Introduction
Mood disorders, primarily depressive disorders, contribute more to worldwide disability than any other health condition (World Health Organization, 2017). Diet may influence mood disorders due to the physiological effects of nutrients on biochemical processes involved in mental health, such as hormones, neurotransmitter activity and the gut–brain axis (Lang et al., Reference Lang, Beglinger, Schweinfurth, Walter and Borgwardt2015). However, the frequency and timing of meals can also have hormonal, neurobiological and microbiome effects, thought to be related to circadian rhythms (Tahara and Shibata, Reference Tahara and Shibata2013; Asher and Sassone-Corsi, Reference Asher and Sassone-Corsi2015). Physical effects include possible influence on cardiometabolic conditions such as diabetes and obesity that are often comorbid with mood disorders (Stunkard et al., Reference Stunkard, Faith and Allison2003; Mattson, Reference Mattson2005; Lowden et al., Reference Lowden, Moreno, Holmbäck, Lennernäs and Tucker2010).
Existing research on the relationship between food timing and mood has largely involved à priori defined dietary behaviours and cross-sectional analyses. For example, skipping breakfast has been consistently cross-sectionally associated with depressive symptoms and poorer mental well-being among both youth (Fulkerson et al., Reference Fulkerson, Sherwood, Perry, Neumark-Sztainer and Story2004; Lien, Reference Lien2007; O'Sullivan et al., Reference O'Sullivan, Robinson, Kendall, Miller, Jacoby, Silburn and Oddy2009; Lee et al., Reference Lee, Han and Kim2017a) and adults (Smith, Reference Smith1998; Begdache et al., Reference Begdache, Chaar, Sabounchi and Kianmehr2017; Lee et al., Reference Lee, Park, Ju, Lee, Han and Kim2017b; Kwak and Kim, Reference Kwak and Kim2018). These associations are often clinically significant, and robust to potential confounders including socioeconomic factors (Lee et al., Reference Lee, Park, Ju, Lee, Han and Kim2017b) and lifestyle practices such as diet quality, smoking or alcohol consumption (Smith, Reference Smith1998; O'Sullivan et al., Reference O'Sullivan, Robinson, Kendall, Miller, Jacoby, Silburn and Oddy2009; Kwak and Kim, Reference Kwak and Kim2018). Other eating behaviours are less well studied, but snacking between meals has been associated with depressive symptoms among adults (Furihata et al., Reference Furihata, Konno, Suzuki, Takahashi, Kaneita, Ohida and Uchiyama2018), while snacking and meal skipping has been associated with higher levels of psychological problems in female adolescents (Farhangi et al., Reference Farhangi, Dehghan and Jahangiry2018). To our knowledge, only one prospective study has examined multiple eating behaviours. This study reported that having at least two out of three unhealthy eating practices of skipping breakfast, snacking after dinner or eating dinner shortly before bed was associated with a higher incidence of depressive symptoms (Huang et al., Reference Huang, Momma, Cui, Chujo, Otomo, Sugiyama, Ren, Niu and Nagatomi2017).
Limitations of previous studies include examining discrete eating behaviours, using non-clinical measures of depression or only considering the concurrent mood. Cross-sectional analyses are unable to identify the directionality of the relationship. Both high and low emotional states have been found to influence food consumption (Cardi et al., Reference Cardi, Leppanen and Treasure2015) meaning bidirectionality should be considered. Furthermore, despite the popularity of methods such as principal components analysis (PCA) to examine the patterns of nutritional intake, it is rare for data-driven approaches to be used to determine time-of-day eating patterns. Two time-of-day eating patterns (a conventional pattern of three main meals, and a snack-dominant pattern) were derived using PCA in a 2011 cross-sectional study (Kim et al., Reference Kim, DeRoo and Sandler2011). However, the outcome in that study was sleep duration, not mood.
There were two important rationales for this study. Firstly, empirical analysis of eating and drinking occasions would allow us to determine common eating patterns that explain variation in the timing of food intake over the day. The term ‘eating patterns’ refers to patterns related to the timing and relative size of meals/snacks as a proportion of daily intake, not the foods, nutrients or energy consumed. Secondly, examining bidirectional associations between eating patterns and clinical diagnosis of depressive episodes over time could help us understand the relationship between eating patterns and mood disorders if one exists. In this study, we aimed to determine if time-of-day eating patterns were longitudinally associated with mood disorders (dysthymia or depression) among an Australian cohort of young to middle-aged adults. We examined if eating pattern score predicted subsequent mood disorders, if tracking of pattern scores was associated with mood disorder over time, and if mood disorders predicted eating pattern scores.
