This is the first longitudinal study which examined prediction of breakfast, lunch and evening meal frequencies over two important transition periods in young people’s life from age 15 to 19, 19 to 27 and 15 to 27 years.
The proportion of respondents who ate breakfast daily decreased continuously from adolescence into early adulthood. The proportion who ate lunch daily was constant over the period, while the proportion who ate an evening meal daily decreased from early adolescence into late adolescence and then increased into early adulthood. Lytle et al.  studied cross-sectional prevalence for all three meal types among third, fifth and eighth graders. The study population was younger than ours but the patterns were the same. Sweeting et al.  found that the prevalence of breakfast and lunch decreased with age and the prevalence of the evening meal remained constant over the period. Niemeier et al.  found that breakfast consumption decreased with age from adolescence into early adulthood. We hypothesize that the decrease in eating breakfast with increasing age is linked to the transition from living at home with the family of origin and moving to live alone. In the family, breakfast may be more available and it may be more appealing to eat breakfast living in a family setting. Regarding the findings for the evening meal, some of the changes observed in prevalence may ascribe to the change from hot evening meal to evening meal in the item formulation. However, it could be hypothesized that people tend to have higher evening meal frequency in adulthood due to family obligations.
We found that low meal frequencies in early adolescence predicted low meal frequencies in late adolescence. Further, low breakfast and lunch frequencies in early adolescence predicted low breakfast and lunch frequencies in early adulthood. Low meal frequencies in late adolescence predicted low meal frequencies in early adulthood. The predictions were stronger over shorter than longer periods. For breakfast frequency, Merten et al.  found that high breakfast frequency in adolescence predicted high breakfast frequency in early adulthood. Their study period was shorter than that of our study, but the findings are consistent. Sweeting et al.  studied the stability in meal frequencies from adolescence into late adolescence. Therefore, it is not possible to compare our results directly with their findings. Their findings showed that evening meal frequencies was constant over the period, while breakfast and lunch frequencies decreased. To some extent, their findings are consistent with our observations.
The presented results of the modifying effect of gender showed that among women, low meal frequencies in early adolescence predicted low meal frequencies in early adulthood, but we did not find similar predictions among men. Further, we found a stronger prediction among men than among women for evening meal frequency from adolescence into late adolescence and from late adolescence into early adulthood. In the study by Sweeting et al.  the changes in meal frequencies are stratified by gender. They found that men to a higher degree than women maintained their evening meal frequency from adolescence into late adolescence. It seems that the predictions, especially among men, are stronger over shorter periods than over longer periods.
We found that family structure modified the prediction between low lunch frequency in early/late adolescence and low lunch frequency in early adulthood. Our findings indicate that living in a non-traditional family structure in adolescence decreases the prediction of low lunch frequency. Nevertheless, no other studies have investigated whether adolescent family structure influences prediction of meal frequencies over time. Cross sectional studies show that living in two-parent families is positively associated with high meal frequency in adolescence [21, 23, 28, 38–41]. This is not consistent with our longitudinal findings. If the measurement of adolescent family structure had been more refined, we would have been able to differentiate the category of other family structures, which covers a broad range of family constellations.
The strengths of this study are high response rates, a national representative sample and a long follow-up period that includes two important transition periods. Further, the prediction analyses have been conducted using both by simple logistic regressions, GEE-models and random effect logistic regression models. The findings are consistent between models. The findings must also be seen in relation to the limitations of the study. There is a risk of selection bias because the original sample was reduced by the 11% of parents who declined to allow their child to participate in the study. We are unable to provide further information to substantiate this issue. The loss to follow-up may also cause selection bias. We found significantly higher loss to follow-up among respondents with the following characteristics: men, low adolescent family social class, and low breakfast and lunch frequency. This may have influenced the generated results. However, worst and best case missing analyses showed no discrepancy from the generated results. Information bias may exist in our study due to using self-reported data. Also the slight change in item formulation for the evening meal for the second follow-up could introduce information bias. However, it is unlikely that this change compromised the observed prediction patterns.
Studies about meal consumption suffer from difficulties in defining a meal. Denmark has a Nordic meal culture of three main meals a day , and people use the concepts breakfast, lunch and evening meal in their daily life. Nevertheless, it is not clear what constitutes a proper meal and how the respondents interpret the term. The literature is limited regarding the use of frequency questionnaires for measuring meal consumption. A frequency question on breakfast consumption was validated in the Health Behavior in School-Aged Children (HBSC) study in 2004–2005. The validation showed a moderate accordance with food habits diaries (kappa statistics 0.47) . The formulation of the question was similar to the formulation in this study but not identical. More validation studies of frequency questionnaires for measuring meal consumption are needed.
In every study possible unmeasured confounders must be considered. Other studies have found ethnic differences in diet and breakfast consumption [44, 45] and it would have had been relevant to include ethnic background as a confounder. However, our dataset does not provide information on ethnic background. Several studies have found an association between meal frequencies and BMI [2, 3, 5, 6], but we abstained from including BMI as a covariate because it is unclear whether BMI confound or mediate the association. Further, aspects of family life such as parents’ working hours, family coherence and child-raring styles may also confound the findings but we didn’t have data to investigate these ideas.