Association between sleep timing and meal and snack patterns in schoolchildren in southern Brazil

Abstract

This study aimed to identify meal and snack patterns and assess their association with sleep timing in schoolchildren. This is a cross-sectional study carried out in 2018/2019 with 1333 schoolchildren aged 7–14 years from public and private schools in Florianópolis, Brazil. Previous-day dietary intake data for breakfast, mid-morning snack, lunch, mid-afternoon snack, dinner and evening snack were collected using a validated online questionnaire. Sleep timing was measured by the midpoint of sleep and classified as quartiles (very early, early, late and very late). Latent class analysis was performed to identify meal and snack patterns, and multinomial logistic regression was used to assess associations. Students with very late sleep timing were less likely to consume the ‘coffee with milk, bread and cheese’ breakfast pattern compared with very early group. Also, the former were more likely to consume the ‘mixed’ breakfast pattern (healthy and unhealthy foods) compared with very early students. The latter were more likely to eat the ‘Brazilian traditional, processed meat, egg and fish’ lunch pattern to the late students and less likely to consume the ‘pasta and cheese’ lunch pattern compared with the students with later sleep timing. Students with later sleep timing were more likely to eat ultra-processed food at mid-afternoon snacks compared with early group. The study findings suggest that morning preference appears to promote healthier breakfast, lunch and afternoon snack patterns, whereas later sleep timing may pose challenges in maintaining healthy patterns at these meals/snacks.

Circadian rhythms are biological rhythms that modulate physiological functions within the body, including the regulation of sleep⁽¹⁾. Sleep timing represents the hours of the day when sleep occurs⁽²⁾ The midpoint of sleep (MSF) is a crucial marker for both sleep timing and circadian phase and is widely used in sleep research due to its association with health outcomes^(3–6). MSF has a strong correlation with Dim Light Melatonin Onset⁽⁷⁾ and identifies circadian preferences by considering individual bedtime and wake-up times, thereby categorising them into earlier or later preferences⁽⁴⁾. Usually, children have a morning preference, and in adolescence, they become progressively later until around 20 years of age. After the age of 20, there is a tendency to return to the morning chronotype with increasing age⁽⁸⁾. This fact is relevant to the discussion of school schedules and explains why starting school later can be beneficial for adolescents^(9,10) .

Evidence suggests that later sleep timing is associated with poorer health indicators as poorer emotional regulation, lower cognitive function/academic achievement, shorter sleep duration/poorer sleep quality, lower physical activity levels and more sedentary behaviours, including unhealthy dietary intake in children and adolescents⁽¹¹⁾. Thus, cross-sectional studies found that late sleep timing correlates with unhealthy eating habits such as increased consumption of unhealthy foods^(12–14), high-energy-dense foods, breakfast skipping^(15,16) and lower intake of fruits and vegetables^(15,17) . Some studies have identified that the impact of late sleep habits on food consumption is mainly related to the increased consumption of more caloric and ultra-processed foods due to the convenience of consuming them at night, as most of these foods are ready to eat or can be prepared quickly^(12,18–22) . The multicentre study by Chaput et al. (2018)⁽¹⁴⁾ conducted with 5873 children between 9 and 11 years old, investigated the association between bedtime and the consumption of sugary drinks. The authors observed a positive association between late bedtime and regular consumption of soft drinks. Similarly, Agostini et al. (2018)⁽²¹⁾ identified that Australian students between 9 and 17 years old who went to bed later (after 9 pm) were likelier to consume fast food five or more times per week. Inadequate sleep routine, such as lack of consistency in bedtime and wake-up times, has been associated with lower consumption of fruits, vegetables, whole grains and higher intake of sugars and meats⁽²²⁾. On the other hand, earlier sleep timing has been linked to healthier behaviours such as eating meals earlier and eating natural, minimally processed foods⁽²³⁾ and a reduced likelihood of schoolchildren being overweight⁽²⁴⁾.

Chrono-nutrition is used to describe the integration of circadian rhythms research into nutrition investigation, as the circadian system is influenced by food intake^(25–27). A study with 21020 Brazilian adults from the 2008–2009 Household Budget Survey found that late meal intake was associated with overweight and obesity⁽²⁸⁾. The studies of global dietary patterns (GDP) and meal and snack patterns (MP and SP) have been used to describe the combinations of food eaten in a day or in specific eating occasions, respectively^(29,30) . This approach offers insights into the actual eating behaviours of a population and facilitates the acquisition of practical information for formulating nutritional recommendations^(31,32) . Investigating eating occasions has emerged as an avenue for exploring associations with diseases or risk factors, aiding in the development of nutritional guidelines and enhancing dietary advice, thus providing tangible benefits to the population in terms of meal planning and preparation of different meals/snacks throughout the day^(30,33) . Additionally, the use of a person-centred approach to derive MP and SP, such as latent class analysis (LCA), involves classifying individuals into distinct groups or classes and estimating the probabilities for each indicator, thus facilitating interpretations and comparisons with characteristics of interest⁽³⁴⁾.

The studies investigating the relationship between GDP and sleep timing in children and adolescents have consistently shown that later bedtimes and short sleep duration are associated with unhealthy GDP^{(12,17,35,36)} , indicating that sleep habits can impact dietary intake. Despite this, limited research has examined the influence of sleep on meal and snack composition, and the results suggest that late sleep timing is associated with skipping breakfast^(13,21,37) . However, the impact of sleep timing on the composition of meals and snacks remains unexplored.