Methods
Participants
In 1985, the Australian Department of Community Services and Health conducted the Australian Schools Health and Fitness Survey (ASHFS) of schoolchildren aged 7–15 years. A two-stage probability design derived a nationally representative sample. Of 121 schools approached, 109 schools participated (90.1% response rate). The student response rate was 67.6% (N = 8498).
During 2001–2002, ASHFS participants were traced and invited to participate in the Childhood Determinants of Adult Health (CDAH) study, resulting in enrolment of 5170 participants (61.0%) (Gall et al., Reference Gall, Jose, Smith, Dwyer and Venn2009). For the first follow-up 2004–2006 (CDAH-1), n = 2410 participants (aged 26–36 years) attended study clinics for physical measurements and completed questionnaires including a food frequency questionnaire (FFQ) and food habits questionnaire (FHQ). At the second follow-up 2009–2011 (CDAH-2), n = 1749 participants (aged 31–41 years) completed a mental health diagnostic interview, questionnaires and the same FFQ and FHQ used in CDAH-1.
Ethical standards
All procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. The State Directors General of Education approved the ASHFS, and signed parental consent was required for all participants. The Southern Tasmanian Health and Medical Ethics Committee approved the CDAH study protocol, and all participants gave informed written consent.
Measures
Eating occasions
At CDAH-1 and CDAH-2, participants were mailed questionnaires that were returned by post or collected at the CDAH-1 clinics. The FHQ included a meal pattern chart that collected information on the types of meals and drinks consumed from 6:00 h the previous day to 6:00 h that morning (Smith et al., Reference Smith, Gall, McNaughton, Blizzard, Dwyer and Venn2010). The 24-h period was divided into hourly periods (e.g. 6:00–7:00 h) from 6:00 to 23:00 h, and an overnight period of 23:00—6:00 h. For each time period, respondents were asked ‘Did you eat anything?’ with responses of ‘No’, ‘A snack’, ‘A small meal’ or ‘A large meal’, and ‘Did you drink anything’, with responses of ‘No’, ‘Alcohol’, ‘Water’ or ‘Something else’. Examples of meal types were given: snacks: a biscuit or a piece of fruit; small meal: beans on toast, boiled egg and bread, breakfast cereal, a pie; large meal: meat and three vegetables or a large serving of fish and chips. Participants were instructed that they could fill in more than one type of drink for each period.
Seven time intervals were defined based on commonly-understood Australian meal windows to aid interpretability of results (Leech et al., Reference Leech, Worsley, Timperio and McNaughton2015): early (6:00–9:00 h), late morning (9:00–12:00 h), midday (12:00–15:00 h), afternoon (15:00–18:00 h), evening (18:00–21:00 h), night (21:00–23:00 h), overnight (23:00–6:00 h). To estimate the proportion of daily food intake consumed during each interval, one point was awarded for a snack, three points for a small meal and five points for a large meal. Water was not awarded any points, but drinks of ‘Alcohol’ or ‘Something else’ were awarded one point, according to accepted methods of including beverages as eating occasions (Kim et al., Reference Kim, DeRoo and Sandler2011; Leech et al., Reference Leech, Worsley, Timperio and McNaughton2015). The number of points consumed during each interval by a participant was divided by their total points consumed that day to calculate the percentage distribution of daily intake across the seven time intervals. This distribution therefore reflected the temporal distribution of daily intake, not nutritional or energy intake.
Participants reported the day of the week they completed the meal pattern chart for. Participants were categorised as weekday (Monday to Friday) or weekend (Saturday or Sunday) reporters.
Mood disorder
Mental health was assessed at CDAH-2 using the lifetime version of the Composite International Diagnostic Interview (CIDI) (World Health Organization, 1997). The computerised CIDI was administered by trained telephone interviewers to collect data on the lifetime prevalence of depressive symptoms, age of onset and age of most recent recurrence. Symptoms were scored using DSM-IV criteria (American Psychiatric Association, 2000) to determine depressive episodes, or dysthymia. Participants, including those who had experienced a mood disorder prior to CDAH-1, were categorised as having a mood disorder only if they had experienced any episodes (first or recurrent) between CDAH-1 and CDAH-2. Sensitivity analyses excluded all participants who had their first mood disorder prior to CDAH-1.
