A data-driven trajectory analysis reveals associations between temporal patterns of within-day energy intake and mortality risk in U.S. adults: NHANES 2005–2018

Study population

As a series of cross-sectional surveys, NHANES was designed to assess nutrition and health status of the noninstitutionalized US population [25]. A nationally representative group of over 5,000 US individuals were recruited by NHANES at each year of the 2-year survey cycle using a stratified, multistage sampling design. All recruited participants were required to attend in-home interviews for questionnaire surveys and in-person visits to a Mobile Examination Center (MEC) for medical exams. The NHANES study protocols were approved by the Ethics Review Board of the National Center for Health Statistics (NCHS), and written informed consent was acquired from all participants [25].

An initial of 61,447 participants with available dietary data were identified across 7 cycles of NHANES (2005–06 through 2017–18). To assess survival status of the participants, we linked NHANES data with the 2019 public-use NCHS Linked Mortality File (LMF) [26], which contains mortality follow-up data from the date of MEC visit through December 31, 2019. All adults in NHANES with sufficient identifying data were eligible for the linkage to 2019 public-use LMF. The linkage rate in NHANES was over 99.5% across different cycles [27]. In this study, we first excluded participants aged under 19 years (n = 26,138), without two valid 24-hour dietary recalls as determined by NHANES standardized protocols (n = 4,856), and who were ineligible for the linkage to LMF (n = 52). Among the remaining 30,435 participants, we further excluded participants with missing demographic information (n = 40), pregnant women due to physiological changes that influence dietary intake and energy metabolism during pregnancy (n = 461), those with extreme daily energy intake (<800 or > 5,000 kcal/d in men and <500 or < 3,600 kcal/d in women; n = 498), and extreme (< 18 or > 50 kg/m²) or missing body mass index (BMI; n = 664). Our final analytical sample was comprised of 28,425 participants (eFigure 1).

Dietary Assessment

Dietary data was collected by two 24-h food recalls from each participant. Participants reported all food and beverage consumption with the clock time of intake during the previous 24 hours (from midnight to midnight of the previous day) [28]. Food recalls were collected under the supervision of trained interviewers. The first dietary recall (Day 1) was collected in person in the MEC using a standard set of measuring guides for the volume and dimensions of the food items consumed. The second recall (Day 2) was collected via telephone 3–10 days after the MEC visit. The US Department of Agriculture (USDA) Automated five-step Multiple-Pass Method (AMPM) was used during the dietary assessment to improve the accuracy of reported food recalls [29]. The energy and nutrient content of the reported food/beverage consumptions were estimated based on the USDA Food Surveys Nutrient Database. Further details of dietary assessment have been reported previously [30].

We extracted total daily energy and macronutrients intake from the collected 24-h recalls. Multiple chrononutrition behaviors were calculated for each recall, including the clock time of each self-identified meal (i.e., breakfast, lunch, and dinner), first and last intake time, eating window, and eating frequency. Briefly, first or last intake time was defined as the first or last food/beverage intake containing > 0 kcal within the 24-hour period of a day; eating window was defined as the length of time period between the first and last intake time; and eating frequency was defined as the total number of eating occasions per day (n/day) when both self-identified eating occasion and clock time changed [30]. Dietary intake and timing from the two 24-hour recalls were averaged to reflect the usual eating behaviors of the study participants [31].

Mortality Ascertainment

The mortality status was assessed by NCHS via the linkage to the National Death Index [32]. The underlying causes of death were derived from the International Classification of Diseases, 10^th Revision (ICD-10) codes. All-cause mortality was determined according to the vital status (variable “MORTSTAT”) in 2019 LMF. Cardiovascular disease (CVD) mortality was defined as death from heart or cerebrovascular diseases, and cancer mortality was defined as death from malignant neoplasms [33]. The follow-up time was defined as person-months from the MEC examination date to the date of death, censored, or end of the mortality follow-up period, December 31, 2019, whichever came first.

Potential confounders

Self-reported demographic characteristics included age, sex (men or women), race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, or others), education level (≤ high school, some college, or ≥ college graduate), marital status (never married, married/living with partner, or single again category that included windowed, divorced, and separated), poverty-income ratio (PIR; low: ≤1.0, middle: 1.0–3.0, and high: >3.0), and employment status (based on participants’ responses about their type of work done in the past week, categorized as ‘Have work’, or ‘Not working/looking for a job’). Body Mass Index (BMI) was calculated as weight (kilograms) divided by height (squared meters), and categorized as underweight (<18.5), normal (18.5–24.9), overweight (25.0–29.9), and obese (≥30.0 kg/m²). Participants also reported whether they attempted to lose weight in the past year (yes or no) and their perceived weight status as underweight, overweight, or about the right weight [34].

Dietary quality was assessed using the collected food recall by the 2015 Healthy Eating Index (HEI) [35] (range 0–100) with a higher score indicating a healthier diet. To account for within-person variation in food/beverage consumption, we extracted the recall days of the week on which the two dietary recalls were conducted (one weekday and one weekend, two weekdays, or two weekends) [36, 37]. Poor appetite or overeating over the past two weeks (not at all, several days, more than half the days, and nearly every day) and food security status (ranging from full to very low food security) [38, 39] were also collected.