Considering the emerging field of research on the interplay between sleep and meal and snack patterns, particularly in children and adolescents, it is crucial to recognise this developmental stage as pivotal for the development of both sleep and dietary behaviours⁽³⁸⁾. Therefore, the objective of this study was to identify meal patterns (MP) and snack patterns (SP) among students aged 7–14 years and assess their association with sleep timing. We hypothesised that those with later sleep timing are more likely to consume unhealthy meal/snack patterns compared with schoolchildren with early sleep timing.

Methods

Study design and sample

This is a cross-sectional study on overweight and obesity prevalences in schoolchildren aged 7–14 years in the city of Florianopolis, Southern Brazil, which is part of a larger longitudinal study entitled EPOCA. The latter has investigated time trends in these prevalences and associated factors among students enrolled in primary education in both public and private schools. The study surveys were conducted in four waves in 2002, 2007, 2012/2013 and 2018/2019. The latter was carried out between November 2018 and December 2019 and composed the sample of the present study.

The study population was made up of children and adolescents of both sexes aged between 7 and 14 who studied in public and private schools in the city of Florianópolis. The city has 82 schools (53 public and 29 private) and 34 318 students enrolled (23 883 in public schools and 10 435 in private schools)⁽³⁹⁾. More details about the sample calculation and sampling have been published in Pereira et al., (2022)⁽⁴⁰⁾.

A total of 6118 students from thirty schools (nineteen were public and eleven private were invited to participate in the research and received the Free and Informed Consent Form (FICF) so that their parents or guardians could authorise their participation in the study⁽⁴⁰⁾.

The inclusion criteria included being present in school on the day of data collection, delivering the signed FICF and the student himself/herself signing the Free and Informed Assent Term at the beginning of the data collection. Body weight and height data and food consumption were collected from 1671 students, but implausible reports were excluded. The latter included reporting no food consumption whatsoever (n 188) and those who presented implausible food consumption data (n 87). The latter consisted of reporting >3 food items per day or the number of items that exceeded three times the standard deviation of the average consumption, assuming a Poisson distribution for reports on the frequency of food consumption⁽⁴¹⁾. Also, all students without sleep data were excluded (n 63) resulting in the analytical sample of 1333 students between 7 and 14 years old (online Supplementary Fig. S1).

This study was conducted in accordance with the guidelines of the Code of Ethics of the World Medical Association (Declaration of Helsinki) and approved by the Human Research Ethics Committee of the Federal University of Santa Catarina (UFSC, protocol number 7 539 718·1·0000·0121).

Sleep data

The questions about sleep were adapted from the School Sleep Habits Survey (Bradley Hospital, 1994) and answered by the parents or guardians. The questionnaire included the following questions: ‘What time does the child usually go to sleep at night on the days they go to school? What time does the child usually wake up in the morning on the days they go to school? What time does the child usually go to bed at night on weekends (days when they don’t go to school)? What time does the child usually wake up in the morning on weekends (days when they don’t go to school)?’. Allowed responses contained hours and minutes (local time), whereas sleep latency and the wake time during the night were not measured. Total sleep time was calculated as the difference between bedtime and wake-up time. The midpoint of sleep (MSFsc) was used as a measure of sleep timing and calculated considering the midpoint between bedtime and wake-up time on non-school days corrected by sleep debt^(8,42) . More details on the MSFsc calculation are available in a previous report⁽²⁴⁾. The MSFsc was divided into quartiles (Q1–Q4 in ascending order), with the first quartile (Q1) representing those with lower MSFsc or very early sleep preference/timing and the fourth quartile (Q4) those with higher MSFsc or very late sleep preference. The second quartile (Q2) represents those with early preference and the third (Q3) those with late sleep preference. The median (interquartile range) cut-off points in local time were, respectively, 2:53 am (2:30–3:04), 3:38 am (3:28–3:47), 4:10 am (4:07–4:28) and 5:22 am (5:00–5:59).

Assessment of dietary intake, physical activity and screen use

Dietary intake and meal and snack definitions

Data on frequency of food consumption, physical activities and screen use from the previous day were obtained from the Food Consumption and Physical Activity for Schoolchildren (Consumo Alimentar e Atividade Física de Escolares, Portuguese acronym Web-CAAFE) questionnaire. This is a web-based, self-report questionnaire developed to monitor food consumption and physical activity in the school environment. The team of trained researchers accompanied the students as they responded to the web-CAAFE⁽⁴³⁾.

The questionnaire was considered adequate in a reproducibility test⁽⁴⁴⁾ and in usability tests⁽⁴³⁾. The average application time was 14 min in the usability test study⁽⁴⁵⁾. The Web-CAAFE food consumption section was validated in two studies with children⁽⁴⁶⁾ and adolescents⁽⁴⁷⁾, both of which used direct observation of school meals as a reference method.

