The timing of macronutrient and major food group intake and associations with mortality among US adults, 1999-March 2020: a serial cross-sectional study

Abstract

Background:

Eating timing has been increasingly linked to health, yet national trends in macronutrient/food group timing and their health implications remain unclear.

Objective:

To characterize trends in timing of energy, macronutrient, and food group intake among US adults and examine their associations with mortality.

Methods:

In this serial cross-sectional study of adults aged ≥20 years with ≥1 valid 24-hour dietary recall (National Health and Nutrition Examination Survey, 1999-March 2020), we examined secular trends in timing of energy, macronutrients, and major food group intake. Associations with mortality (through December 2019) were examined using Cox models.

Results:

Among 50,264 adults, evening (6–10pm) accounted for the highest daily energy intake (weighted mean proportions across years, 31.9%–33.3%), followed by noon (10am-2pm, 24.7%–26.8%), afternoon (2–6pm, 19.9%–21.8%), morning (6–10am, 13.5%–14.9%), and overnight (10pm-6am, 5.6%–6.5%); midnight (10–2am) eating occurred in 23.4%–28.0% of the population. Macronutrient and food groups followed similar patterns, except whole grain (peaked in the morning) and fruit, egg, and dairy intake (more evenly distributed). Over years, noon and midnight energy intake proportions declined, while afternoon proportion increased; secular trends varied by macronutrients/food groups. Fasting started at 8:34–8:51pm and ended at 8:41–8:52am; intake midpoint was 2:38–2:48pm; intake duration was 11.9–12.2 hours. Male, non-Hispanic black, and socioeconomically disadvantaged groups had greater midnight intake proportions and later intake midpoints. Reallocating 5% of daily energy to midnight was associated with higher cardiovascular mortality (HR, 1.09; 95% CI, 1.02,1.17), driven by carbohydrates; reallocating 5% to predawn (2–6am) was associated with higher cancer mortality (1.22;1.05,1.41), driven by proteins. Each 1-hour delay in fasting and intake midpoint was associated with an 8%–9% higher cardiovascular mortality.

Conclusion:

Overnight intake and delayed eating timing are prevalent among US adults, especially among socioeconomically disadvantaged groups, and were associated with higher mortality, particularly for specific macronutrients/foods, supporting eating timing recommendations integrating food composition.

Introduction

The 2020–2030 Strategic Plan for NIH Nutrition Research highlights the need to understand the role of dietary behaviors for optimal health, including when we should eat [1]. Eating timing, a key aspect of chrononutrition, is conceptualized by timepoints (when intake occurs), duration (length of intervals between eating occasions), and distribution (proportions of daily intake across time blocks) [2]. Though linked to cardiometabolic health [3], chrononutrition is still in its infancy [2], prompting its identification as a knowledge gap by the 2025 Dietary Guidelines Advisory Committee [4] and multiple societies [5, 6].

Eating timing remains poorly characterized among US adults, although two studies from National Health and Nutrition Examination Survey (NHANES; 2009–2014 and 2011–2018) described energy intake timing using self-reported eating occasions (e.g., “breakfast”) [7, 8], which vary across individuals and populations, limiting between-study comparability and generalizability [4]. Using clock time as a standardized anchor is needed. Additionally, different macronutrients and food groups may exert different timing-dependent health effects [9] given their distinct circadian metabolism [10–12], however, beyond energy intake timing, macronutrient- and food-specific eating timing has not been comprehensively assessed. Furthermore, social determinants of eating timing are essential for developing targeted interventions to promote health equity but remain understudied.

While shifting energy intake earlier in the day (e.g., avoiding breakfast skipping and late-night eating) and prolonging fasting intervals have shown health benefits [3], macronutrient/food-group-specific eating timing in relation to health outcomes remains unclear. Using NHANES (1999-March 2020) data, we characterized trends in the timing of energy, macronutrient, and food group intake among US adults based on a comprehensive eating timing framework. We further examined their sociodemographic differences and associations with mortality as an exploratory analysis.

