Relationship between raised BMI and sugar sweetened beverage and high fat food consumption among children


  • Disclosure: the authors have no competing interests.

  • Author contributions: LM developed the major concepts, LM & BR conceived the analysis plan and analyzed the data. All authors were involved in writing the article and had final approval of the submitted and published versions.



Longitudinal evidence of relationships between unhealthy diets and BMI in children is crucial for appropriately targeting obesity prevention activities. The objective was to determine the relationship between frequency of consumption of sugar sweetened beverages (SSBs) and high fat foods (HFFs) and body weight in Australian children aged from 4 to 10 years.


Data from 4,164 children participating in four waves (wave 1, 2004; wave 2, 2006; wave 3, 2008; and wave 4, 2010) of the Longitudinal Study of Australian Children were analyzed. A multi-level growth model tested relationships between consumption of SSB and HFF and BMI z-scores.


BMI z-scores were associated with daily consumption of HFF, SSB and maternal BMI independent of BMI z-scores at wave 1 (baseline); with each additional occurrence of SSB and HFF consumption intake per day, BMI z-score increased by 0.015 U (P < 0.01) and 0.014 U (P < 0.001), respectively. With each additional maternal BMI unit, BMI z-score increased by 0.032 (P < 0.001).


Higher BMI z-scores were strongly associated with the consumption of SSBs and HFFs. Future efforts to prevent obesity should consider urgent action to address the impact of the consumption of SSBs and HFFs in childhood.


The level of obesity for children in developed countries, such as Australia, has reached epidemic proportions, and is now a major preventative health priority for most countries. Low levels of physical activity, high levels of television and computer screen time and high consumption levels of energy dense and nutrient poor food have all been identified as significant contributors to this epidemic [1-3]. In Australia, the reduction of sugar sweetened beverages (SSB) and high fat foods (HFF) has been the focus of many prevention measures [4-6] but as yet there is not clear evidence about the relationship between high consumption of SSB and HFF and childhood obesity.

Studies in developed and developing countries have demonstrated that increases in SSB are associated with increased body mass index (BMI) [7-10]. These studies have principally examined BMI at two time points where higher levels of SSB consumption at the earlier time point was found to be predictive of greater levels of BMI at a later time point. Systematic reviews of cross sectional and two-time point studies have identified associations between consumption of SSBs and increasing body weight in adults and children [11-13].

Critics of the focus on reducing SSB to prevent obesity argue that the existing studies are limited because they are often cross sectional or examine risk at baseline against outcomes at time two without measures across multiple time points [14, 15]. In this context, it remains clear that stronger evidence from studies that allows for temporal influences is required, especially if preventative measures targeting SSB are to be sustained and supported in the long-term by funders and policy makers [11, 16]. Longitudinal studies whereby data on food and drink consumption, and individual, family and environmental are collected over several time points (>2) is one way of developing such evidence.

Research from Barclay and Brand-Miller [17] point to a possible paradox; they identify decreasing sugar consumption and SSB consumption in Australia alongside increasing prevalence of obesity in Australia. Many have argued that this indicates no association between sugar consumption and obesity patterns at population level. Others have found that industry funding of nutrition-related scientific articles may bias conclusions in favor of sponsors' products [12, 18] and so potential conflicts of interest exist in the three key reviews demonstrating no association between high levels of SSB consumption and high body weight as they were industry funded [15, 16, 19]. Moreover, all data used in the analysis to support the proposed Australian paradox were cross sectional. Longitudinal designed studies can assist to untangle and better understand this seemingly paradoxical relationship.

High fat food (HFF) consumption has also been implicated in the obesity epidemic among children and adolescents. Australian studies examining HFF consumption and obesity among children have shown mixed results; for example, cross-sectional studies have found positive relationships between more frequent fast food consumption at home and a person being overweight [20]. Conversely, lower fat intake in girls and boys and lower consumption of energy dense snacks in boys has been shown to be associated with a greater risk of overweight/obesity in a sample of Australian adolescents [21]. Caution needs to be applied when interpreting cross-sectional studies. For example, without temporal data they cannot establish causality nor consider the possibility of reverse causality (that is being overweight causing a reduction in HFF consumption).