Web-CAAFE is divided into three sections: registration, food consumption section and physical activities and sedentary behaviors section. The food consumption section is a previous-day recall of the intake on three meals (breakfast, lunch and dinner) and three snacks (mid-morning snack, mid-afternoon snack and evening snack), presented in chronological order without specifying the time of eating events (discussed in section below). For each meal and snack, the questionnaire presents a list of thirty-one pre-defined icons of healthy and unhealthy food items for the student to select the items that were consumed in the respective eating event on the previous day: water, rice, vegetables, green leaves, vegetable soup, beans, manioc flour, pasta, instant pasta, French fries, beef/poultry, eggs, fish/seafood, maize/potatoes, sausage, breakfast cereal, fruits, bread/biscuits, cheese bread, cake without icing, cheese, coffee with milk, milk, yogurt, chocolate milk, fruit juices, cream cookies, soda, sweets (chocolate bars, ice cream, candies and cake with icing), chips and pizza/hotdog/hamburger.

On the day of data collection, students received instructions from trained researchers on how to complete the Web-CAAFE and the definitions of meals and snacks. The questionnaire includes six eating events ordered chronologically and presented sequentially on the screen (breakfast, mid-morning snack, lunch, mid-afternoon snack, dinner and evening snack). It is not possible to include additional meals or snacks. The presence of an animated character (avatar) helped identify which meal was being questioned at that moment, placing it at the time of day in which it occurs, using quick definitions. For breakfast, the avatar explains: ‘Breakfast is the first meal we have the day after waking up’. For the mid- morning snack: ‘It’s what you ate after breakfast and before lunch’. Lunch is considered the meal that takes place in the middle of the day. The mid-afternoon snack is explained as: ‘is what you ate after lunch and before dinner’. Dinner: ‘is the main meal we have at night’. The evening snack: ‘is what you ate after dinner and before bed’. These sentences are repeated for each meal and snack. At the end of each eating event, the avatar explains ‘Remember, if you didn’t eat anything, click on the “nothing” button’ ⁽⁴⁵⁾. Thus, when the student reported consuming at least one food item (except water), the meal or snack was considered as meal or snack consumed.

The Web-CAAFE is a qualitative questionnaire based on the daily frequency of consumption of food items considered healthy and unhealthy foods and does not allow identifying the exact time of meal or snack consumption or the amount of food intake⁽⁴⁰⁾. Therefore, each meal or snack had its consumption frequency based on the consumption or non-consumption of each food item, assuming that each item can be selected once at each eating event. Each student answered the questionnaire once.

The Demo version of the web-CAAFE, including English subtitles, is available on http://caafe.ufsc.br/portal/10/detalhes.

Physical activity

The physical activities/sedentary behaviours section are divided into morning, afternoon and evening, with thirty-two drawings depicting the activities they carried out on the previous day: basketball/volleyball, catch, soccer, running, martial arts, tennis, dancing, table tennis, marbles, hopscotch, rope jumping, gymnastics, swimming, cycling, rollerblading/skateboarding, surfing, kite flying, dodgeball, hide-and-seek, playing with the dog, studying/reading/drawing, board games, playing with dolls, playing with toy cars, watching TV, listening to music, using a smartphone/tablet, using a computer, playing videogames, doing the dishes and sweeping the floor.

The physical activity score (PAS) was calculated considering the frequency of twenty-eight physical activities (except screen-based activities) and the Compendium of Energy Expenditures for Youth⁽⁴⁸⁾. Metabolic equivalent values were considered for each physical activity and multiplied by the daily frequency (ranging from 0 to 3). The PAS was obtained by summing the scores across all physical activities and subsequently categorised into tertiles⁽⁴⁹⁾.

Screen use

Daily frequency of screen use was described for each period of the day based on the following activities: watching television, using a computer, using a smartphone/tablet and playing video games, in the three periods of the day and categorised into never, once a day, twice a day and more than three times a day. The information on duration of physical activity or screen time we not measured. Web-CAAFE was applied in the school environment and took place from Monday to Friday during both morning and afternoon shifts. Therefore, the data on food consumption, physical activities and sedentary behaviour were all obtained from Sunday to Thursday and on different days of the week.

Analysis of meal and snack patterns

Meal and snack patterns were identified using LCA, which groups individuals based on their probabilities of class membership⁽³⁴⁾. In LCA, the consumption of at least one food item in the meals or snacks were considered, unless students selected only water. All thirty-one food items were included in the LCA in each eating event, except the maize/potatoes item, which was excluded from breakfast, as no schoolchildren selected it.

The models were evaluated considering Akaike information criterion, sample size-adjusted Bayesian information criterion, entropy, the Lo–Mendell–Rubin test and the percentage of schoolchildren allocation in classes (online Supplementary Table1). Lower Akaike information criterion and size-adjusted Bayesian information criterion values and higher entropy values indicate better-adjusted models^(50,51) .

To identify the food items belonging to each meal or snack pattern, the ratio of the average frequency of consumption of each food item and Confidence Intervals of 95 % (95 % CI) were calculated by dividing item class average frequency of consumption of the item in the class and divided by the overall (all classes) average consumption of the meal/snack. An ratio of the average frequency of consumption value whose 95 % CI does not include the value of one was used as the criterion for inclusion of the food item in the MP or SP^(52,53) .

The meal and snack patterns were named according to the food items that compose them and the recommendations of the Dietary Guidelines for the Brazilian Population⁽⁵⁴⁾. Those patterns that described combinations of foods traditionally consumed in Brazil were named ‘traditional Brazilian’. Patterns that contained more ultra-processed foods were named as ‘ultra-processed’. Those patterns that contained both healthy foods and unhealthy foods were named as “mixed”⁽⁵³⁾.