Methods

Study Design

Since 1999, the NHANES has annually employed a stratified, clustered, four-stage, probability sampling design to recruit a nationally representative sample [13]. Data were released in 2-year cycles; however, data collection was interrupted by the COVID-19 pandemic in March 2020, and 2017-March 2020 data were combined to provide nationally representative estimates [13].

NHANES collected demographic, health, and nutrition information from questionnaire interviews and standardized physical examinations [13]. The NHANES protocol was approved by the National Center for Health Statistics research ethics review board. All participants provided written informed consent. This analysis included non-pregnant adults ≥20 years with at least one valid dietary recall (defined in Supplemental Methods) recruited between 1999 and March 2020 (10 cycles).

Dietary Assessment

NHANES employed 24-hour dietary recalls (midnight to midnight) using the multiple-pass method and standardized measuring guides to ensure complete and accurate recalls [14]. Participants reported time, name (e.g., breakfast, lunch, dinner, snack), and each food/beverage intake amount for each eating occasion. In all cycles, one in-person recall was conducted at the Mobile Examination Center; since 2003, a second telephone recall was administered [14]. Nutrient and food group intake were calculated using the updated USDA Food and Nutrient Database for Dietary Studies (FNDDS) and MyPyramid Equivalents Database and Food Patterns Equivalents Database (MPED/FPED; convert foods into food pattern components standardized as serving equivalents) [15].

Eating Timing

According to a methodological review, eating timing has three conceptualizations (Supplemental Figure 1) [2] and was conceptualized separately for energy, macronutrients (including carbohydrate, protein, and fat), and major food groups (including whole grains, refined grains, whole fruits, fruit juice, vegetables, red/processed meat, poultry, seafood, eggs, dairy, and lentils/nuts/soy). First, timepoints described when consumption occurred, including (1) whether energy and each macronutrient/food group were consumed within each time block of the day, (2) the daily intake midpoint for energy and each macronutrient/food group, and (3) the start and end time of fasting for energy and each macronutrient (fasting was defined as the longest interval without consuming energy or certain macronutrient between two reported eating occasions). Second, for energy and each macronutrient/food group, distribution was assessed by the proportion of daily intake consumed within each block of the day. Third, durations of energy/macronutrient intake (also referred to as the eating window) were calculated as the time difference between the end and the start of fasting. The start/end time of fasting or eating duration was not calculated for food groups, as many participants consumed certain food groups once or fewer per day.

Based on a previous study [16], the 24-hour cycle was divided into six 4-hour blocks – 2:00–5:59 am (predawn), 6:00–9:59 am (morning), 10:00 am-1:59 pm (noon), 2:00–5:59 pm (afternoon), 6:00–9:59 pm (evening), and 10:00 pm-1:59 am (midnight). To support this segmentation scheme, we calculated the mean hourly intake of energy, macronutrients, and food groups across the day. Each 4-hour block was also split into two 2-hour blocks for finer resolution. Secondary analyses used self-reported eating occasion names (breakfast/lunch/dinner/snack).

Primary analyses used the first recall, assuming that evenly distributed recalls throughout the year and week ensure accurate population-level estimates on a given day [17]. A sensitivity analysis used the average of two recalls. Secondary outcomes included macronutrient subtypes (high- and low-quality carbohydrates, defined by food sources; Supplemental Table 1; animal and plant proteins; saturated [SFAs], monounsaturated [MUFAs], and polyunsaturated fatty acids [PUFAs]) [18] and mean Healthy Eating Index (HEI) 2020 score within each block (calculated with the population ratio method, Supplemental Table 2) [19, 20].

Mortality Assessment

Mortality was ascertained via National Death Index through December 31, 2019, including all-cause, cancer (International Classification of Diseases, 10th Revision codes, C00-C97), and cardiovascular disease (CVD; I00-I09, I11, I13, I20-I51, and I60-I69) mortality. To reduce the potential for reverse causation, participants with <5 years of follow-up were excluded [21].

