview and the NAPLAN tests. A continuous variable for SES was created from standardized scores for three variables: parents’ years of education; parents’ occupation as determined by the status of their main occupation; and combined annual income including pensions and allowances before tax (with natural log transformation) [15]. The primary caregiver reported their own health (excellent, very good, good, fair, poor); whether they currently smoked cigarettes (yes/no), whether the child had two parents living at home (yes/no) and the child’s ethnicity. Financial hardship was assessed using the Family Hardship Scale, which sums the number of positive responses to seven indicators of household hardship. To calculate the Family Hardship Scales, parent/caregivers were asked to report if they had experienced any of the following situations in the previous 12 months because they were short of money: inability to pay bills on time; unable to pay mortgage or rent on time; went without meals; were unable to heat or cool the home; pawned or sold something because needed cash; sought assistance from welfare or community organization; unable to send child to kindergarten/preschool/child care for as much time as they would like (potential score 0-7) [16]. Ethnicity was considered as a potential confounder in the analysis, but was not included in any of the final models as it did not change the coefficient of breakfast skipping by at least 10% (our criterion for including a potential confounder [17]) when included in the model. Year 3 NAPLAN test results were used in the propensity model.
Statistical analysis
Log-link ordinal regression [18] was used to compare the probability of being in a lower level of teacher-reported scholastic performance for skippers and non-skippers. Results are reported for the continuation ratio probability model because for this model a test of the constraint underlying the ordinal assumption was satisfied in most cases. Differences between skippers and non-skippers in mean scores on the behavior subscales and NAPLAN results were estimated using linear regression. Model 1 included adjustment for age and sex. The teacher reported outcomes were adjusted for age at the time of the face-to-face interview, and the NAPLAN results were adjusted for age at the time the child sat the test. Model 2 had additional adjustment for SES, which was associated with both skipping breakfast and poorer academic performance. The other covariates included in model 3 were those that changed the coefficient of the covariate for skipping breakfast by at least 10% [17].
Inverse propensity weighting was used to take into account data missing from the baseline sample. The propensity model included baseline variables associated with missingness: SES; the primary caregiver’s sex, self-reported health status, education level and smoking status; household income; two-parent home; financial hardship; change in primary caregiver, child academic achievement (teacher-report and Year 3 NAPLAN) and child school attendance at wave 3. To ensure a full set of weights, missing observations in the variables required by each propensity model were imputed using multiple imputations by chained equations [19]. Ten imputations were performed.
A sensitivity analysis was conducted using only two measures of breakfast (the parent interview and one time-use diary) to allow children who were missing one diary to be included in the analysis. For children who had both diaries, one diary was randomly selected. We also examined cross-sectional associations between skipping breakfast and the wave 3 teacher-reported outcomes (collected the same way as in wave 4), but not the national standardized tests because most children sat the tests (May 2008) before the breakfast data were collected (March-December 2008, Additional file 1: Figure S1).
All analyses were conducted using Stata SE (version 12.1, 2011, StataCorp, College Station, TX). P-values ≤0.05 were considered statistically significant.