Anthropometric measurements and socio-economic data

Weight and height measurements were performed at school by trained researchers according to standardised protocols (International Society for the Advancement of Kinanthropometry ISAK)⁽⁵⁵⁾. Weight was measured with a portable digital scale (Marte, model LS200P, 200 kg maximum capacity, 50 g precision). A portable stadiometer (AlturExata, 2·13 m of maximum capacity and 1 mm precision) was used for height. The body mass index (BMI) was calculated as weight (kg) divided by height squared (m). Age-and sex-specific BMI z-scores were calculated according to the World Health Organization (WHO) criteria for children and adolescents aged 5–19 years⁽⁵⁶⁾. The weight status was categorised into non-overweight (underweight and normal weight, BMI z-score for age < +1) or overweight including obesity (BMI z-score for age ≥ +1).

School management provided information on the students’ dates of birth, classes, school shift (morning or afternoon) and type of school (public or private). Maternal education was self-reported by parents or guardians and classified into three categories according to years of study (0–8, 9–11 and ≥ 12 years of study), thus corresponding to primary, secondary and university educational levels.

Statistical analysis

Sample characteristics were described as absolute and relative frequencies, 95 % CI for categorical variables and mean or median and interquartile range (p25–p75) for continuous variables. The differences in categorical variables between MSFsc quartiles were analysed using Pearson’s χ² test, whereas Kruskal–Wallis test was applied to investigate MSFsc differences. A statistical significance level of P < 0·05 was used as a cut-off point for the type I error.

The association between quartiles of midpoint of sleep (main exposure variable) and meal or snack patterns (dependent variable) was calculated for each eating event using multinomial logistic regression analysis adjusted for the following exposure variables: sex, age group (7–10 or 11–14 years), type of school (public or private), weight status (non-overweight or overweight including obesity), physical activity score tertiles, daily frequency of screen use (never, once a day, twice a day and more than three times a day), maternal education (0–8, 9–11 or ≥12 years of schooling), day of food intake report (weekday or weekend) and school shift (morning or afternoon). The adjustment variables were selected considering their relationship with the outcome and exposure and were therefore included in the models: physical activity⁽⁵⁷⁾, use of screens⁽⁵⁸⁾, sex and age⁽⁴⁾, type of school and maternal education (family income proxy)⁽⁴¹⁾. Also, in order to assess whether there are differences in the association according to the school shift the child/adolescent studies, the analyses were stratified by this variable.

Marginal distributions for each MP and SP were presented in terms of predicted probabilities with the corresponding 95 % CI adjusted for all exposure variables (using Stata command ‘margins’). Statistically significant differences were detected by non-overlapping 95 % CI of the marginal effects.

Stata^® version 14·0 (StataCorp LLC) was used for descriptive analysis and multinomial regression, whereas Mplus^® version 6·12 was used for LCA. The analyses were adjusted considering the survey design effect (using the Stata command ‘svy’).

Results

The analytical sample consisted of 1333 children aged 7–14 years. For each meal or snack, only the children who reported a plausible food consumption were considered, thus resulting in different sample sizes (online Supplementary Fig. 1). Table 1 presents key characteristics of the study sample. Most of children were female (53·1 %), between 7 and 10 years old (57·8 %) and studied in the morning shift (52·4 %). About a third (33·8 %) of the sample were overweight (including obesity). Most reports referred to weekdays (87·6 %). Lunch was the most frequent meal consumed (97·8 %), followed by dinner (92·2 %), mid-afternoon snack (88·0 %), breakfast (83·0 %), mid-morning snack (58·7 %) and evening snack (54·8 %). A higher proportion of public school students shared the fourth quartile of MSFsc compared with the first quartile (65·8 % v. 54·3 %), whereas a lower proportion of private school students were in the fourth quartile compared to the first quartile (34·2 % v. 45·7 %). The first quartile of MSFsc had higher proportion of morning shift students than fourth quartile (71·8 v. 27·2 %). Breakfast consumption was significantly different across MSFsc quartiles (P = 0·040). Mid-morning snack was more frequently consumed by the children in the first quartile compared with the fourth quartile (71·3 v. 44·6 %). All sleep variables were different between quartiles, except sleep duration on weekend (Table 1). Students aged 11–14 years had lower median of total and weekday sleep duration (9·43, 9·00 v. 10·00, 10·00), later bedtime on weekdays and weekend (10:15 pm, 11:30 pm v. 10:00 pm, 11:00 pm) and earlier wake-up time on weekdays (6:50 am v. 7:30 am) and later on weekend (10:00 am v. 9:10 am) compared with students aged 7–10 years (online Supplementary 1).

Table 1. Description of the sample of 7- to 14-year-old schoolchildren according to midpoint of sleep quartiles (MSFsc group). Florianopolis. Brazil. 2018/2019

Q, quartile; p25; p75: interquartile range; MSFsc, midpoint of sleep on free days corrected.

† Classified according to WHO (2006).

‡ Pearson’s χ² test.

^§ Kruskall–Wallis test. Bold values denote statistical significance at the P < 0.05 level.

^a The contrast between the first and fourth quartile is significantly different at 5 % level.

^b The contrast between the second and third quartile is significantly different at 5 % level.