Statistical Analysis

All analyses considered sampling weights, stratification, and clustering of the complex sampling design to derive nationally representative estimates. The sampling weights considered race/Hispanic origin, age, sex, weekday/weekend for dietary recalls, and nonresponse. Invalid recalls were treated as nonrespondents and excluded; this does not affect representativeness because the sampling weights incorporate nonresponse adjustment. First, for energy and each macronutrient/food group, eating timing variables were summarized by survey cycles using weighted means/proportions with 95% confidence intervals (CIs). To illustrate the secular trend, we regressed each eating timing variable on the survey cycle (detailed in Supplemental Methods); survey cycle was treated as a continuous (to estimate P values for overall secular trends) and a categorical variable (to calculate differences between 2017-March 2020 and 1999–2000), respectively. Second, we explored social determinants of eating timing by comparing energy/macronutrient/food group intake timing across age, sex, race/Hispanic origin, education, income, employment status, food security, and working schedule (2005–2010 only) subgroups (detailed in Supplemental Methods). Participants with missing data on subgroup variables were excluded from the corresponding subgroup analyses.

Third, for energy and each macronutrient/food group, Cox models estimated hazard ratios (HRs) and 95% CIs for mortality associated with consumption within each block (yes vs no), daily intake midpoint and fasting start/end time (per hour delay), proportion of daily intake within each block (per 5% increase), and duration (per hour increase). Models adjusted for age, sex, race/Hispanic origin, education, income, employment status, food security, daily energy intake, daily intake of the examined macronutrient/food group, number of daily eating occasions, daily HEI score, sleep duration, cigarette smoking, alcohol consumption, leisure-time physical activity, body mass index, and prevalent hypertension, high cholesterol levels, diabetes, cardiovascular disease, and cancer. Missing values in continuous covariates were imputed at the year- and sex-specific median, and a “missing” category was generated for missing values in categorical covariates [22]. Each timing conceptualization for carbohydrate, protein, and fat was mutually adjusted due to high correlations (r range, 0.48–0.83). Within each block, food group intake with high correlation was also mutually adjusted (Supplemental Methods). Unlike other analyses estimating population-level means/proportions, this analysis linked individual-level eating timing to their mortality risk. Using two dietary recalls can improve the reliability of eating timing estimates for each participant [23]. Therefore, only participants with two dietary recalls were included (averaged across two days), and two-day dietary sampling weights were used to generate nationally representative estimates [13]. A sensitivity analysis was conducted by excluding participants with evening/night/rotating shifts (data are only available in three cycles – 2005–2010; Supplemental Methods). Analyses were performed in R 4.4.1 (The R Foundation). Two-sided P <0.05 was considered statistically significant.

Results

Study Design and Population

Between 1999 and March 2020, 51,688 NHANES participants aged 20 years or older provided at least one valid dietary recall. After excluding 1,424 pregnant participants, 50,264 participants (weighted mean age, 47.5 years [SE, 0.19 years]; 25,326 [51.2%] females) were included in this analysis. From 1999 to March 2020, compared to the earlier cycles, mean age was higher in later cycles (Supplemental Table 3), and a lower proportion of non-Hispanic white adults was observed in later cycles; secular trend analyses showed increasing trends in socioeconomic levels and the prevalence of low/very low food security.

Timing of Energy Intake

Between 1999 and March 2020, three energy intake peaks were consistently observed at 8:00–8:59 am, 12:00–12:59 pm, and 6:00–6:59 pm, with similar peaks for macronutrients and most food groups (Figure 1 and Supplemental Figure 2). The first two peaks occurred near the midpoint of predefined morning (6:00–9:59 am) and noon (10:00 am-1:59 pm), while the third aligned with the onset of evening (6:00–9:59 pm). Furthermore, across twelve 2-hour blocks, four blocks spanning 10:00 pm-5:59 am contributed the least to energy, macronutrient, and food group intake (Supplemental Figures 3–4), corresponding to two blocks with minimal intake: midnight and predawn (collectively referred to as overnight). These findings support the rationale for the current 4-hour block structure.

Figure 1.