Overall, studies examining the relationship between children's body weight and both SSB and HFF have been inconclusive. These studies have found weak relationships between consumption of HFF and increased body weight but more robust findings between SSB consumption and increased body weight. This has been found in one study within Australia [22, 23] and other high income countries [8, 24, 25]. The mixed findings may be due to the differences in foods included in studies and the limitations of dietary assessment methods. Measurement and interpretation of SSB consumption is less complex than measurement of HFFs as the kilojoule content of SSBs is quite uniform and the beverages are usually presented in standard sizes so measurement can be more precise. Conversely HFF includes a range of diverse food items, and studies have variously defined HFFs according to frequency of consumption, calculated energy density or number of convenience foods (not fast foods) [24, 26, 27].

Stronger evidence is needed to adequately support government decisions on investing in strategies that reduce HFF and SSB consumption. Longitudinal data with more than two time points, using national representative samples, and which uses strong and reliable measures will be vital in providing this information. Such data provide the opportunity for analysis that allows for trajectory modeling, and the assessment of influences at a variety of levels. Such an opportunity is provided by the Longitudinal Study of Australian Children (LSAC); a large, high quality, observational longitudinal cohort study over multiple time-points that includes robust data collected every 2 years on a representative cohort of Australian children. Such a data set combined with an analysis method, such as Multilevel modeling, that allows individual growth trajectories and multiple influences over time to be modeled will assist to provide stronger and more robust evidence.

This article determines the longitudinal relationship between frequency of consumption of SSBs and HFFs and change in body weight over 6 years (four waves) in Australian children aged from 4 to 10 years. It uses a multilevel growth model to identify whether SSB and HFF consumption is changing over time, and whether this consumption is linked to changes in BMI z-scores.


Sample and study design

The LSAC is an ongoing nationally representative cross-sequential longitudinal survey study which aims to examine Australian children's development and wellbeing and how this relates to social, economic, and cultural aspects of their environment. There are two cohorts; the B cohort (birth) and K cohort (kindergarten). The sample for this analysis comprised all children in the K-cohort (4-5-years old at wave 1—baseline) with complete height and weight data from the four waves of LSAC (wave 1—baseline, 2004; wave 2—first follow-up, 2006; wave 3—2nd follow-up, 2008; and wave 4—3rd follow-up, 2010) [28]. Briefly, LSAC employed a two-stage clustered sampling design stratified by a state and capital city statistical division /rest of state division and clustered by postcode within each stratum. Children born between March 1999 and February 2000 were randomly selected to achieve a cohort aged between 4.3 and 5.2 years at interview with all birth months represented. There were 4,169 children who completed all four waves of data collection and this cohort was representative of the Australian population on most demographic features [29]. Trained professional interviewers conducted face-to-face interviews in the child's home with the study child's primary caregiver (“Parent 1,” usually the biological mother), who also completed a written questionnaire.

There were 4,983 respondents who participated in the baseline (wave 1) survey, 4,464 at wave 2, 4,331 at wave 3 and 4,169 at wave 4 (Table 1). Compared to the baseline, the attrition rate for waves two, three, and four were, 10.4, 13.1, and 16.3%, respectively. Full details of the characteristics of responders are available from The LSAC technical papers [29-31]. Common characteristics that were associated with continued participation in the study included; if Parent 1 was female, had a bachelor degree and the study child lived in a home that was being paid off rather than rented [29-31]. Additionally, t test analysis of the key variables used in this study showed that the BMI z-scores at baseline of the children who remained in the study (mean = 0.59) were similar to those who left the study (mean = 0.58; P = 0.65), average SSB consumption was lower for those who remained in the study (mean = 1.4) compared to those who left (mean = 1.7; P < 0.001) as was average HFF consumption (mean = 2.0: mean = 2.1; P = 0.005).

Table 1. Participant characteristics and differences between child BMI-z scores, frequency of consumption of SSB and HFF and mother BMI at the first wave (2004) and the subsequent waves (2006, 2008, and 2010, respectively)
Characteristic Wave 1; N = 4,983Wave 2; N = 4,464Wave 3; N = 4,331Wave 4; N = 4,169
  1. a

    WHO Reference 2007 (43);

  2. *significant at P < 0.05;

  3. b

    **significant at P < 0.001. Bold text indicates a significant difference from wave1, differences tested using linear regression.

