Stability of the timing of food intake at daily and monthly timescales in young adults

The study addresses whether the timing of food intake is stable over time and whether stability differs when referenced to clock time versus circadian phase. Rising global obesity, with college-aged individuals exhibiting rapid weight gain, underscores the need to identify behavioral factors such as meal timing that influence cardiometabolic health. Prior cross-sectional work links later eating (especially closer to dim-light melatonin onset, DLMO) with higher BMI and body fat, but the reliability of single-week or single-day assessments and the stability of meal timing relative to physiological circadian markers remain unclear. The authors hypothesized: (i) timing of food consumption would be stable across multiple months; (ii) circadian-referenced timing would be more stable than clock timing; and explored (iii) whether lean individuals exhibit more stable timing and caloric intake than non-lean individuals. Understanding stability across daily and monthly timescales informs the interpretability of cross-sectional studies and potential interventions for health.

Prior research suggests meal timing affects cardiometabolic outcomes, with later eating associated with weight gain and higher BMI. Studies indicate that non-lean individuals consume a greater share of daily calories later, closer to DLMO. Literature on meal regularity shows associations between irregular meal patterns and adverse metabolic profiles; however, stability has often been evaluated over only several days within a week, commonly operationalized as consuming or skipping predefined meals rather than examining continuous timing of caloric events. These limitations motivate assessing stability using objective timing metrics across longer intervals and relative to circadian phase.

Design: Observational study over a spring semester with three assessment time points approximately one month apart. Participants: 14 undergraduates (5 female), age 19.1±0.3 years (range 18–21), BMI 22.9±0.8 kg/m² (18.3–28.5). Inclusion criteria included smartphone access for logging, no night-shift work, no travel >1 time zone in prior 3 months. IRB-approved; informed consent obtained; trial registered (NCT02846077). Field procedures: Participants recorded all caloric intake for 7 consecutive days at each of the three time points using MealLogger, a smartphone app capturing time-stamped photos, meal labels, and descriptions. They included an object of known size; a second photo documented leftovers. Dietitians scored caloric content and macronutrients using the Nutrition Data System for Research. Events within ≤15 min of a previous entry with the same meal label were combined into a single caloric event. Timing metrics included first daily caloric event, caloric midpoint (time when 50% of daily calories were consumed), last daily caloric event, and participant-labeled meals (breakfast, lunch, dinner, snacks). In-laboratory procedures: At each time point, participants completed a ~16-h dim-light (∼4 lx) overnight for circadian phase assessment and body composition. Saliva samples were collected hourly from ~16:00 to ~07:00 for melatonin. DLMO was determined as the linear interpolated time when salivary melatonin exceeded and remained above 5 pg/mL. Body composition was measured via four-lead bioelectrical impedance (three measures averaged). Derived measures: Each caloric event was assigned a circadian phase relative to DLMO (0°). Circadian metrics included: percent of calories consumed within 4 h of DLMO to sleep onset; circadian phase of peak caloric timing; and difference between caloric midpoint and DLMO. Day-to-day stability was quantified for first, midpoint, and last caloric events using Composite Phase Deviation (CPD), which combines deviation from the previous day (regularity) and deviation from an individual’s mean timing (alignment). Standard deviations of these daily timings were also computed per week. Participants were categorized as lean (n=8) or non-lean (n=6) using sex-specific body fat cutoffs (non-lean: ≥31% females, ≥21% males); lean were below these thresholds. Statistical analysis: Mixed-effects models (variance components) tested month effects on weekly averages (months 1–3), with month as fixed factor and participant as random factor. Planned t-tests with Bonferroni correction (p<0.017) followed significant main effects. Intra-class correlation coefficients (ICC; two-way mixed-effects, single measures) quantified intra-individual stability across months; strength interpreted as slight (0.00–0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), almost perfect (0.81–1.00). Pearson correlations examined between-month associations, with normality assessed via skewness and kurtosis. Analyses used SAS 9.4.