Intake of Energy, Macronutrients, Grains, and Fruits during Each Hour among US Adults, 1999-March 2020. X-axes indicate the survey year, which were grouped into three periods (1999–2006, 2007–2014, and 2015-March 2020) with roughly similar time spans. Instead of analyzing all original 2-year cycles separately, we combined them into three groups to increase statistical power and reduce random variability. Y-axes indicate the hour of the day (in 24-hour format), e.g., 0 indicates 0:00–0:59 am, 16 indicates 4:00–4:59 pm. Z-axes indicate (A) the mean intake of energy, macronutrients, grains, and fruits during each hour or (B) the proportion of US adults consuming energy, macronutrients, grains, and fruits during each hour. Sampling weights, stratification, and clustering of the complex sampling design were considered. Food group intakes were derived using the MyPyramid Equivalents Database (MPED) and Food Patterns Equivalents Database (FPED), which translates reported foods into standardized “serving equivalents” and allows consistent comparison of food group contributions across survey cycles. Specifically, grains (including whole grains and refined grains) are expressed in ounce equivalents; fruits (including whole fruits and fruit juice) are expressed in cup equivalents; total vegetables are expressed in cup equivalents; protein foods (including red/processed meat, poultry, seafood, eggs, and lentils/nuts/soy) are expressed in ounce equivalents; dairy foods are expressed in cup equivalents.

From 1999 to March 2020, over 82.5% of US adults consumed energy in the evening and noon, respectively; over 70.5% in the morning and afternoon, respectively; 23.4%–28.0% at midnight; 7.6%–10.7% in the predawn (Supplemental Figure 5). US adults consumed the highest proportion of daily energy in the evening (mean proportions across 1999-March 2020, 31.9%–33.3%), followed by noon (24.7%–26.8%), afternoon (19.9%–21.8%), morning (13.5%–14.9%), midnight (4.2%–5.3%), and predawn (0.9%–1.6%; Figure 2).

Figure 2.

Proportion of Macronutrient and Food Group Intake in Each Block among US Adults by National Health and Nutrition Examination Survey Cycles from 1999 to March 2020. Sampling weights, stratification, and clustering of the complex sampling design were considered. For each nutrient or food group, the intake proportion at each block was calculated among participants who reported consuming the certain nutrient or food group on the survey day. Six 4-hour blocks were predefined – 2:00–5:59 am (predawn), 6:00–9:59 am (morning), 10:00 am-1:59 pm (noon), 2:00–5:59 pm (afternoon), 6:00–9:59 pm (evening), and 10:00 pm-1:59 am (midnight). * P for trend <0.05. The difference between 2017-March 2020 and 1999–2000 and P for trend are reported in Supplemental Table 4.

Secular trend analyses showed decreasing trends in mean proportion of daily energy intake consumed at noon (from 26.8% to 24.7%) and midnight (from 5.2% to 4.6%), while an increasing trend (from 19.9% to 21.6%) in the afternoon (P≤0.028 for trend; Figure 2 and Supplemental Table 4), with no statistical change at other blocks. Similar secular trends were observed for the proportion of adults with energy intake within each block (Supplemental Figure 5).

Consistently, self-reported dinner contributed to the highest proportion of energy intake (35.6%–36.6% across years), followed by lunch (24.3%–25.8%), snack (20.9%–23.6%), and breakfast (15.3%–17.3%; Supplemental Figures 6–7). Results were largely consistent when using the average from two dietary recall data (Supplemental Figure 8).

From 1999 to March 2020, fasting consistently ended at 8:41–8:52 am and started at 8:34–8:51 pm. The midpoint of energy intake remained stable at 2:38–2:48 pm; energy intake duration ranged 11.9–12.2 hours (Figure 3).

Figure 3.

Fasting End and Start Time, Intake Midpoints, and Eating Durations for Energy and Macronutrients among US Adults by National Health and Nutrition Examination Survey (NHANES) Cycles from 1999 to March 2020. X-axes denote the NHANES survey cycle; left y-axes denote the clock time (in 24-hour format, e.g., 16:00 means 4:00 pm) for the mean fasting end/start time and intake midpoints, corresponding to the dots in the figure; right y-axes denote the eating duration (hours), corresponding to the bar in the figure. Sampling weights, stratification, and clustering of the complex sampling design were considered. Only participants who consumed certain macronutrients on the survey day were included. Between March 2020 and 1999, difference in fasting end time for carbohydrate intake was 0.20 (95% confidence interval, 0.01, 0.39) hours; difference in carbohydrate intake duration was −0.26 (−0.42, −0.10). There are declining trends in energy and protein intake duration (P<0.003 for trend), and the difference between March 2020 and 1999 was −0.12 (−0.28, 0.04) hours and −0.04 (−0.22, 0.13) hours, respectively. No secular trends were observed for other parameters.

