The current study presents tracking coefficients for fruit, vegetables and unhealthy snacks covering a 15-year period. To our knowledge, this is the first tracking study presenting separate tracking coefficients for fruit and unhealthy snacks after a dietary intervention.
Overall tracking of fruit, vegetables and unhealthy snacks
Based on the FFQ, covering a 15-year period in the total sample, tracking coefficients of fruit and vegetables were 0.33 and 0.36, respectively, which is considered low to moderate and is in line with the literature [13]. Further, the tracking coefficient for unhealthy snacks were 0.46 for males and 0.40 for females, indicating that males had a more stable consumption over time. Predictability for weekly vegetable consumption was significantly higher for males that females, but predictability of fruit and unhealthy snacks consumption did not differ between the sexes.
Our analysis based on the 24-h recall found that subjects who had parents with higher education had a higher tracking coefficient of fruit than their counterparts. Although this result was not confirmed in the analysis based on the FFQ, it might indicate that stability of fruit consumption may be related to socioeconomic status, however this needs to be examined and explored in other studies.
The existing evidence of tracking of fruit, vegetables and unhealthy snacks in children and adolescents derives from studies with different design, various methods used and variation in number of years followed [13]. We have not been able to identify other studies using mixed models for assessing tracking of fruit, vegetables and unhealthy snacks which makes comparison difficult. Furthermore, no other studies have previously reported tracking of dietary habits after a dietary intervention period. A study by te Velde et al., however, used a similar statistical method (GEE) reported comparable tracking coefficients for fruit and vegetable over a 24-year period with six measurements [15]. Patterson et al. found poor tracking for vegetables, fair tracking for fruit and weak tracking for different snacks using Cohens’s weighted kappa over a 6-year period and the design of the study allowed for just two measurements [17]. The weak tracking coefficients in the latter study can be explained by the dietary method on which they based their analysis; one single 24-h recall, which is less suitable for assessing tracking. Further, Lien et al., analysed by stability and rank [16], and reported good tracking of fruit, vegetables and sweets/chocolate. Lake et al. used Pearson’s correlation coefficient and reported tracking for fruit and vegetables (combined) but found no correlation of sugary foods over a 20-year period [18]. The statistical methods and/or design used by Patterson, Lien and Lake has the disadvantage of only using two measurements as opposed to our study where the statistical method allowed for utilization of all available data. In our study, we only found a significant difference for tracking of snacks by gender, where boys had a higher tracking coefficient, as opposed to Lien et al. who found that girls had higher stability of consumption of sweets/chocolate and soft drinks [16].
Tracking of fruit, vegetables and unhealthy snacks by intervention
Ideally, public health interventions and initiatives should have lifelong effects. The main argument used for initiation of interventions promoting increased FV consumption at an early age is that dietary habits learned in childhood “tracks” into adulthood. So far, no study has actually evaluated if tracking of diet over years has been modified by a dietary initiative. It is important to realize that the tracking coefficient concerns the relative position of a subject, so if everyone increases their intake the same amount (because of an intervention), the tracking coefficient will not change. It will only change if some subjects change more than others and when their relative position changes.
We found no significant difference between the intervention and the control group when analysing tracking of weekly fruit consumption. The tracking coefficient for unhealthy snacks was significantly higher in the intervention- compared to the control group from follow-up one (while the intervention was operating), but not from baseline and follow-up two (right after the intervention ended). Indicating that from follow-up one, the intervention group had a higher stability of unhealthy snacks compared to the control group. The subjects in the intervention group generally had a lower mean consumption of unhealthy snacks compared to the control group at all time-points (Table 1). Therefore, a higher tracking of unhealthy snacks shown in the present study indicates that the intervention group maintained a lower consumption over time. Further, the tracking coefficient for vegetables (times/week) was significantly higher in the control group from follow-up two, indicating a higher stability of consumption. However, the unadjusted consumption of vegetables (times/week) was lower in the control group compared to the intervention group at most timepoint, Table 1.
Predictability of remaining in the highest tertile of weekly fruit consumption from follow-up one was higher in the control group than the intervention group, but not from baseline and follow-up two. At follow-up one, the intervention was still operating and the subjects in the intervention group had a higher mean consumption of fruit. Therefore, it is likely that when the intervention ended, more subjects fell into the second and first tertile of consumption, thus lower predictability of remaining in the highest tertile. Further, within both groups and at all timepoints, subjects in the highest tertile at the initial measure had higher odds of remaining in the highest tertile over time, indicating that if you are a high consumer in childhood you are more likely to be a high consumer in adulthood (Table 3).
From follow-up two (just after the intervention ended) the participants in the control group had a significantly higher odds compared to the intervention group of remaining in the highest tertile of unhealthy snacks, but not from baseline and follow-up one, indicating that participants in the intervention group had a lower odds ratio of remaining in the highest tertile of unhealthy snacks.