Covariates
At CDAH-1 and CDAH-2, questionnaires collected data on age, marital status (married/living as married, single/separated/divorced), highest education (university, vocational, school), occupational status (professional, non-manual, manual, not in workforce) and current smoking status (never, ex-smoker, smoker). Total weekly minutes of leisure-time physical activity were measured using the validated International Physical Activity Questionnaire long form (Craig et al., Reference Craig, Marshall, Sjöström, Bauman, Booth, Ainsworth, Pratt, Ekelund, Yngve and Sallis2003), and converted to hours per week for interpretability. Parenting status (no children, have children) was determined using the date of birth data for biological children reported at CDAH-2. Social support at CDAH-1 and CDAH-2 was measured using the Henderson Index of Perceived Social Support (potential range 15–75), with a higher score indicating higher self-perceived social support (Henderson et al., Reference Henderson, Duncan-Jones, McAuley and Ritchie1978). At CDAH-2 only, participants reported the hours and minutes of usual sleep duration and the preferred amount of sleep they need to feel they have rested properly. Discrepancy of sleep preference was calculated as preferred minus usual sleep duration. At CDAH-2, participants reported their usual type of work schedule [regular daytime, evening/night/rotating, irregular (e.g. split shift, on call), not employed].
Dietary data were collected using a 127-item FFQ based on a validated FFQ developed for Australian populations (McLennan and Podger, Reference McLennan and Podger1998; Hodge et al., Reference Hodge, Patterson, Brown, Ireland and Giles2000). Diet quality was calculated using a validated Dietary Guidelines Index (DGI) that reflects the 2013 Australian Dietary Guidelines (Wilson et al., Reference Wilson, Blizzard, Gall, Magnussen, Oddy, Dwyer, Venn and Smith2019). A higher score on the scale 0–100 indicated higher diet quality. At CDAH-2, participants were asked how many days per week they usually ate breakfast (range 0–7). Participants were categorised as never skip breakfast, sometimes skip (skip 1–3 days/week) or regularly skip (skip 4–7 days/week).
For CDAH-1 clinic participants, weight was measured to the nearest 0.1 kg in light clothing using Heine portable digital scales (Heine, Dover, NH, USA), and height to the nearest 0.1 cm with a Leicester stadiometer (Invicta, Leicester, UK). Body mass index (BMI) was calculated as weight in kilograms divided by squared height in metres (kg/m2). For CDAH-1 participants who did not attend clinics, and at CDAH-2, BMI was calculated from self-reported height and weight with a correction factor applied. The correction factor was determined based on discrepancies between the self-reported and measured height and weight of CDAH-1 clinic participants (Smith et al., Reference Smith, Gall, McNaughton, Cleland, Otahal, Dwyer and Venn2017).
Transition variables reflect change in circumstance between CDAH-1 and CDAH-2: parenting status (no children, first child born since CDAH-1, additional children born since CDAH-1, same number of children as CDAH-1); marital status (stayed living as married, became living as married, stayed living as single, became living as single); smoking (non-smoker, stopped smoking, started smoking, continued smoking); change in education level (advanced education, same level of education); and change in employment (remained employed, became employed, became unemployed, remained unemployed). For continuous variables (BMI, social support, DGI and leisure-time physical activity), the transition variable was calculated by the value at CDAH-2 minus the value at CDAH-1.
Statistical analyses
All analyses were performed in Stata Version 15 (StataCorp, College Station, Texas, USA, 2017). Time-of-day eating patterns were determined by PCA of the percentages of daily food intake consumed during each time interval (6:00–9:00 h, 9:00–12:00 h, 12:00–15:00 h, 15:00–18:00 h, 18:00–21:00 h, 21:00–23:00 h, 23:00–6:00 h). The number of components was selected based on visual examination of the scree plot, and size of the eigenvalues. Orthogonal varimax rotation was applied to improve the interpretability of the identified components. Bartlett's test of sphericity was used to test whether the variables were unrelated and therefore unsuitable for PCA. The Kaiser–Mayer–Olkin statistic for sampling adequacy was not generated due to singular correlation matrices arising from standardisation of the eating interval variables to sum to one for each participant.