Timing of Macronutrient Intake

Within each block, the proportion of adults reporting intake of each macronutrient was similar to that of energy intake (Supplemental Figure 5). Similar to energy intake, US adults consumed the highest proportion of daily macronutrient intake in the evening, followed by noon, afternoon, morning, midnight, and predawn between 1999 and March 2020 (Figure 2). For example, in 2017-March 2020, 31.6% of daily carbohydrates occurred in the evening, 25.0% at noon, 21.0% in the afternoon, 15.9% in the morning, 4.6% at midnight, and 1.9% in the predawn; within each block, proportions of carbohydrates, protein, and fat consumed relative to their respective daily intake totals were similar.

However, secular trends in eating timing varied by macronutrients. From 1999 to March 2020, the proportion of daily carbohydrate intake slightly decreased in the morning, noon, and midnight, while increasing in the afternoon and evening; all changes were within ±1.8% (P≤0.021 for trend; Figure 2 and Supplemental Table 4). A similar trend was observed for low-quality (including added sugars) rather than high-quality carbohydrates (Supplemental Figure 9). In contrast, there were declining trends in the proportion of daily protein (from 27.4% to 25.4%) and fat (from 27.9% to 25.3%) intake at noon, with minimal changes in proportions at other blocks (Figure 2). For protein subtypes, the proportion of daily animal protein intake at noon decreased over these years, while the noon proportion of plant protein intake remained stable; additionally, the morning proportion of animal protein intake increased, while that of plant protein intake decreased (Supplemental Figure 9). For all fat subtypes, the proportions of daily intake increased in the morning but decreased at noon over these years. Within each block, carbohydrates consistently contributed to the highest energy intake, followed by fats and proteins (Supplemental Figure 10).

Figure 3 shows that fasting start/end time, intake midpoint, and eating duration for macronutrients are similar to those of energy intake and largely stable between 1999 and March 2020. However, the mean fasting end time for carbohydrate intake was delayed by 12 minutes (from 8:48 to 9:00 am over these years; P=0.045), which shortened the mean carbohydrate intake duration by 0.3 hours (from 12.1 hours to 11.8 hours; P=0.002).

Eating Timing for Major Food Groups

For refined grains, total vegetables, red/processed meat, poultry, seafood, and lentils/nuts/soy, US adults consumed the highest proportion in the evening, followed by noon, afternoon, morning or midnight, and predawn (Figure 2). Particularly, morning, midnight, and predawn collectively contributed to <12% of daily intake for total vegetables, red/processed meat, poultry, and seafood, and <12.9% adults consumed these food groups during these three blocks (Supplemental Figure 5). In contrast, US adults consumed the largest proportion of whole grains in the morning (range, 32.8%–39.9%), followed by noon (19.6%–28.1%), evening (17.2%–20.3%), afternoon (10.8%–14.9%), midnight (4.3%–5.6%), and predawn (1.8%–3.3%). Whole fruit, fruit juice, egg, and dairy intake showed more balanced distributions across morning, noon, and evening, e.g., 23.9%–27.4%, 24.1%–27.4%, and 19.6%–24.5% across these three blocks for whole fruit intake, respectively. The mean HEI score ranged 29.6–39.4 across morning, noon, afternoon, and evening and 3.0–11.2 across predawn and midnight (collectively referred to as overnight; Supplemental Figure 11), suggesting low dietary quality during the overnight period.

Social Determinants of Eating Timing

Eating timing distribution for macronutrients and major food groups is largely consistent across age groups (Supplemental Figure 12); however, adults aged 35–64 years had longer durations for energy/macronutrient intake and ended fasting earlier compared to those aged 20–34 years (Supplemental Figure 13). Compared to females, males consumed greater proportions of energy/macronutrients in the evening, midnight, and predawn, but smaller proportions in the morning, noon, and afternoon (Supplemental Figure 14); males had overall later eating timing (later fasting end/start time and intake midpoint) and a longer macronutrient intake duration (Supplemental Figure 15).