^c The contrast between the second and fourth quartiles is significantly different at 5 % level.

^d The contrast between the third and fourth quartiles is significantly different at 5 % level.

^e All quartiles are significantly different from each other at a 5 % level.

The most consumed food items for breakfast were bread (50 %), coffee with milk (24 %) and chocolate milk (22 %). During the mid-morning snack, fruits (23 %), water (22 %) and bread (22 %) were most frequently consumed. At lunch, rice (65 %), beef/poultry (55 %) and beans (49 %) were preferred. During mid-afternoon snack, bread (31 %), cream cookies (21 %) and fruits (18 %) were the most popular choices, and for dinner, these were rice (40 %), beef/poultry (32 %) and beans (22 %). During the evening snack, water (33 %), fruits (21 %) and sweets (11 %) were the most selected items (online Supplementary Table 2).

The criteria used to select the best LCA model are provided in online Supplementary Table 3 and resulted in three classes for breakfast, mid-morning snack, dinner and evening snack on the one hand and four classes for mid-afternoon snack and lunch on the other hand. The food items included in patterns in each eating occasion are described in online Supplementary Tables 4–9, respectively.

For breakfast, the most common pattern included 45 % of the sample. It was named ‘coffee with milk, bread and cheese’ due to the high probability of consuming these items. The second breakfast pattern, called ‘mixed’, included 35 % of the sample with a higher probability of consuming water, rice, vegetables, beans, instant pasta, French fries, beef/poultry, fruits, cheese bread, cream cookies, breakfast cereal, yogurt, fruit juice, soda, sweets, chips, Pizza/hot-dog/hamburger and cake. The third breakfast pattern included 20 % of the students and was characterised by a higher probability of consuming chocolate milk (online Supplementary Table 4).

The first mid-morning snack pattern was shared by 69 % of the students and was labelled ‘ultra-processed and fruits’ because of a higher probability of consuming fruits, cream cookies, pizza/hotdog/hamburger and cake. The second mid-morning SP was observed in 24 % of the sample and termed ‘coffee with milk, bread, cheese and processed meat’ due to a higher probability of consuming sausages, bread, cheese, coffee with milk and chocolate milk. The third SP included 9 % of the students and was labelled ‘traditional Brazilian lunch with soda’ due to preference for consuming rice, vegetables, green leaves, vegetable soup, beans, manioc flour, corn/maize, pasta, beef/poultry and soda (online Supplementary Table 5).

The most common lunch MP termed ‘Brazilian traditional’ was observed in 41 % of students, with a higher probability of consuming rice, vegetables, green leaves, beans, manioc flour and beef/poultry. The second lunch MP included 29 % of the sample and was labelled ‘Brazilian traditional, processed meat, egg and fish’ characterised by a higher probability of consuming rice, vegetables, beans, manioc flour, sausages, eggs and fish/seafood. The third lunch pattern comprised 19 % of students and was identified as ‘mixed’ because of mixing both healthy and unhealthy diet markers, such soup, instant pasta, sausages, breads, cheese bread, cream cookies, milk, pizza/hotdog/hamburger and cake. The fourth lunch pattern was labelled ‘pasta and cheese’ due to the predominance of these items and was identified in 11 % of schoolchildren (online Supplementary Table 6).

Most of the students (51.6 %) preferred the ‘ultra-processed’ mid-afternoon SP. It included cheese bread, cream cookies, soda, sweets, chips, pizza/hotdog/hamburger and cake as the most frequent food choices. The second SP was denominated ‘coffee with milk, bread, cheese and processed meat’, with 30.5 % of the children allocated to it, characterised by the preference for sausages, bread, cheese and coffee with milk. The third mid-afternoon SP was termed ‘fruits’, with 14.6 % of the students included. The fourth SP was shared by 3 % of the students and labelled ‘traditional Brazilian lunch’. It was composed of rice, vegetables, green leaves, vegetable soup, beans, manioc flour, corn/maize, pasta, beef/poultry and French fries (online Supplementary Table 7).

The most common dinner patterns labelled ‘mixed’ was identified in 47 % of the sample and indicated a higher probability of consuming vegetable soup, pasta, instant pasta, sausages, fruits, bread, cheese bread, cream cookies, cheese, coffee with milk, chocolate milk and cake. The second pattern included 41·3 % of the schoolchildren and was labelled ‘traditional Brazilian, fish and water’ because of a higher probability of consuming water, rice, vegetables, green vegetable leaves, soup, beans, manioc flour, maize/potatoes, beef/poultry and fish/seafood. The third dinner pattern denominated ‘ultra-processed and sweets’ was identified in 12 % of the sample, who had higher probability of consuming soda, sweets and pizza/hotdog/hamburger (online Supplementary Table 8).

The most common evening SP comprised 64 % of the students and was labelled ‘ultra-processed, sweets, dairy and fruits’ due to a higher probability of consuming fruits, cream cookies, milk, yogurt, chocolate milk, sweets and pizza/hotdog/hamburger. The second evening snack patterns was named ‘water’ due to a high preference for this drink and included 27 % of the sample. The third evening SP labelled ‘traditional Brazilian lunch and ultra-processed foods’ included 9 % of students and indicated a higher probability of consuming rice, vegetables, green leaves, vegetable soup, beans, manioc flour, maize/potatoes, beef/poultry, fish/seafood, instant pasta, French fries, sausages and eggs (online Supplementary Table 9).