ChildMale, n (%)2,537 (50.9)2,277 (51.0)2,212 (51.1)2,132 (51.1)
  Mean (SD)Mean (SD)Mean (SD)Mean (SD)
 Age (years)4.8 (0.2)6.9 (0.2)8.8 (0.2)10.9 (0.3)
 BMI z-scoresa0.65 (1.00)0.51** (1.1)0.59* (1.17)0.59* (1.19)
Sugar sweetened beverage (mean number of times consumed in the 24 hours prior to survey)All1.7 (1.2)1.4** (1.2)1.3** (1.1)1.5** (1.2)
 Males1.7 (1.2)1.5** (1.2)1.3** (1.1)1.6** (1.2)
 Females1.6 (1.2)1.4** (1.1)1.2** (1.1)1.5** (1.2)
High fat foods (mean number of times consumed in the 24 hours prior to survey)All1.9 (1.2)1.9 (1.2)1.8* (1.2)2.5** (1.8)
 Males2.0 (1.3)2.0 (1.3)1.9 (1.2)2.6** (1.9)
 Females1.9 (1.2)1.9 (1.2)1.8* (1.1)2.3** (1.7)
MotherAge (years)34.6 (5.2)36.8 (5.1)38.9 (5.2)41.0 (5.2)
 BMI, (kg m−2)25.3 (5.2)25.4 (5.2)25.8** (5.3)26.5** (5.7)


Written informed consent was obtained for each participating child, and the LSAC study was approved by the Australian Institute of Family Studies Ethics Committee.

Anthropometric measurement

Children's weight was measured in light clothing to the nearest 50 g using glass bathroom scales (Salter Australia, Code 79985; Springvale, Victoria, Australia) and height to the nearest 0.1 cm using a portable rigid stadiometer (Invicta, (Leicester, UK), Model IPO955). The averages of two height measurements were used in analyses; where the two differed by more than 0.5 cm a third measurement was taken and the average of the two closest was used.

Outcome measure

BMI z-scores were the outcome measures and these were calculated using the age- and sex-specific WHO Growth Standard for 6- to 60-month-old children [32] and Growth Reference for 5- to 19-year-old children [33], where applicable.

Independent variables


Both the SSB and HFF were derived variables from the LSAC dataset. They were both reported by parent 1 during the face to face interview. The stem of the items was identical, “In the last 24 h has your child had the following foods or drinks once, more than once or not at all? Frequency was coded 0 for “not at all,” 1 for “once” and 2 for “more than once.” Responses were summed and a final score for frequency of SSB and HFF consumption was allotted.

SSBs comprised two parent-reported survey questions including the frequency of consumption of: [1] fruit juice; [2] soft drink or cordial (not diet), in the 24 h prior to the survey.

HFFs was measured by the reported frequency of consumption of: [1] meat pie, hamburger, hot dog, sausage or sausage roll; [2] hot chips or French fries; [3] potato chips or savoury snacks such as “Twisties”; 4) biscuits, doughnuts, cake, pie or chocolate, in the 24 h prior to the survey.


The LSAC composite household socio-economic position (SEP) variable was constructed by Blakemore et al. and was derived from standardized scores for: combined annual household income (with natural log transformation); parents' years of education; and parents' occupations (main occupation and occupational status) [34].

Demographic information such as child age and gender and mother age and self-reported anthropometry was also collected. Mother's BMI was calculated using weight (kg)/height2 (m2). All continuous independent predictors were centered at the sample mean.

Missing data

There was very little missing data for most variables; BMI z-scores <4%, SSB <1%, HFF <2%, mother age <2% and household SEP <5%. An exception to this was mother's self-reported BMI which had missing data from 21% of cases altogether (ranging from a high of 29% at wave 2 to a low of 10% at wave 4).

Statistical analyses

Demographic data were examined using descriptive statistics. A multilevel growth model (MGM) was used to examine the BMI z score changes over the four waves of data collection. A growth model was used as it permitted [1] modeling of change over time; and [2] simultaneous modeling of change occurring within an individual over time, and between individuals over time.

To systematically build a MGM, the data were organized into a long format and the variables were categorized into three levels. Consistent with MGM methods, level-one variables included the continuous dependent variable BMI z-score at each of the four waves of data collection and the variable used to model time (wave). Using a long format, and categorizing these variables as level one is a unique characteristic of MGM. From a data perspective, this effectively treats each wave of data for each person as a different case; the fundamental difference between each case is the time at which the dependent variable information is collected. Thus, using the ID number for each case to cluster individuals together, a growth trajectory for each person can be estimated (see analytical strategy below). To assist in interpretation, wave was centered at baseline (baseline = 0, wave 2 = 1, wave 3 = 2, wave 4 = 3), and all other variables used in growth models were centered at their grand mean.