Compared to non-Hispanic white adults, Hispanic adults consumed larger proportions of energy/macronutrients in the afternoon, larger proportions of fats in the morning, and smaller proportions of energy/macronutrients in the evening and predawn (Supplemental Figure 16); they ended fasting later and thus had shorter durations for all macronutrient intake (Supplemental Figure 17). Non-Hispanic black adults consumed smaller proportions of energy/macronutrients in the morning, noon, and evening, while larger proportions in the afternoon and midnight; they had overall later eating timing (later fasting end/start time and intake midpoint) and shorter eating durations for all macronutrients.

Generally, adults with favorable socioeconomic factors (i.e., high income, employment, and food security) ended their fasting earlier but had similar fasting start times, compared to their counterparts (Supplemental Figures 18–21). Consequently, they had earlier intake midpoints and longer durations for all macronutrients. They also consumed greater proportions of energy, macronutrients, and most food groups in the evening but smaller proportions in the afternoon or midnight (Supplemental Figures 22–25), and they consumed greater proportions of whole fruits, lentils/nuts/soy, or whole grains in the morning. Particularly, adults with evening/night/rotating shift had an overall postponed eating timing (i.e., later fasting end/start time and intake midpoint) compared to those with regular daytime work schedules (Supplemental Figures 26), and they had higher macronutrient intake proportions in the afternoon, midnight, and predawn, while lower proportions in other blocks, particularly for refined grains, red/processed meat, poultry, eggs, dairy, and vegetables (Supplemental Figures 27).

Associations between Eating Timing and Mortality

Among 24,993 participants with at least one valid dietary recall during 2005–2014, 22,030 had two valid recalls. We further excluded 1,300 participants followed up for <5 years, leaving 20,730 participants in this analysis (Supplemental Figure 28), and 1,783 deaths (520 from CVD, 409 cancer) were documented over 200,802 person-years (median=9.7 years).

Several findings supported that delayed eating timing is associated with higher mortality risk (Figure 4 and Supplemental Figure 29). First, results for the distribution of intake should be interpreted in a substitution manner, due to the adjustment of daily intake of energy and corresponding macronutrient/food group, and other macronutrient/correlated food group intake within the same block (detailed in Supplemental Methods). Reallocating 5% of daily energy intake from other blocks to midnight or predawn was associated with higher mortality. HRs (95% CIs) per 5% increase at midnight were 1.06 (1.01, 1.10) for all-cause, 1.09 (1.02, 1.17) for CVD, and 1.04 (0.93, 1.17) for cancer mortality; the corresponding HRs for predawn energy intake were 1.12 (1.03, 1.21), 1.06 (0.95, 1.19), and 1.22 (1.05, 1.41), respectively. Particularly, higher midnight intake proportions for carbohydrates, refined grains, whole fruits, and fruit juices were associated with higher CVD mortality; higher predawn intake proportions for proteins, red/processed meat, poultry, eggs, refined grains, vegetables, and fruit juices were associated with higher cancer mortality. These results were largely consistent with comparisons between consumers and non-consumers at midnight and predawn.

Figure 4.

Associations of Distribution for Energy and Macronutrient Intake with Mortality. Sampling weights, stratification, and clustering of the complex sampling design were considered. Analyses only included participants recruited in 2005–2014, because 1) participants with less than 5-year follow-up were excluded from the analyses to reduce the possibility of reverse causation, and those recruited after 2014 were excluded (mortality data were updated until the end of 2019); 2) only participants providing two 24-hour dietary recalls were included to more accurately reflect participants’ dietary habits, and the National Health and Nutrition Examination Survey (NHANES) only administered one 24-hour dietary recall for each participant before 2003; 3) the NHANES started to collect sleep information in 2005. Red dots with horizontal lines indicate the hazard ratios (HRs) with their 95% confidence intervals (CIs) related to per 5% increase in the intake in each block to daily intake. We adjusted for age, sex, race/Hispanic origin, education, income, employment status, food security, daily energy intake, daily intake of the examined macronutrient, number of eating occasions, Healthy Eating Index-2020 score, sleep duration, cigarette smoking, alcohol consumption, leisure-time physical activity, body mass index, and prevalent hypertension, high cholesterol levels, diabetes, cardiovascular disease, and cancer. Macronutrient intake at each block was mutually adjusted given high correlations (r range, 0.48–0.83).