The probability of belonging to the meal and snack patterns in each eating occasion across the quartiles of the midpoint of sleep (MSFsc group) is shown in Table 2. There was a statistically significant decrease in the probability of belonging to the ‘coffee with milk, bread and cheese’ breakfast pattern in the fourth quartile compared with the first quartile (35·4 %, 95 % CI 27·2, 43·6 v. 56·0 %, 95 % CI 48·5, 63·4) as well as a significant increase in the probability of belonging to the ‘mixed’ breakfast pattern in the fourth quartile compared with the first quartile (40·0 %, 95 % CI 32·4, 46·7 v. 28·0 %, 95 % CI 23·8, 32·0) (Table 2). As for mid-morning snack, there was a significant increase in the probability of belonging to the ‘traditional Brazilian lunch with soda’ SP in the third quartile compared with the first quartile (7·8 %, 95 % CI 4·6, 11·0 v. 2·0 %, 95 % CI 0·2, 3·9). For lunch, there was a decrease in the probability of belonging to the ‘Brazilian traditional, processed meat, egg and fish’ MP in the third quartile compared with the first quartile (21·5 %, 95 % CI 15·2, 27·8 v. 35·4 %, 95 % CI 30·3, 40·5). Also, there was an increase of the probability of belonging to the ‘pasta and cheese’ MP in the third quartile compared with the first quartile (17·1 %, 95 % CI 13·0, 21·1 v. 10·1 %, 95 % CI 8·4, 11·9) (Table 2). A significant increase in the probability of belonging to the ‘ultra-processed’ SP in the third quartile compared with the second quartile (56·3 %, 95 % CI 52·4, 60·2 v. 47·2 %, 95 % CI 43·5, 50·8) was found for the mid-afternoon snack. No differences were observed at dinner and evening snack (Table 2).

Table 2. Probability (%) of belonging to a latent class at different meals/snacks in schoolchildren by midpoint of sleep quartiles (MSFsc group). Florianópolis. Brazil 2018/2019 (n 1333)

MSFsc, midpoint of sleep on free days corrected; Q, quartile.

* Adjusted for sex, age, screen use, type of school, maternal education, weight status, school shift, physical activity and day of food intake report.

After stratifying by school shift, at breakfast, we observed that afternoon shift differences were similar to total sample in ‘Coffee with milk, bread and cheese’ and ‘Mixed’ patterns. At the mid-morning snack, the students from afternoon shift showed the same result as the total sample in ‘traditional Brazilian lunch with soda’. At lunch, the morning shift presented a similar result to the total sample, while in the afternoon shift was observed a decrease in the ‘pasta and cheese’ pattern in fourth quartile compared with second quartile of MSFsc (5·3 %, 95 % CI 0·4, 10·2 v. 16·7 %, 95 % CI 10·3, 23·1). At dinner, it was observed that those on the morning shift had a lower probability of belonging to the ‘mixed’ pattern in fourth quartile compared to second quartile (34·6 %, 95 % 28·0, 41·1 v. 50·5, 95 % CI 45·1, 56·0). At the evening snack, the afternoon shift showed an increase in ‘ultra-processed’ SP in second quartile compared with first quartile (75·9 %, 95 % CI 66·9, 84·8 v. 58·6 %, 95 % CI 51·0, 66·3) and a decrease in the ‘water’ pattern in second quartile compared with first quartile (19·0 %, 95 % CI 10·2, 27·7 v. 37·7 %, 95 % CI 28·1, 47·3). No differences were observed at mid-afternoon snack (online Supplementary Table 10).

We performed a sensitivity analysis using MSFsc as a continuous variable; however, we only found difference at breakfast in ‘mixed’ (β = 0·31, P = 0·020) and ‘chocolate milk’ (β = 0·38, P = 0·026) pattern (data not shown).

Discussion

To the best of our knowledge, this is the first study to assess differences in meal and snack patterns between sleep timing groups (midpoint of sleep quartiles) in children and adolescents. Three patterns were identified for breakfast, mid-morning snack, evening snack and dinner on the one hand and four patterns for mid-afternoon snack and lunch on the other hand. The results suggested a link between sleep timing and meal/snack patterns in schoolchildren. Specifically, very late and late sleep preferences were associated with a higher probability of consuming the mixed breakfast pattern (healthy and unhealthy foods), the pasta-and-cheese lunch pattern and the ultra-processed mid-afternoon pattern. Very early and early sleep preferences were associated with an increased probability of consuming typical Brazilian foods at breakfast (coffee with milk, bread and cheese) and lunch (rice, beans and beef/poultry). To date, no studies have been identified that evaluated the association of sleep timing with meals or snacks patterns with the methods used in the present study, thus limiting a direct comparison with other GDP studies. It is also important to note that dietary patterns derived from a posteriori inherently reflect cultural and regional variations and may be influenced by the methodology employed for their derivation^(30,32,59) .