- Data coverage: Participants were studied over ~104 days across the semester; monitoring weeks occurred during weeks 2–6, 7–11, and 12–16. - Circadian phase: Mean DLMO 23:26 (SEM 0:29; range 20:06–03:32) with no significant month effect. - Day-to-day variability (CPD across all months): First event 3.1 h (0.4; range 0.8–5.3), midpoint 3.7 h (0.4; 1.2–5.5), last event 4.9 h (0.3; 3.0–6.7). Standard deviations: first 2.2 h (0.2; 0.2–5.1), midpoint 3.5 h (0.2; 1.1–6.0), last 2.6 h (0.2; 0.4–5.7). No significant month effects for CPD or SD (all p>0.36). CPD ICCs were slight-to-fair: first 0.12, midpoint 0.13, last 0.34; SD ICCs: first 0.43, midpoint 0.18, last 0.43. - Weekly average clock timing across months: Population-level main effect for snacks (p=0.04) and first daily caloric event (p=0.01); first event earlier in month 1 vs month 2 (p<0.01). No significant month effects for breakfast, lunch, dinner, caloric midpoint, or last event. Variation across months in timing ~1 h (vs ~3 h day-to-day). - Intra-individual stability (ICC) across months by clock time: Breakfast ICC 0.29 (fair), lunch 0.27 (fair), dinner 0.63 (substantial), snacks 0.56 (moderate). First daily caloric event 0.63 (substantial), caloric midpoint 0.54 (moderate), last event 0.55 (moderate). - Circadian-referenced stability: Group-level no month effects for percent of calories within 4 h of DLMO to sleep onset, phase of peak caloric timing, or caloric midpoint relative to DLMO. ICCs indicated lower stability than clock time: percent within 4 h ICC ~0.36 (moderate by authors’ description), peak phase ICC ~0.33 (fair), caloric midpoint relative to DLMO ICC ~0.33–0.41 (fair-to-moderate). - Lean vs non-lean exploratory comparisons: Non-lean participants showed more significant between-month correlations in timing metrics (11/30) than lean (4/30). Non-lean group displayed strong between-month associations for several meal times and for first/mid/last caloric events (e.g., months 1–2 r≈0.84–0.96, p≤0.05 in multiple metrics). Non-lean also had more stable DLMO timing between months, which may contribute to their greater stability relative to circadian time. - Caloric and macronutrient intake: Mean daily calories 1627 (SEM 127; range 840–2346) with a main effect of month (p=0.04) but no significant Bonferroni-corrected pairwise differences. Average meals/day 2.9 (0.3; 1.7–6.4), no month effect. Macronutrient percentages showed no month effects. Intra-individual ICCs across months: daily calories 0.70 (substantial), fat % 0.57 (moderate), carbohydrate % 0.76 (substantial), protein % 0.83 (almost perfect). Meal number showed moderate consistency; lean participants had significant correlations in meal number between months 1–2 (r=0.90, p=0.002) and 1–3 (r=0.89, p=0.04); non-lean did not.

Findings demonstrate two relevant timescales for eating behavior stability in young adults: (1) day-to-day timing within a week shows poor stability (CPD and SD metrics slight-to-moderate ICCs), and (2) weekly average timing is fairly to substantially stable across months by clock time (ICC 0.54–0.63). Thus, while single-day assessments are unlikely to reflect habitual behavior, a week of monitoring yields stable patterns at the individual level across months. Contrary to the hypothesis that circadian-referenced timing would be more stable, clock-time timing was more stable than circadian-referenced timing (lower ICCs relative to DLMO). This discrepancy may be driven by social scheduling and external constraints (e.g., class schedules, eating with peers) that anchor meals to clock time rather than internal circadian phase. Dinner and last caloric event timing exhibited relatively strong stability, consistent with socially coordinated evening meals. Exploratory analyses showed non-lean participants had greater stability across months in several timing metrics and more stable DLMO, potentially contributing to higher circadian-referenced stability; this contrasts with prior associations between irregular eating and poorer metabolic health, possibly reflecting differing definitions of "stability" (timing versus meal frequency) across studies. Variability measures may require larger numbers of observations to achieve reliability, which could explain the weaker stability of day-to-day metrics.

The timing of caloric intake in college-aged adults is fairly stable across months when assessed over a week, but exhibits poor day-to-day stability within weeks. Clock-time metrics of eating are more stable across months than circadian-referenced metrics, likely due to social and environmental constraints. A single day of dietary logging does not represent habitual eating timing, whereas a week-long assessment is more representative. Exploratory results suggest non-lean individuals may show greater temporal stability of eating and circadian phase than lean peers, warranting further study. Future work should include larger and more diverse populations, investigate social determinants of eating timing, evaluate reliability of variability metrics with longer sampling, and examine links between stability of eating timing, body composition, and cardiometabolic outcomes.

- Small sample size (n=14) limits power, particularly for lean vs non-lean comparisons and for linking timing to body composition/health outcomes; underpowered for regression analyses. - Study limited to college students over one semester; generalizability to other populations and settings is uncertain. - Semester-related and seasonal factors (e.g., exams) and inclusion of both school/work and free days may introduce exogenous variability; however, this enhances real-world relevance. - Variability metrics (e.g., CPD, SD) may require many repeated observations to achieve high reliability; 7-day windows may be insufficient for robust individual variability estimates. - Self-logging could alter behavior, though longer recording periods may mitigate this effect.