Second, for energy intake, delayed fasting end and start times and intake midpoint were associated with higher CVD mortality, with HRs (95% CIs) per 1-hour delay of 1.08 (1.02, 1.15), 1.09 (1.02, 1.16), and 1.09 (1.02, 1.16), respectively (Supplemental Figure 29); similar associations were observed for all-cause mortality. Particularly, higher CVD mortality was only linked to delayed fasting start time and intake midpoint for proteins, but not for carbohydrates or fats. Third, consuming proteins, poultry, eggs, and refined grains in the morning was associated with lower cancer mortality.

Notably, associations between afternoon intake proportions and CVD mortality varied by macronutrient. Reallocating 5% of daily intake from other blocks to afternoon was associated with an HR (95% CI) of 0.93 (0.88, 0.98) for carbohydrates while 1.07 (1.00, 1.14) for proteins. Results remained similar after excluding participants with evening/night/rotating shifts (Supplemental Figure 30).

Discussion

Over the past 2.5 decades, US adults consistently consumed the largest proportions of energy, macronutrients, and most foods in the evening, followed by noon, afternoon, morning, midnight, and predawn; mean energy/macronutrient intake duration was ~12.0 hours. Previous NHANES-based studies reported similar patterns in 2009–2014 [8] and 2011–2018 [7] – highest energy intake at dinner and a mean eating duration exceeding 12 hours. By dividing the day into six 4-hour blocks based on clock time, our study overcame limitations of prior studies based on self-reported eating occasions, enabling comparability across populations and future studies [4]. We also analyzed more recent and extensive data (1999-March 2020) and employed a comprehensive framework incorporating timepoint, distribution, and duration.

Our findings raise concerns about late eating timing among US adults. Chrononutrition research increasingly links late-night eating and low morning intake to cardiometabolic risk [3]. Consistently, we observed that higher midnight/predawn intake of energy and several macronutrients/foods was associated with higher mortality, whereas protein, poultry, egg, and refined grain consumption in the morning was associated with lower cancer mortality. Despite a decreasing secular trend, midnight intake still contributed ~5% of daily energy, with 23.4% and 10.7% of adults consuming food at midnight and predawn, respectively. On average, fasting began after 8:34 pm, raising concerns about late-night eating. Moreover, we found that the evening consistently accounted for the highest proportion of daily energy, macronutrient, and most food group intake, with an increasing secular trend in the proportion of evening carbohydrate intake, while morning intake only contributed to <15% of daily energy.

By investigating the timing of macronutrient and food group intake, our study offered additional insights beyond energy intake timing alone. First, certain food groups have unique eating timing patterns, e.g., disproportionately higher proportions of whole grains and whole fruits consumed in the morning among US adults. These findings prompt a reconsideration of previously reported cardiovascular benefits associated with greater morning energy intake [3], as it remains unclear whether such benefits are attributable to healthier food consumption in the morning or to the timing itself. Our analysis adjusted for daily macronutrient/food intake and intake of macronutrients and correlated foods at each block, showing that only higher morning intake of refined grains and eggs was associated with lower cancer mortality, rather than total energy or other nutrients/foods, consistent with previous studies [9, 24]. This suggests that reallocating certain food intake to the morning might be associated with lower cancer mortality than broadly increasing morning energy intake.

Second, associations between eating timing and health outcomes may vary by nutrient/food group, even within the same time block. We observed opposite directions for associations of afternoon carbohydrate (protective) and protein (risk) intake with CVD mortality. Individuals with higher afternoon carbohydrate intake tended to consume less later in the day (Supplemental Figure 31), potentially benefiting from better glycemic responses earlier in the day [10]. Meanwhile, amino acid metabolism is more active in the afternoon [12], and branched-chain amino acid metabolites have been linked to cardiometabolic risk [25], which may explain the adverse associations with afternoon protein intake.