Combined, the tracking- and prediction analyses used in the present study suggest that one year of school fruit provided without parental payment for 11.8-year-old children does not lead to different tracking coefficients of weekly fruit consumption from childhood to adulthood.
If the school fruit scheme actually modified tracking of fruit, we would expect lower tracking from baseline and higher tracking coefficients from follow-up one and two in the intervention group compared to the control group. A lower tracking coefficient from baseline would indicate a higher degree of fluctuation (within and between subject), preferably subjects with a low consumption increasing their intake over time. A higher tracking coefficient from follow-up one and two in the intervention group compared to control group would indicate that that the subjects to a higher degree maintained their high consumption over time. However, our prediction analysis suggests that if you are a high consumer in childhood you are also more likely to be a high consumer in adulthood.
On the other hand, we did observe significantly different tracking of unhealthy snacks between the groups from follow-up one; subjects in the intervention group had a higher tracking coefficient (and a lower consumption). These results may indicate that the school fruit scheme influenced tracking of unhealthy snacks. Yet, this was not sustained from follow-up two.
Tracking coefficients by dietary methods
When interpreting the tracking coefficients, several considerations need to be kept in mind. Firstly, the magnitude of the tracking coefficient is dependent on the length of the time measurement(s) [29]. A high tracking coefficient over a short period of time may just be evidence of reliability of the used method, while a lower tracking coefficient over a long period of time, as found in this study, may be a stronger indication of tracking [29]. Secondly, tracking coefficients are influenced by measurement error. For instance, tracking coefficients for dietary consumption is lower than tracking of biological properties [30]. The reason for lower tracking of dietary consumption compared to i.e. biological properties is probably partly due to lower tracking, but also because it is difficult to measure dietary consumption, as eating behaviour can be highly variable both within and between days for an individual [30]. Hence, it is important to use a dietary measure which reflects habitual dietary consumption, i.e. FFQ, as used in this study, reflects habitual dietary (i.e. less measurement error) consumption better than a single 24-h recall. As mentioned, one single 24-h recall does not reflect usual dietary consumption at the individual level and may be more susceptible to daily variations and consequently provide low tracking coefficients. The current study confirmed that assumption. Still, the results of the 24-h recall are interesting and were thus presented, as it can be useful for other researchers to see how choice of dietary methods may influence results from tracking analysis.
Strengths and limitations
The main strength of this study was that we assessed tracking over 15 years according to a dietary intervention, which has not been done before. Further, we used all six measurements and a suitable statistical test, providing knowledge into trajectories over time. We compared tracking coefficients derived from two dietary methods; FFQ and 24-h recall, where FFQ, is suitable for assessing tracking and gives an indication of habitual consumption [31]. Another strength of the present study is that both unhealthy and healthy dietary variables were studied, which provides a broader insight into dietary behaviour from childhood to adulthood.
Longitudinal studies are prone to loss of participants at follow-up, and this study is no exception. In 2009, the survey was sent to participants by regular mail. In addition, the participants had to send their completed survey back by regular mail. The delivery and return of the survey most likely led to the low participation in 2009. However, drop-out was much smaller for the other surveys. We performed two drop-out analyses which showed that drop-outs or subjects with one or several missing measurements, generally had a lower weekly consumption of fruit and vegetables and higher consumption of unhealthy snacks. Further, girls and subjects with higher educated parents were better represented at the follow-ups. The rather large drop-out may have affected the results, however, we used mixed models for repeated measurements, which is an analysis that handles data with missing observations. Although we adjusted for sex and parental education, there might be other confounders e.g. total energy consumed which we did not have the opportunity to control for. The population in the FVMM is not a representative sample of the Norwegian population regarding parental education. In 1999, 28,5% of the population in Norway aged 34–44 years had higher education, which is lower than observed in the FVMM study [32].
The main challenge of estimating tracking coefficients for eating behaviour lies in measurement challenges, as eating behaviour can be highly variable both within and between days for an individual. Therefore, we don’t know how much of the low to moderate tracking coefficient may be explained by measurement error. In addition, all methods for collecting self-reported dietary data have well-known limitations [33].
Another limitation of this study is the Norwegian School Fruit Scheme (NSFS). The NSFS is a paid subscription program providing daily fruit to pupils at school, while the FVMM intervention was operating. Initially, schools could decide if they wanted to participate in the program. At participating schools’ parents could decide if they wanted to subscribe. All 29 FVMM control schools in the were given the opportunity to participate in the NSFS. Nine control schools participated, and 20 control schools declined to participate in the programme in the school year 2001/2002. In the nine control schools participating, there were both subscribers and non-subscribers (41% of the pupils subscribed).
Another limitation of the present study was the use of a non-validated measure for unhealthy snacks. Additionally, unhealthy snacks are a combined measure, which included both information about soda, crisps and candy/chocolate consumption. Therefore, we were not able to identify the contribution of the different snacks items to the tracking coefficient and how the foods would track separately.