Every participant received a score for each pattern and scores were categorised by tertiles into low, middle and high thirds. A tracking variable for change in pattern scores from CDAH-1 to CDAH-2 was created: consistently low (lowest third of pattern scores at both time-points), decreased (decrease from high or middle to a lower third), consistently middle (middle third at both time-points), increased (increase from low or middle to a higher third) or consistently high (highest third at both time-points). Tracking of pattern scores was determined by examining per cent agreement of the categories and Cohen's κ coefficient for inter-rater reliability (Landis and Koch, Reference Landis and Koch1977). At CDAH-2, per cent agreement was also used to assess the concordance of eating pattern score categories with the reported frequency of eating breakfast.
Multiple imputation was performed to complete the 1985 ASHFS data for missing variables that predicted loss-to-follow-up. Inverse probability weighting on these variables was used in the regression analyses (motivated by Seaman et al., Reference Seaman, White, Copas and Li2012). Firstly, we examined if eating patterns at CDAH-1 predicted the risk of mood disorder during the follow-up period using log binomial regression to calculate relative risks (RR). Secondly, we examined if tracking of eating pattern scores from CDAH-1 to CDAH-2 was associated with the prevalence of mood disorder during the intervening period using log binomial regression to calculate prevalence ratios (PR). Thirdly, to explore bidirectionality, we examined whether experiencing a mood disorder during follow-up predicted eating pattern category at CDAH-2. We used adjacent categories ordered log-link regression to calculate the RR for being in a higher adjacent score category for those who experienced a mood disorder during the follow-up period compared to those who did not (Blizzard et al., Reference Blizzard, Quinn, Canary and Hosmer2013). Males and females were analysed together as there was no evidence of differences by sex in the estimates.
Minimally adjusted models (Model 1) adjusted for sex and age. Purposeful model building procedures were used to determine the fully adjusted models (Model 2) with adjustment for variables thought to be causally associated with the outcome and that changed the coefficient of the principal study factor by at least 10% (Greenland, Reference Greenland1989). Model 2 for the prediction of mood disorder based on CDAH-1 eating pattern adjusted for sex, age, social support, BMI and smoking at CDAH-1. Model 2 for the tracking analyses adjusted for sex, CDAH-2 age and work schedule, and transitions between CDAH-1 and CDAH-2 in social support, marital status, smoking and BMI. Model 2, for the analysis of mood disorder predicting eating pattern at CDAH-2, adjusted for sex and CDAH-2 age, education, BMI, work schedule, parental status, smoking status and self-perceived social support. Model 3 further adjusted for eating pattern category at CDAH-1. Statistical significance was deemed if p < 0.05.
Two PCA sensitivity analyses were conducted to check the robustness of the patterns by stratifying separately by: (1) sex and (2) weekday/weekend. Two separate log binomial regression sensitivity analyses were conducted: (1) excluding weekend reporters; (2) excluding all participants who had experienced a mood disorder prior to CDAH-1.
Results
The meal pattern chart at CDAH-1 was completed by 2853 participants; however, 78 were excluded due to pregnancy. Of the remaining 2775 participants, 1435 completed the meal pattern chart at CDAH-2, with 39 participants excluded for pregnancy. Of the 1396 participants with meal data at both time-points, 1374 also completed the CDAH-2 CIDI. PCA was performed separately on the CDAH-1 and CDAH-2 time-of-eating data for this group. After exclusion of 70 participants missing covariate data, the final sample for regression analyses was n = 1304 (Fig. 1). Participant characteristics are shown in Table 1.
Fig. 1. Childhood Determinants of Adult Health (CDAH) study participant flow chart and related analyses.
Table 1. Participant characteristics by the experience of mood disorder during follow-up (CDAH-1 to CDAH-2)
CDAH, Childhood Determinants of Adult Health study; s.d., standard deviation; BMI, body mass index.
a Henderson Index of Perceived Social Support, possible score range 15–75. A higher score indicates higher self-perceived social support.
b Dietary Guidelines Index, possible score range 0–100. A higher score indicates greater compliance with the 2013 Australian Dietary Guidelines.
c Discrepancy between preferred and usual minutes of sleep per night.
Time-of-day eating patterns
Three similar patterns were obtained at both time-points, cumulatively explaining 65% (CDAH-1) and 64% (CDAH-2) of the variation in the timing of daily food intake. Factor loadings, which indicate the strength of association between the variable and component, and scree plots are shown in online Supplementary Table S1 and Fig. S1, respectively.
Bartlett test of sphericity results for CDAH-1 and CDAH-2 was p < 0.001. Sensitivity analyses of PCA on subgroups male, female, weekday and weekend day produced the same three dominant patterns, with similar loadings to whole-of-group patterns (data not shown).