Analytical strategy

Following Singer and Willet (2003), a four-part analytical strategy was used [35]. First, an unconditional means model (only a random intercept, varying by participant ID) was established. An unconditional means model is used to identify a baseline of unexplained variance, and to establish baseline fit statistics. As other variables are introduced into the model, this information was used to make judgments as to how well the unexplained variance is reduced, and how well model fit is improved. Fit statistics used included Aikake's information criteria (AIC) and the Bayesian Information Criteria (BIC).

Second, an unconditional growth model was established. This model is also described as a simple growth model, as it only includes the time (wave) variable as a predictor. In this model the association between the time variable (wave) and the dependent variable (BMI z-score) is allowed to vary by individual; in effect, this is how a growth trajectory for each person is achieved. Like the unconditional means model, the unconditional growth model establishes a baseline of unexplained variance, and model fit. To aid interpretation of the intercept, wave was centered at baseline. Thus, for all the models when time was zero, the intercept represented the average BMI z-score.

Model three examined whether frequency of SSB and HFF consumption over the four waves were predictors of baseline BMI z-score (intercept) and rate of BMI z-score change (slope of wave predictor). Model four introduced covariates and explored possible interactions. In all cases, P < 0.05 was considered statistically significant. Analyses were conducted using Stata release 12.0 (StataCorp, College Station, TX, 2011).


The characteristics of the participants and their mothers are shown in Table 1. The average age of the study child was 4.8 years at wave 1 and increased by 2 years at each subsequent wave. BMI z-score was higher at wave 1 than at subsequent waves. The mean frequency of consumption of SSBs varied over the four waves from 1.7 times per day at baseline decreasing to 1.3 in wave 3 then increasing to 1.5 times in wave 4. The mean frequency of consumption of HFFs varied from 1.9 times per day at baseline and increased to 2.5 times in wave 4.

Table 2 presents the results of the growth model building process. The conditional mean model (model I) identified a baseline variance within individuals of 0.288. A significant Log Likelihood test (see model II) indicated that individuals' BMI z-scores not only differed significantly between individuals, but also further varied between local areas (post codes) (χ2(1, n = 4961) = 23.32 P < 0.01).

Table 2. Multilevel growth models for children's BMI over four waves of data collection
 Model IModel IIModel IIIModel IVModel VModel VI
 Conditional meanConditional mean 2Unconditional growth modelWave & HFF & SSB random effectsFinal ModelUsing imputed data
 β (95CI)β (95CI)β (95CI)β (95CI)β (95CI)β (95CI)
  1. *P ≤ 0.05;

  2. **P ≤ 0.01;

  3. ***P ≤ 0.001; Model 1 : unconditional mean; Model 2: unconditional mean, three level; Model 3: unconditional growth, three level; Model 4: BMIZ start and rate of change function of high fat and high sugar; Model 5: final model.

Constant/intercept0.598 (0.569, 0.626)***0.602 (0.568, 0.637)***0.615 (0.579, 0.651)***0.617 (0.583, 0.651)***0.632 (0.601, 0.664)***0.623 (0.592, 0.654)
Wave  0.009 (−0.018, 0.001)−0.012 (−0.022, −0.003)**−0.028 (−0.038, −0.017)***−0.025 (−0.035, −0.015)***
Wave#HFF   −0.000 (−0.007, 0.007)- 
Wave#SSB   −0.000 (−0.006, 0.006)- 
SSB   0.018 (0.005, 0.032)***0.015 (0.004, 0.025)**0.017 (0.007, 0.027)**
HFF   0.025 (0.013, 0.038)**0.014 (0.005, 0.023)***0.021 (0.014, 0.029)***
Mothers BMI   -0.032 (0.028, 0.035)***0.020 (0.017, 0.023)***
Household SEP   -−0.036 (−0.058, −0.014)***−0.031 (−0.051, −0.011)**
Female   -−0.124 (−0.180, −0.068)***−0.135 (−0.188, −0.082)***
N1 (ind)17,62017,62017,62017,53313,37917,947
N1 (groups)4,9614,9614,9614,9584,5354,983
N2 (area)      
σ2 within postcode 0.0290.0260.0240.0240.015
σ2 within person0.2880.2880.1870.1840.1840.227
σ2 in initial status (intercept)0.9630.9350.8640.8660.8660.793
σ2 in rate of change (wave)  0.0620.0640.0640.062
σ2 covariance (intercept#slope)  −0.004−0.006−0.006−0.019

Model III in Table 2 is the unconditional growth model, where wave is the only predictor in the model. As wave was centered, the intercept indicated that the average BMI z score at baseline was 0.614. The coefficient (β) for wave also indicated that that BMI z-scores decreased by ∼0.01 (P = 0.072) for each wave, but this was not statistically significant. For model II the residual representing the average variation between an individual's predicted growth trajectory and their actual trajectory, was reduced from 0.288 to 0.187, a 35% reduction. This suggested that 35% of the individual change in BMI z score was possibly systematically related to time. A reduction in the AIC and BIC fit statistics for model III, when compared with the unconditional mean model suggested the model was a better fit to the data.