Schoolchildren with very early sleep preference were more likely to consume the ‘coffee with milk, bread and cheese’ breakfast pattern, considered a common combination of foods usually consumed for breakfast in some regions of Brazil, as described in similar studies with schoolchildren^(45,53) . The Dietary Guidelines for the Brazilian Population recognise this pattern as a healthy combination of breakfast foods, particularly when it includes fruits⁽⁵⁴⁾. A study in Florianopolis, conducted between 2013 and 2015 with public school students, derived meal and snack patterns using LCA and identified the second most common breakfast pattern, labelled ‘traditional Brazilian breakfast’, shared by a quarter of the students, characterised by coffee with milk, bread and cheese⁽⁵³⁾, similar to the most common BP found in the present study. Although evidence is scarce about the role of sleep timing influence on breakfast food choices, some studies focused on skipping breakfast. The results suggest that those with later sleep timing are more likely to skip breakfast^{(15,16,21,37,60)} .

In the present study, students with later sleep timing were more likely to consume the ‘mixed’ breakfast pattern which includes healthy food items such as fruits, vegetables, rice and beans, but also unhealthy ones, such as ultra-processed foods, pizza/hotdog/hamburger, sweets, cream cookies and soda. This BP suggests a less structured breakfast, with higher dietary variability of food items at a very late breakfast. This may be related to a typical mealtime upon waking up. For instance, the children who wake up close to breakfast time are more likely to consume the foods typically eaten at breakfast, while those waking up closer to lunchtime may gravitate towards lunch-type foods. Although the mixed patterns cannot be considered globally healthy or unhealthy, this BP requires monitoring due to the presence of ultra-processed foods and those rich in fat, sugar and salt in the first meal of the day. A nationwide representative study of 7425 Brazilian children and adolescents aged 10–19 years also found frequent mixing of both healthy and unhealthy breakfast patterns⁽⁶¹⁾.

Some studies found associations between later sleep timing with the consumption of unhealthy foods during the day^{(12,13,15,16,62)} . A study carried out with 465 children between 9 and 11 years old from New Zealand found that those with later sleep preferences had lower scores in the GDP composed by fruits and vegetables than those with earlier sleep preferences⁽¹⁷⁾. Similarly, Thellman et al. (2017)⁽³⁶⁾ conducted a study with 119 North American children aged 9–15 years and found that later sleep timing (fourth quartile of weekly midpoint of sleep) was associated with a higher probability of consuming high fat, sugar and salty foods during the day, compared with children with morning preference (first quartile of weekly midpoint of sleep). The authors highlighted that these differences were higher in the afternoon and evening hours of the day⁽³⁶⁾, this is in line with the present study result that those with later sleep timing were more likely to consume unhealthy foods in the ‘mixed’ BP.

Later sleep timing is not only associated with unhealthy eating but also with lower consumption of healthy foods such as milk⁽⁶²⁾, fruits and vegetables⁽¹⁶⁾ and less physical activities^(17,62) in children and adolescents. Furthermore, a study from Florianopolis found that schoolchildren with early sleep timing were less likely of being overweight including obesity than intermediate types⁽²⁴⁾. Likewise, Yang et al. (2023)⁽⁶²⁾ found that non-morning types (intermediate and evening chronotype) Chinese children between 10 and 12 years of age had a higher risk of being overweight compare with the morning types.

Late sleep timing was associated with the ‘traditional Brazilian lunch with soda’ at mid-morning SP when compared with very early sleep timing. The children who wake up later are more likely to skip mid-morning snack⁽²⁴⁾. However, this snack could be considered a large meal (rice, beans and beef/poultry) provide by public schools in Brazil^(53,63) or eaten at home before going to school in the afternoon shift. Another possibility is that children had difficulty in reporting mid-morning snack and lunch separately, due to the short time interval between them.

The schoolchildren with very early sleep preference were more likely to consume the ‘Brazilian traditional, processed meat, egg and fish’ lunch pattern compared with those with late sleep preference. The latter were more likely to consume the ‘pasta and cheese’ lunch pattern compared with those with very early preference. These results suggest that lunch also can be affected by sleep timing, whereby morning habits may contribute to a healthy lunch pattern. Although the ‘Brazilian traditional, processed meat, egg and fish’ lunch pattern contains processed meat such as sausages, it may still be considered a healthy combination of foods for lunch due to the presence of rice, beans and vegetables, which are recommended by the Dietary Guide for the Brazilian Population⁽⁵⁴⁾. Also, the guide states that pasta can be part of a healthy meal if accompanied with some source of protein as chicken and vegetables⁽⁵⁴⁾, thus making it difficult to classify the ‘pasta and cheese’ lunch pattern as either healthy or unhealthy. The students with late sleep preference were more likely to consume the ‘ultra-processed’ mid-afternoon snack pattern compared with those with early sleep preference. Roberto et al. (2022)⁽⁵³⁾ found a similar most common mid-afternoon SP labelled ‘ultra-processed’ composed by energy-dense foods such as cream cookies, soda, sweets and pizza/hotdog/ hamburger. However, we did not find studies investigating the composition of afternoon snacks and sleep timing. Our results support the hypothesis that later sleep timing is associated with a higher consumption of unhealthy foods during the afternoon snack.