The mean percentages of daily intake consumed at each of the seven time intervals by those in the highest third of pattern scores were examined to further describe and name the patterns as Grazing, Traditional and Late (Fig. 2). Those high on the Grazing pattern had intake spread across the day from 6:00–18:00 h and consumed the highest average percentage of their daily food intake during the afternoon 15:00–18:00 h. The Traditional pattern was characterised as three main intakes, with the largest mean percentages reflecting breakfast, lunch and dinner times. The Late pattern was characterised by low intake during 6:00–9:00 h, with slightly higher mean percentages of intake during the night and overnight periods than the other patterns.
Fig. 2. Mean percentage of daily intake by eating interval* among participants scoring in the highest third of each time-of-day eating pattern at CDAH-1 and CDAH-2.
There was evidence of tracking of participant scores for all patterns from CDAH-1 to CDAH-2, with participants more likely to be in the same score category at CDAH-2 than the two other score categories (online Supplementary Table S2). For example, of the 33.4% of participants who were in the highest third of the Late pattern at CDAH-1, 16.0% were also in the highest third at CDAH-2, 8.6% in the middle third and 8.8% in the lowest third.
Only the Late pattern was associated with skipping breakfast. Of the 426 participants in the highest third of the Late pattern who had breakfast frequency data, 239 (56.1%) reported skipping breakfast at least once per week (online Supplementary Table S3).
Associations between eating patterns and mood disorder
Time-of-day eating patterns at CDAH-1 were not significantly associated with mood disorder outcomes during the 5-year follow-up (Table 2). A borderline significant increased risk for those in the highest compared to the lowest third of the Late pattern [RR = 1.33; 95% confidence interval (CI) 0.97–1.83] was attenuated in Model 2 (RR = 1.13; 95% CI 0.82–1.55).
Table 2. Associations between time-of-day eating pattern category at CDAH-1 or tracking of eating pattern category from CDAH-1 to CDAH-2, with mood disorder during follow-up between CDAH-1 and CDAH-2
CDAH, Childhood Determinants of Adult Health study; RR, relative risk; PR, prevalence ratio; CI, confidence interval.
Statistically significant (p < 0.05) results are highlighted in bold.
a Prediction analysis models adjusted for sex and age at CDAH-1.
b Tracking analysis models adjusted for sex and age at CDAH-2.
c Prediction analysis models adjusted for sex and CDAH-1 age, BMI, social support and smoking status.
d Tracking analysis models adjusted for sex, age and work schedule at CDAH-2, and change from CDAH-1 to CDAH-2 in social support, smoking, marital status and BMI.
Associations between pattern score tracking categories from CDAH-1 to CDAH-2 and mood disorder during follow-up are also shown in Table 2. After adjustment, compared to those in the consistently low category of the Late pattern, there was a higher prevalence of mood disorder among those in the increased (PR = 1.85; 95% CI 1.11–3.09) and consistently high (PR = 2.04; 95% CI 1.20–3.48) categories. A significant trend for the Late pattern was observed, with higher pattern category associated with a higher prevalence of mood disorder. Indications of higher prevalence of mood disorder among those in the consistently high category of the Grazing pattern and lower prevalence among those in the consistently high category of the Traditional pattern were not statistically significant.
Results for the analysis of mood disorder predicting eating pattern scores are presented in Table 3. After adjustment for covariates, participants who experienced a mood disorder during the follow-up period had a 7% increased risk (RR = 1.07; 95% CI 1.00–1.14) of being in a higher adjacent score category (e.g. high rather than middle, or middle rather than low), compared to participants who had not experienced a mood disorder during follow-up. Having a mood disorder during follow-up was not associated with the Grazing or Traditional patterns at CDAH-2.
Table 3. Relative risk of being in a higher score category of CDAH-2 eating pattern for participants who experienced a mood disorder during follow-up between CDAH-1 and CDAH-2, compared to participants who did not experience a mood disorder during follow-up
CDAH, Childhood Determinants of Adult Health study; RR, relative risk; CI, confidence interval.
Statistically significant (p < 0.05) results are highlighted in bold.
a Model 1: adjusted for sex and CDAH-2 age.
b Model 2: adjusted for sex and CDAH-2 age, BMI, education level, work schedule, parenting status, smoking status and social support.
c Model 3: model 2 plus additional adjustment for eating pattern category at CDAH-1.
