A Smartphone App Reveals Erratic Diurnal Eating Patterns in Humans that Can Be Modulated for Health Benefits

The study addresses how the timing and duration of daily eating in free-living humans relates to circadian biology and health, and whether restricting the daily eating window yields benefits. Circadian clocks coordinate physiology with daily cycles of activity, sleep, and feeding-fasting. In model organisms, restricting food intake to an 8–12 hr window (time-restricted feeding, TRF) improves multiple metabolic and physiological outcomes without necessarily changing calories. However, human eating timing is rarely measured objectively, hindering translation. The authors sought to (1) objectively characterize real-world temporal eating patterns in healthy non–shift-working adults using a smartphone-based method with minimal feedback, and (2) test in a pilot whether shortening daily eating duration to ~10–11 hr, without prescribed caloric or nutritional changes, can improve body weight and sleep in overweight individuals.

Prior work shows circadian disruption (genetic or behavioral, e.g., shift work) perturbs metabolic regulation and predisposes to disease. Frequent caloric intake in animals dampens circadian rhythms. TRF in mice and Drosophila (8–12 hr windows) enhances metabolic fitness, reduces adiposity, improves sleep, endurance, inflammation, cardiac aging, and gut homeostasis, and alters diurnal gene expression in peripheral organs. Human epidemiology links meal timing, social jetlag (weekday/weekend sleep timing differences), short sleep, and metabolic risk. Despite these insights, human data on diurnal patterns of food intake timing are limited because standard dietary assessment tools emphasize quantity/quality over timing.

Baseline monitoring: Healthy adult men and women (non–shift workers) were recruited (IRB-approved; consent obtained) and monitored for 3 weeks during normal routines using a custom iOS smartphone app (Salk Metabolic App). Participants logged every ingestion event (food, beverage, water) just prior to consumption primarily via photographs (JPEG downscaled on-device to 1/10 original size); text entries were allowed when photos were impractical. Each entry captured timestamp and geolocation and was uploaded to a server; data were then deleted from the device to minimize feedback effects. Randomly timed push notifications (1–2/day during stated wakeful periods) asked whether participants had consumed anything in the past 30 min to estimate underreporting; the false negative rate was 10.34%. Cohort and data: Data from 156 individuals over 21 days were analyzed, yielding 26,676 events (2.1% text-only). Of these, 22% were water, 28% pre-packaged items (with readily accessible nutrition info), and 50% mixed meals. Caloric content was estimated via CalorieKing or FNDDS. Average estimated intake was 1,947 kcal/day (95% CI 1,917–1,977), about 1.233× maintenance calories (95% CI 1.214–1.251). No significant weight change occurred during the 3-week logging. Actigraphy subset: A subset of 47 participants wore wrist actigraphy (CamNTech MotionWatch 8) to measure activity and light exposure, enabling alignment of ingestion with activity/sleep timing. Temporal definitions and processing: Because ingestion events had a population-level trough near 4 a.m. (and digital activity trough 2–4 a.m.), the "metabolic day" was defined to start at 4 a.m. Events within 15 min of each other were merged into a meal based on observed average meal duration (14 min 36 s). Inter-meal interval distributions were computed. Cumulative caloric timing profiles were generated relative to local time (daylight savings considered). The daily eating duration for each person was defined as the 95% interval (2.5th–97.5th percentile) of ingestion event times starting at 4 a.m., aggregated over the monitoring period to tolerate occasional missed logs. Intervention study: From the baseline cohort, eight overweight/obese adults (BMI >25 kg/m²) with baseline eating duration >14 hr were enrolled in a 16-week pilot intervention. After review of their baseline pattern (including a personalized raster plot "feedogram"), each selected and adhered to a consistent 10–12 hr daily eating window across weekdays and weekends, limiting all non-water intake (including coffee/tea) to that window. No advice on diet quality, quantity, or calories was given. Participants continued using the app and received weekly summaries (feedograms and eating duration). Anthropometrics were measured at baseline, after 16 weeks, and at 1 year (36 weeks after supervised monitoring ended). Subjective ratings (energy morning/overall, hunger at bedtime, sleep satisfaction; 1–10 scale) were collected. A fasting metabolic panel was obtained pre-intervention to screen for undiagnosed hypoglycemia. Data annotation included duplicate removal via MD5, manual de-duplication, dual independent annotations with consensus resolution. Analyses used Mathematica and GraphPad Prism; raster plots binned events into 15-min intervals with day–night coloring.