School shift appears to influence the relationship between MSFsc quartiles and meal/snack patterns, with schoolchildren attending the afternoon shift appearing to be more affected by changes in MSFsc quartiles, although we also observed changes in the morning shift. However, our results do not allow us to assess the impact of the school shift on the association between quartiles of MSFsc and meal/snack patterns since some associations were observed between the first and second quartiles of MSFsc and not between the opposite quartiles, and some associations with meal/snack patterns were with a mix of foods considered healthy and unhealthy. Peng et al. (2024)⁽⁶⁴⁾ in a study with 305 Mexican students between 9 and 17 years old identified that who attend to the morning shift had higher adherence to a global dietary pattern ‘meat and starchy’ (chips, refined grains, sugar-sweetened beverages, processed meat and high-fat dairy, sweets, pork, Mexican foods, potatoes and fried plantains, soup, legumes and corn tortillas) compared with students in the afternoon shift. In contrast, a Brazilian study with 635 high school students found no differences between adherence of two global dietary pattern ‘processed’ and ‘unprocessed’ among those who attend school in the morning compared with those who studied in the afternoon shift⁽⁶⁵⁾. Future studies may help understand how chronotype and school shifts attended can influence the composition of meals and snacks.

In the present study, the most common morning and evening snack patterns were characterised by the presence of ultra-processed foods and fruits, thus suggesting a combination of both healthy (e.g. fruits) and unhealthy (e.g. ultra-processed) foods in daily snacks. Similarly, a nationwide representative study entitled Brazilian Dietary Survey conducted in 2017–2018 with 8264 adolescents aged 10–19 years found that morning, afternoon and evening snacks were mostly composed by sweeteners added to food and beverages, cookies/crackers, coffee/tea, fast food, fruit juices and fruits⁽⁶⁶⁾. Furthermore, a study with 5264 USA adolescents aged 12–19 years from the National Health and Nutrition Examination Survey (NHANES), identified the most frequent sources of snacks came from fruits, refined grains, oils, solid fats and added sugars⁽⁶⁷⁾. On the other hand, fruits did not compose any most common snack patterns in the study carried out with students from public schools in Florianópolis, Brazil, between 2013 and 2015⁽⁵³⁾. Socio-economic differences between public and private students may explain these results, as attending a private school in Brazil requires higher income, also associated with higher fruit and vegetable intake in Florianopolis^(68,69) .

It is important to highlight that consuming food late at night has been associated with disruption in circadian rhythms. In our study, we highlighted that most children are consuming ultra-processed foods, rich in fat, sugar and salt as an evening snack, possibly close to bedtime, which can cause changes in circadian cycles responsible for hormonal and metabolic oscillations related to overweight and obesity⁽⁷⁰⁾.

The strengths of the present study include a large representative sample of both private and public school students – a proxy of socio-economic status. The use of the midpoint of sleep to measure sleep timing is a new approach in sleep research⁽¹¹⁾ that considers the dimension of the time of day that sleep occurs, while traditional research focus on sleep duration. Moreover, some authors consider LCA less subjective than factor analyses to derive the meal and snack patterns^(30,59) . The former is also described as person-centred analysis which facilitates data interpretation and may be adjusted for the covariates known to influence both sleep and dietary intake⁽⁴⁾. The approach based on time-tagged eating events – a relatively new field of dietary patterns research – has been growing fast, producing important practical applications based on time-of-the-day food intake. Also, nutritional recommendations based on specific eating occasions may be easier to follow⁽³⁰⁾.

The present study’s limitations include a cross-sectional design centred on the analysis of associations that do not always imply a causal relationship. Sleep data were provided subjectively by parents or guardians, who may be prone to recall errors. However, this methodology is commonly used in epidemiological studies⁽⁷¹⁾. Screen use frequency, physical activities and dietary intake were all self-reported by schoolchildren and therefore depend on their attention, memory and perceived social desirability⁽⁴³⁾. Although Web-CAAFE was verified for usability, reproducibility and external validity, we cannot exclude the possibility of errors in the allocation of specific foods across eating occasions⁽⁵³⁾. The dietary data from a single day may not fully represent individuals’ habitual consumption⁽⁷²⁾. Nevertheless, this method is still widely applied for population-level assessment⁽⁵²⁾. Furthermore, to mitigate potential bias, data collection was conducted across different days of the week, encompassing both weekdays and weekends. Lastly, Web-CAAFE does not provide the exact timing of eating events or the amount of food intake, thus limiting the possibility of investigating associations between energy intake and meal/snack timing.

Conclusion

The study findings suggest that schoolchildren with very late sleep timing are more likely to consume the ‘mixed’ pattern and less likely to consume the ‘coffee with milk, bread and cheese’ pattern at breakfast when contrasted with those with very early sleep timing. Also, later sleep preferences are associated with a lower intake of the ‘Brazilian traditional, processed meat, egg and fish’ lunch pattern and a higher intake of the ‘pasta and cheese’ pattern. Later sleep timing was also associated with higher consumption of ultra-processed foods at mid-afternoon snacks compared with students with earlier sleep timing.

For breakfast, a morning preference for sleep timing was associated with a healthier breakfast pattern, while later preferences seem to have an inverse association. Similarly, a morning preference appears to be advantageous for promoting healthy lunch and afternoon snack patterns. Future longitudinal studies are needed to clarify the causal nature of these associations and those between sleep timing and energy intake on different eating occasions. The present study findings provide a basis for practical dietary recommendations, focused on specific eating events, bedtime and wake-up time. With a growing expert agreement on the importance of sleep timing in a healthy diet, school intervention projects should give more weight to early sleep routines to improve healthy meal/snack patterns in children and adolescents.