Results of the sensitivity analyses are presented in the online Supplementary Tables S4 and S5. Among participants who experienced their first mood disorder between CDAH-1 and CDAH-2, those in the consistently high category of the Late pattern had a higher prevalence of mood disorder compared to those in the consistently low category (PR = 2.84; 95% CI 1.06–7.58). For the Traditional pattern, compared to those in the lowest category at both time-points, a lower prevalence of mood disorders during the follow-up period was observed among those in the consistently middle category (PR = 0.34; 95% CI 0.12–0.99), and the consistently high category (PR = 0.31; 95% CI 0.11–0.87). After excluding weekend reporters, compared to those in the lowest category of the Late pattern at both time-points, those in the increasing (PR = 2.30; 95% CI 1.01–5.24) and consistently high categories (PR = 3.46; 95% CI 1.47–8.14) had an increased prevalence of mood disorder during follow-up. Those who increased their Grazing pattern score category between follow-ups also had a higher prevalence of mood disorders during follow-up (PR = 2.67; 95% CI 1.19–5.99) compared to those in the consistently low category.
Discussion
Three distinct time-of-day eating patterns were identified. The Traditional pattern described a conventional eating schedule of breakfast, lunch and dinner, and the Grazing pattern had intake spread more evenly across the daytime hours. The Late pattern was characterised by low intake in the early morning (6:00–9:00 h) but higher intakes late morning, indicating skipped or delayed breakfast, and proportionally more food consumed during the evening and night than the other patterns. High compared to low scores on the Late pattern at both time-points were associated with a higher likelihood of experiencing a mood disorder, and a nearly three times higher prevalence of first ever onset of a disorder during the intervening 5-year period. However, there was also weak evidence of bidirectionality, with mood disorder during follow-up associated with slightly increased risk of being in a higher Late pattern score category at CDAH-2. Participants who consistently scored in the middle or highest third of the Traditional pattern had a lower prevalence of the first onset of mood disorder during the follow-up period. These results suggest that a more traditionally structured pattern of eating may be associated with better mental health.
Preference for a later-in-the-day style of eating could be a biological or social trait that is implicated in, or predisposes an individual to, poorer mental health. Chronotype characteristics relating to the difference in preference for morning or evening activity may contribute to the observed associations. Evening chronotypes are more likely to skip or delay breakfast, consume higher intakes of food later in the day compared to morning types (Meule et al., Reference Meule, Roeser, Randler and Kübler2012; Roßbach et al., Reference Roßbach, Diederichs, Nöthlings, Buyken and Alexy2018), and have a higher risk of major depressive disorder (Antypa et al., Reference Antypa, Vogelzangs, Meesters, Schoevers and Penninx2016; Au and Reece, Reference Au and Reece2017). It is suggested that preference for evening activity may be a pre-existing trait of the individual rather than symptom of mood disorders (Drennan et al., Reference Drennan, Klauber, Kripke and Goyette1991; Hidalgo et al., Reference Hidalgo, Caumo, Posser, Coccaro, Camozzato and Chaves2009). A later pattern of eating may precede onset of mood disorders, and contribute to ‘social jetlag’ which has been associated with depressive symptoms (Levandovski et al., Reference Levandovski, Dantas, Fernandes, Caumo, Torres, Roenneberg, Hidalgo and Allebrandt2011). Social jetlag refers to a discrepancy between biological and social or work schedules, where evening chronotypes are unable to fulfil their sleep timing preferences (Wittmann et al., Reference Wittmann, Dinich, Merrow and Roenneberg2006). In our cohort, a larger mean discrepancy between preferred sleep and actual sleep times at CDAH-2 was reported by participants who experienced a mood disorder (46 min) than those with no mood disorder (33 min). However, the amount of reported usual sleep was very similar at 7 h 22 min for those who had experienced a mood disorder compared to 7 h 25 min for those who had not. Usual sleep duration and sleep preference were not included in our adjusted models as they did not have sufficient effect on the prevalence estimates after inclusion of other covariates.
There were indications of bidirectionality, as participants with mood disorders during follow-up were slightly more likely to be in a higher score category of the Late pattern at CDAH-2 compared to participants who had not experienced a mood disorder. Mood disorders may influence lifestyle and dietary behaviours, but this does not preclude the influence of chronobiology. Mood disorders and emotional stress may reduce capacity to adhere to morning or daytime work/life schedules, or what are considered favourable behaviours such as making healthy food choices (Lopresti et al., Reference Lopresti, Hood and Drummond2013). Therefore, bidirectionality and the concept of social jetlag and chronobiology should be considered when exploring the nexus between diet, time-of-day eating patterns and mood disorders.