Male, non-Hispanic black, and socioeconomically disadvantaged adults tend to have less healthy diets defined by dietary quantity and quality [18, 26], and our findings add new evidence from the eating timing perspective – these populations consumed greater proportions of energy/macronutrients later in the day and had later intake midpoints, which are associated with adverse health outcomes. Dietary guidelines and interventions may need to place greater emphasis on reducing overnight intake and promoting earlier eating timing, particularly in these populations. Tailoring public health strategies to address both dietary quality and timing could help reduce disparities in diet-related health outcomes. For example, Women, Infants and Children (WIC) can integrate eating timing into the required individualized nutrition counseling, and Supplemental Nutrition Assistance Program (SNAP) could pilot targeted incentives (e.g., enhanced subsidies for healthy breakfast foods) to encourage earlier eating timing. Of note, previous studies observed shorter eating durations among Hispanic and non-Hispanic black adults compared to non-Hispanic white adults [7, 8], while our results suggest different causes – Hispanic adults ended fasting later but started fasting at a similar time, while non-Hispanic black adults had both delayed fasting end and start times, aligning with the lower evening and predawn intake proportions among Hispanic adults but higher midnight and lower morning intake proportions (linked to adverse health outcomes) among non-Hispanic black adults.

This study’s strengths include applying a comprehensive eating timing framework, examining macronutrient- and food-group-specific eating timing, leveraging nationally representative data collected during an extensive period, and considering multiple social features. However, several limitations should be acknowledged. First, although supported by prior research [16] and observed energy intake peaks, the time classification of six predefined 4-hour blocks needs validation in other studies. Future investigations using finer time bins (e.g., 2- or 1-hour) in studies with larger sample sizes are warranted to provide more precise evidence. Second, although 24-hour dietary recall is the most feasible way to collect eating timing data currently, measurement errors are inevitable, and affordable wearable devices combined with artificial intelligence techniques are warranted in future studies for more accurate dietary evaluation. Third, for the mortality analysis, although evidence reported moderate reliability in estimating the eating timing for each individual by using two dietary recalls, using more days of recall data can improve the reliability. Besides, two recalls were conducted within 3–10 days, which may not fully capture long-term eating timing behaviors relevant to mortality, and the results should be interpreted with caution. Fourth, given the observational design, causal inference between eating timing and mortality cannot be made, and reverse causation and residual confounding are possible. Although we adjusted for sleep duration in the main analysis and observed consistent results after excluding participants with evening/night/rotating shifts (2005–2010), residual confounding from unmeasured sleep timing may remain.

In summary, from 1999 to March 2020, US adults consumed the largest proportions of energy, macronutrients, and most foods in the evening, followed by noon, afternoon, morning, and overnight, with a quarter reporting midnight intake and a relatively late fasting start time (after 8:34 pm on average). Higher overnight intake and delayed eating timing might be associated with higher mortality, though associations varied by nutrients and foods. Future chrononutrition research should assess nutrient- and food-specific eating timing to inform evidence-based recommendations. Adults with lower socioeconomic status had greater midnight intake and delayed eating timing, suggesting nutritional disparities in eating timing. Future studies from other countries are warranted to enhance generalizability and guide global dietary recommendations, and objective dietary monitoring tools (e.g., smartphone applications) are needed to obtain long-term, accurate data. Cross-over interventional studies are also needed to evaluate whether reallocating specific macronutrients/foods to the morning, or advancing their intake timing, can improve health.

Abbreviations:

CI

confidence interval

CVD

cardiovascular disease

FNDDS

Food and Nutrient Database for Dietary Studies

FPED

Food Patterns Equivalents Database

HEI

Healthy Eating Index; HR, hazard ratio

MPED

MyPyramid Equivalents Database

MUFA

monounsaturated fatty acid

NHANES

National Health and Nutrition Examination Survey

PUFA

polyunsaturated fatty acid

SFA

saturated fatty acid

Data Availability

NHANES data and code book are available at https://wwwn.cdc.gov/nchs/nhanes/default.aspx. Analytic code will be made available by contacting Dr. Zhang (yanbo.zhang@einsteinmed.edu).