Model IV examined whether baseline BMI-z (intercept) and the rate of change over the four waves (slope coefficient for wave) was systematically related to the amount of SSB and HFF consumed. In multilevel growth modeling terms, these specifications are described as identifying whether SSB and HFF are random effects associated with each individual growth trajectory. The equations for this model are presented below. Specifying a random effect to be associated with the slope of a variable always results in an interaction term [36]. This is shown in the composite model below, where the substitution of level two equations into the level one equation results in an interaction term between HFF and wave, and an interaction term between SSB and wave. A reduction in the AIC and BIC fit statistics for model IV, when compared with the unconditional mean model suggested the model was a better fit to the data.

display math
display math
display math

Model four composite equation:

display math

Multilevel growth trajectory for each individual i at time j : Error terms capture individual stochastic variation for each trajectory: ζ01 = individual residual associated with overall intercept (γ00) for individual i; ζ1i = individual residual associated with overall slope (γ10) for individual i at wave j; εij= residual associated with predicted BMI-z for individual i at wave j. As wave is centered at baseline, the intercept (γ00) is the average baseline BMI-z value when time = 0, and all other variables in the equation are held constant. The interaction terms are modeling systematic variation of the wave coefficient as a function of SBS and HFF.

As model IV in Table 2 indicates, the interaction term for wave and HFF (β = 0.000; P > 0.05) and the interaction term for wave and SSB (β = 0.000; P > 0.05) were not significant, suggesting that SSB and HFF consumption were not systematically changing over time. Wave in this model was a significant predictor of child's BMI (−0.012). Before removing insignificant interactions from the model, possible confounding variables were introduced.

With the confounders introduced, the coefficients for both interaction variables remained statistically insignificant and thus were removed. Model V in Table 2 presents the final model, where only statistically significant variables have been retained. The equations associated with model five are presented below. A reduction in the AIC and BIC fit statistics for model V, when compared with the unconditional mean model suggested the model was a better fit to the data.

display math
display math
display math

Model 5 composite equation:

display math
display math

Multilevel growth trajectory for each individual i at time j: Error terms capture individual stochastic variation for each trajectory: ζ01 = individual residual associated with overall intercept (γ00) for individual i; ζ1i = individual residual associated with overall slope (γ10) for individual I at wave j; εij = residual associated with predicted BMI-z for individual i at wave j. As wave is centered at baseline, the intercept (γ00) is the average baseline BMI-z value when time = 0, and all other variables in the equation are held constant.

For the final model (Model V), as wave was centered at baseline and all other variables centered at grand mean average, the intercept indicated that when all the variables were held constant at their average, BMI z-score at baseline was 0.63. The coefficient for wave suggested that BMI z-score was decreasing at a rate of 0.028 units per wave. When controlling for confounders, SSB consumption was significantly related to average BMI z-score. With each additional occurrence of SSB intake per day BMI z-score significantly increased by 0.015 units (P < 0.01). Similarly, with each additional occurrence of HFF consumption per day, BMI z-score significantly increased by 0.014 units (P < 0.001). Although not of primary interest in this study, maternal BMI was also a significant predictor of BMI z-scores in this sample; with each additional maternal BMI unit, BMI z-score increased by 0.032 (P < 0.001).

As there was a substantial proportion of missing data for the covariate “mother's BMI,” to ensure that the final model was a reasonable and accurate representation of the data, the model building process was repeated with all missing data imputed. Twenty imputations were made, and the model building produced similar results (full details available from the corresponding author). For comparison, the final model with imputed data is presented in the final column of Table 2. The coefficients in the imputed model were similar to the final model without imputed data; P values were also similar. Variance explained in the imputed model was greater at a postcode level and less at an individual level. However, variance of the intercept and slope were both reduced.