Our results concerning the Late pattern complement existing literature reporting cross-sectional associations between skipping breakfast and depressive symptoms (Fulkerson et al., Reference Fulkerson, Sherwood, Perry, Neumark-Sztainer and Story2004; Lien, Reference Lien2007; O'Sullivan et al., Reference O'Sullivan, Robinson, Kendall, Miller, Jacoby, Silburn and Oddy2009; Lee et al., Reference Lee, Han and Kim2017a, Reference Lee, Park, Ju, Lee, Han and Kim2017b; Kwak and Kim, Reference Kwak and Kim2018). However, ‘breakfast’ has often been poorly defined or not defined at all (Szajewska and Ruszczynski, Reference Szajewska and Ruszczynski2010) making it difficult to determine whether associations are due to not eating a morning meal, or delaying first consumption until later in the morning. In the current study, the Late pattern is likely to reflect both skipped and delayed breakfast. Participants who scored highly on the Late pattern had greater intake during late morning (9:00–12:00 h) compared to other patterns, and more than half of these participants reported they usually skipped breakfast at least once per week. Although this demonstrates the need for clarification around what constitutes breakfast, previous studies examining various concepts of ‘skipping breakfast’ have highlighted the physiological and hormonal mechanisms that could explain the associations between omitting or delaying breakfast and mood disorders. Skipping breakfast has been shown to be associated with poorer diet quality and obesity which may affect mood due to long-term nutritional imbalance as well as metabolic co-morbidities (Smith et al., Reference Smith, Gall, McNaughton, Blizzard, Dwyer and Venn2010; Szajewska and Ruszczynski, Reference Szajewska and Ruszczynski2010; Horikawa et al., Reference Horikawa, Kodama, Yachi, Heianza, Hirasawa, Ibe, Saito, Shimano, Yamada and Sone2011). Eating breakfast lowers cortisol levels so skipping or delaying this meal may affect mood due to higher levels of cortisol and immune system dysregulation (Witbracht et al., Reference Witbracht, Keim, Forester, Widaman and Laugero2015; Lee et al., Reference Lee, Park, Ju, Lee, Han and Kim2017b). Lower appetite for breakfast first thing in the morning could also indicate reduced levels of the appetite regulating hormone ghrelin. Ghrelin has been shown to have an anti-depressant effect in mice (Lutter et al., Reference Lutter, Sakata, Osborne-Lawrence, Rovinsky, Anderson, Jung, Birnbaum, Yanagisawa, Elmquist, Nestler and Zigman2008) and affect plasma cortisol (Kluge et al., Reference Kluge, Schüssler, Dresler, Schmidt, Yassouridis, Uhr and Steiger2011). Proximity of the last eating occasion can influence the amount of food consumed at the following eating occasion, so higher intake at night may result in a less subsequent hormonal drive to eat early the next day. People with night eating syndrome (NES), typified by >50% of daily calorie intake during the evening and waking at night to eat, have been shown to have lower ghrelin levels than controls during the early morning period to 9:00 h (Allison et al., Reference Allison, Ahima, O'Reardon, Dinges, Sharma, Cummings, Heo, Martino and Stunkard2005). We do not suggest that participants who scored high on the Late pattern meet criteria for NES, but later eating combined with skipping breakfast could be eating practices that warrant further attention.
Associations between the Grazing pattern and mood disorder only reached statistical significance in the sensitivity analyses excluding weekend reporters, with those who increased their score category between CDAH-1 and CDAH-2 having a 2.7 times higher prevalence of mood disorder during the follow-up period. The Grazing pattern's spread of food intake across daytime hours could represent snacking type behaviour and varied eating schedules. Irregular meal schedules, including skipped meals, snacking and delayed lunch, have been associated with unfavourable health outcomes including obesity, depressed mood and hypertension (Gill and Panda, Reference Gill and Panda2015; Furihata et al., Reference Furihata, Konno, Suzuki, Takahashi, Kaneita, Ohida and Uchiyama2018; Leech et al., Reference Leech, Timperio, Worsley and McNaughton2019).