- Human eating is frequent and temporally erratic across wakeful hours with minimal evidence of a strict three-meals-a-day structure. Only a 5-hr window between 1–6 a.m. had <1% of events per hour. - Caloric timing is skewed late: <25% of calories are consumed before noon; 37.5% after 6 p.m.; 12.2% after 9 p.m.; 3.9% after 11 p.m. The median times to reach 50%, 70%, 90%, and 100% of maintenance calories were 3:32 p.m., 5:04 p.m., 6:11 p.m., and 6:36 p.m., respectively. - Inter-meal patterns: Median inter-meal interval 3 hr 6 min; 25% of meals occur within 1 hr 25 min of another meal; only 25% follow fasting >6 hr 41 min. - Sleep parallels fasting: Median time from waking to first caloric intake was 1 hr 18 min; median time from last caloric intake to bedtime was 2 hr 22 min; overnight fasting roughly matched time in bed. - Weekday/weekend shifts indicate "metabolic jetlag": Mean first caloric intake was 9:21 a.m. on weekdays vs 10:26 a.m. on weekends; 40% delayed breakfast by ≥1 hr on weekends; last caloric intake timing was more variable on Fri/Sat. - Eating duration: Median daily eating duration (95% interval) was 14 hr 45 min; more than half the cohort exceeded 14.75 hr; only 9.7% had <12 hr. Eating duration correlated weakly with last caloric time (r²=0.215) and minimally with breakfast time (r²=0.035) or BMI (r²=0.017). - Underreporting estimate from push notifications was 10.34%. - Intervention (n=8, BMI>25, baseline eating >14 hr): Participants reduced eating duration by an average 4 hr 35 min (95% CI 3 hr 30 min–5 hr 40 min; p<0.001) and reduced weekday/weekend timing variability to <1 hr. Average body weight decreased by 3.27 kg (95% CI 0.91–5.62 kg) with BMI reduction of 1.15 kg/m² (95% CI 0.325–1.98). Subjective energy improved (morning and overall), hunger at bedtime decreased, and sleep satisfaction improved (all p<0.05). At 1 year, weight loss and sleep improvements were maintained. - Estimated daily caloric intake fell by 20.26% on average during intervention (95% CI 4.92%–35.6%; p<0.05), potentially reflecting the timing-dependent omission of certain items rather than shifting them into the eating window.

An image-based, minimally feedback smartphone method can objectively and scalably capture the timing of human ingestion events in free-living conditions, overcoming limitations of traditional self-report tools and enabling integration with activity and light data. Findings reveal that many healthy, non–shift-working adults eat across most waking hours with short overnight fasts and a late-day caloric bias. Variability between weekdays and weekends creates a form of "metabolic jetlag" that may contribute to metabolic dysregulation, potentially linking observed associations between short sleep, social jetlag, and metabolic disease to reduced fasting duration and altered meal timing. Despite no simple cross-sectional correlation between eating duration and BMI, a feasibility intervention shortening the eating window to 10–12 hr reduced body weight and improved sleep and energy without increased hunger, suggesting benefits may arise from restored feeding-fasting rhythms, reduced metabolic jetlag, and incidental caloric reduction. The approach also sets the stage for assessing timing of medications and supplements relative to circadian and feeding cycles, which could influence efficacy and safety.

This work introduces a scalable, evidence-based framework to quantify diurnal eating patterns in humans and demonstrates that the typical daily eating window often spans ~15 hr with late-day caloric concentration and weekday–weekend variability. A pilot time-restricted eating intervention (10–12 hr window) in overweight adults reduced body weight and improved sleep and energy, with benefits maintained at 1 year. Future research should test generalizability across diverse ages, occupations (including shift workers), cultures, and geographies; examine populations with non-modern lifestyles; incorporate randomized controlled designs; and comprehensively assess metabolic biomarkers to disentangle effects of meal timing from caloric and dietary composition changes.

- Baseline-to-intervention design without a randomized control group; small intervention sample (n=8) limits generalizability. - Short-term baseline monitoring provides a snapshot of long-term behaviors; cohort heterogeneity (age, gender, activity) may confound associations (e.g., weak BMI correlation). - Potential underreporting of ingestion events (estimated false negatives ~10.34%); reliance on self-logging may bias event capture, particularly for small snacks. - Actigraphy data available only for a subset (n=47), limiting sleep–eating alignment analyses. - No comprehensive metabolic biomarker assessment; primary outcomes were anthropometrics and subjective measures. - Intervention led to reduced caloric intake, confounding isolation of timing effects vs caloric restriction. - Self-selected eating windows and free-living conditions introduce variability; participants were smartphone users who consented to monitoring (possible selection bias).