This article examines the longitudinal relationship between frequency of consumption of SSBs and HFFs and change in body weight over 6 years (four waves) in Australian children aged from 4 to 10 years. Using a representative sample of Australian children aged from 4 to 10 years, we found no evidence that SSB and HFF consumption had changed over this period. We found a positive association between frequency of consumption of SSBs and BMI z-scores and between frequency of consumption of HFFs and BMI z-scores; where higher consumption was associated with a greater proportion of BMI z-scores. These effects were significant over and above the strong effect of maternal BMI which is an established determinant of childhood BMI. These findings indicate that continued investment in reduction of children's consumption of HFF and SSB is warranted and should remain a national priority.

The findings that SSB was not decreasing over time is not consistent with Barclay and Brand-Miller [17]. This study failed to find a consistent decrease in the frequency of SSB consumption in this sample but there was a plateauing of consumption which is similar to that reported by Rangan et al., who found that after an initial rapid increase in SSB consumption of 240% from 1969 to 1996 among Australians, from 1997 to 2006 this increase has slowed to 5% per year [37]. Moreover, the finding that SSB consumption was related to increased BMI z-scores is consistent with reviews by Malik, Olsen, and Vartanian [11, 12, 18].

These results may also provide insights into the rapid unhealthy weight gain seen in adolescence, which is an age group also associated with large increases in SSB consumption. Striegel-Moore et al., [9] reported a threefold increase in the consumption of SSBs in a study following a cohort of girls from 10 to 20 years of age. They also estimated that BMI z-scores increased by 0.01 U for every 100 g of regular soft drink consumed. This is similar to the findings of the current study, where each additional time per day that SSB were consumed was associated with an increase in BMI z-scores of 0.015 U. An Australian study in 93 secondary schools in the state of New South Wales found that in 2004, over half the boys and one-third of girls reported drinking SSBs daily [38]. A Victorian study found that about 70% of children/adolescents consumed SSBs on the day prior to the survey [39]. Given the high prevalence of consumption of SSBs and the strong link to obesity, interventions to prevent obesity need to focus on limiting access to and availability of SSBs to children.

The consumption of HFFs was also related to higher BMI z-scores. This finding was consistent with Niemer et al. [26] who found that fast food consumption at baseline predicted increased BMI z-scores at follow-up. The widely varying definitions of HFFs, which in some studies includes only “fast foods” and in others a range of high fat products is likely to be responsible for at least some of the heterogeneity of findings. These results are, however, similar to other cross-sectional studies that reported relationships between high fat food consumption and increased body size among children [20, 27, 40].

There have been mixed findings from studies that analyzed consumption of both SSBs and HFFs simultaneously. Most studies reported relationships between the consumption of SSBs and body weight but not between consumption of HFF and body weight (8,22-25) whereas the current study found a relationship between HFF consumption and BMI z-scores and also between SSB consumption and BMI z-scores. It is important to understand the relationship between the two as they often consumed together (including when offered together in “meal deals” which aim to increase consumption) and both form major components of high energy dietary patterns [38].

The major strength of this analysis is that the data are drawn from four longitudinal waves of a large, representative cohort of Australian children. The rigorous data collection methods are well-documented and transparent. Large cohort studies provide the opportunity to detect relationships between variables over time, and the multi-level method of analysis can account for different levels of influence that may affect the variables of interest.

These analyses did not control for physical activity as the LSAC data set does not include measures of intensity of physical activity but this may have impacted only slightly on the obtained results as Jago found that physical activity was only weakly related to obesogenic diets among adolescents [41]. Serving sizes were not measured in the survey, only the number of times the study child consumed SSB or HFF in the 24 h prior to the survey, but the same items were used over the four waves of data collection so internal comparisons are valid. In addition, 24-h recall data are not well suited to identifying relationships in individual-level analyses due to the fact that reported intakes may not be representative of usual or average intake. This may be especially the case for less frequently consumed food and drink items, including SSBs and snack foods. It is likely, therefore, that these findings may under-estimate the true relationships. On other limitation was that no data on metabolic outcomes such as glucose, insulin or HbA1c levels were collected. Future research would be strengthened by inclusion of these data.


This longitudinal study of Australian children found higher BMI z-scores were strongly associated with the consumption of HFFs and SSBs. While these findings provide further that BMI z-scores have decreased or remained stable recently, the prevalence of overweight/obesity is still high. For additional gains in the prevention of obesity urgent action to address the impact of the consumption of SSBs and HFFs in childhood must be considered.


This article uses unit record data from Growing Up in Australia, the Longitudinal Study of Australian Children. The study is conducted in partnership between the Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA), the Australian Institute of Family Studies (AIFS) and the Australian Bureau of Statistics (ABS). The findings and views reported in this article are those of the author and should not be attributed to FaHCSIA, AIFS or the ABS.