Consistently high scores on the Traditional pattern, characterised by distinct meal times, was associated with a non-statistically significant lower prevalence of mood disorder during follow-up. Furthermore, in the sensitivity analyses, high scores on the Traditional pattern at CDAH-1 was associated with a lower risk of first ever onset of mood disorder during follow-up. Structured and regular meal times may indicate healthier behaviours. In a previous study, healthier lifestyle behaviours were protective against mood disorders among the CDAH cohort (Gall et al., Reference Gall, Sanderson, Smith, Patton, Dwyer and Venn2016).
Limitations of this study include potential bias as the meal pattern chart was reliant on recall and only covered a single 24-h period at each time-point which may not reflect usual eating patterns. However, there was evidence that the pattern scores tracked from CDAH-1 to CDAH-2, indicating possible habituality of time-of-day eating. There was no guidance given to participants about entering multiple meal types in the same hourly period, or which time period they should use when entering food or drink consumed on the hour (e.g. whether a drink at 7:00 h should be entered as 6:00–7:00 h or 7:00–8:00 h). The 23:00–6:00 h period meant there was no differentiation between overnight eating and an early breakfast. Bias from loss to follow-up between the nationally representative baseline youth sample and the adult surveys may limit the generalisability of our results. However, there was wide variation in the characteristics of participants in the adult sample and loss-to-follow-up was mitigated by inverse probability weighting. There is also the possibility of bias from misreporting of covariate measures, such as self-reported weight (mitigated by using a correction factor) and physical activity; or unmeasured confounding such as lifestyle (e.g. work schedule or sleep hours at CDAH-1) or psychological factors.
Strengths of the study include the use of the CIDI, which is considered the ‘gold-standard’ measure for retrospective assessment of the history of mental disorders in epidemiological studies (Steel et al., Reference Steel, Marnane, Iranpour, Chey, Jackson, Patel and Silove2014). Participant recollection may have resulted in some misreporting. However, the time-related questions in the CIDI around the first and last occurrence of a disorder have shown good reliability (Wittchen, Reference Wittchen1994). Although misreporting of snack and beverage intake is common in dietary surveys, primarily as under-reporting (Poslusna et al., Reference Poslusna, Ruprich, de Vries, Jakubikova and van't Veer2009), converting each individual's eating occasion to a proportion of their total intake may have helped address systematic misreporting by individuals, or variation in concepts of snack or meal sizes between participants. The assessment of BMI, overall diet quality and physical activity as covariates in our models considered potential confounding or mediation from energy and nutritional aspects of diet. Diet quality and physical activity did not change the coefficients sufficiently to be included in our models, indicating they were not confounding measures. The sensitivity analyses on the PCA and regression analyses confirmed that the patterns and associations were robust to influence of factors such as sex, prior mood disorder and differences between weekday and weekend eating practices. Another strength is the novel application of PCA to derive patterns that capture dietary behaviours, and in the case of the Late pattern, multiple behaviours of skipping breakfast and eating later into the evening. Furthermore, the longitudinal design builds on existing cross-sectional research.
Longitudinal studies that replicate the eating patterns observed in this study, or specifically examine clustering of several habits, may be useful in determining lifestyle and chronobiological influences on mood disorders. Repeat measures and more detailed information about the timing and size of meals would help determine the nature of the relationship between eating patterns and mental health outcomes.
In conclusion, delaying or skipping breakfast and eating higher proportions of intake later in the day may be an unhealthy behaviour associated with a higher likelihood of mood disorder among adults. Whereas more traditional eating patterns of main meals at breakfast, lunch and dinner may be associated with a lower likelihood of mood disorder over time. These relationships may be bidirectional, and a pre-existing preference for certain eating patterns due to chronobiological traits of the individual should be considered.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291719002800.
Acknowledgements
We gratefully acknowledge the Australian Council for Health, Physical Education and Recreation and contributions of the CDAH study's project manager Ms Marita Dalton, all other project staff and the study participants.
Financial support
The CDAH study was funded by the National Health and Medical Research Council (grant number 211316), the National Heart Foundation of Australia (grant number GOOH0578), the Tasmanian Community Fund, and Veolia Environmental Services. Additional sponsorship was received from Sanitarium, ASICS and Target. The authors are supported by the National Health and Medical Research Council Early Career Fellowship (K.J.S., grant number APP1072516), National Heart Foundation of Australia Future Leader Fellowships (C.G.M., grant number 100849), (S.L.G., grant number 100446) and National Heart, Lung and Blood Institute (T.D., grant number R01 HL121230-01A1). The funding bodies and sponsors had no role in the study design, conduct, analysis or reporting of results.
Conflict of interest
None.