Total physical activity was assessed via an interviewer-administered physical activity questionnaire during the MEC visit. We extracted time spent in vigorous and moderate leisure-time physical activity (MVPA, min/week) separately, based on American College of Cardiology/American Heart Association guidelines [40]. Alcohol consumption (yes/no) was assessed by the reported drinking frequency over the past 12 months. Smoking status (never, former, or current smoker) was acquired based on current smoking status and whether the participant had smoked ≥ 100 cigarettes in life. Self-reported usual sleep duration on weekdays (hours/day) and general health status (excellent, very good, good, fair, or poor) were also collected. Based on medical history questionnaires, we calculated a composite score (ranges from 0–13) to quantify the number of chronic conditions (including 13 conditions, i.e., hypertension, hypercholesterolemia, diabetes, kidney conditions, asthma, arthritis, congestive heart failure, coronary heart disease, stroke, emphysema, chronic bronchitis, liver conditions, and cancer) [41, 42].

Group-based trajectory modeling of within-day temporal patterns of energy intake

We used a group-based trajectory model [43] to identify distinct subgroups within the US adult population that follow similar temporal patterns of energy intake throughout the day, by modeling the heterogeneity in the population. This model is designed to classify individuals into clusters based on similar trajectories of the outcome of interest over time [43]. We first divided the 24-hour dietary intake data into four 6-hour intervals (i.e., 00:00–05:59, 06:00–11:59, 12:00–17:59, and 18:00–23:59) and calculated the energy intake within each interval for all participants. The 6-hour intervals were chosen after comparing trajectory results across different time intervals (1, 2, 3, 4, and 6 hours), with the 6-hour division yielding the most optimal results and the least discrepancy between the observed and fitted trajectory curves. Next, we calculated the proportion of total daily energy intake (% of kcal/day) consumed during each of the 6-h intervals using the averaged values from two recalls were used. A censored normal group-based trajectory model was employed to analyze the patterns of 6-hourly energy intake (% kcal/day) across the 24-hour period within a day. Here we identified four distinct trajectory groups that characterize temporal patterns of within-day energy intake, i.e., Skewed to Morning, Skewed to Midday, Skewed to Evening, and Midday-Evening Balance (Figure 1, Panel A). Each participant was assigned to the corresponding dietary energy trajectory groups based on the maximum likelihood estimation. Different numbers of groups and polynomial functions for each trajectory group were tested to determine the best-fitting model, with model fit assessed using the Bayesian Information Criterion (BIC) [44] and by ensuring that each trajectory group contained at least 5% of the overall population. Finally, spaghetti plots were generated to visualize between-individual variation in temporal patterns of energy intake across trajectory groups and to further assess the model fitting.

Statistical Analyses

NHANES complex sampling design, including cluster, strata, and weight, were accounted for in our analyses and Taylor series linearization was performed in variance estimates [45] to derive nationally representative associations.

Given the relatively low percentage of missing data across covariates, median or mode substitution [46, 47] was used to impute missing values of the following covariates with missing: smoking status (0.04%), PIR (7.9% missing), MVPA (0.1% missing), food security (1.3% missing), perceived weight status (0.3% missing), self-reported general health condition (4.9% missing), sleep duration (0.3%), and alcohol consumption (17.8% missing). Additionally, we conducted sensitivity analyses using multiple imputation as an alternative method to evaluate the potential impact of different imputation methods on our results.

Descriptive analyses were used to summarize the participant characteristics and dietary intakes according to the identified trajectory groups of temporal patterns of within-day energy intake. The differences across the groups were assessed using survey-weighted chi-square tests for categorical variables, and ANOVA for continuous variables.

Survey-weighted Cox proportional hazards regression was used to assess the association between trajectory groups of within-day temporal patterns of energy intake and all-cause and cause-specific (i.e., CVD and cancer) mortality risk. Hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) were estimated from unadjusted (Model 1) and two multivariable Cox models (Model 2 and 3), with covariates selected from potential confounders identified through correlation analyses. In Model 2, we adjusted our Cox models for age, age² (to account for the non-linear relationship between age and mortality [48]), sex, race, poverty-income ratio, BMI categories, total daily energy intake, diet quality, physical activity, alcohol intake, smoking status, sleep duration, food security, and self-report health status. In Model 3, we further adjusted our models for perceived body weight, marital status, and recall days of the week. The proportional hazards assumption was tested via using Schoenfeld residuals for each exposure and covariate.

To assess potential effect modification by demographic characteristics, we further conducted subgroup analyses to examine the associations between trajectory groups of temporal patterns of energy intake and mortality risk stratified by age (<70 or ≥70 years), sex (men or women), and race/ethnicity (Non-Hispanic White or Non-White) groups. Here, due to the low mortality rate among younger individuals (97 / 9,810) and middle-aged adults (1,053 / 13,514), we combined the young (20–39 years) and middle-aged (40–70 years) into one ‘young to middle-aged (< 70 years)’ group to ensure higher statistical power and reliable estimates. Similarly, four race/ethnicity categories (Hispanic, Non-Hispanic Black, and other/multiracial) were combined into the ‘Non-White’ group.

We also conducted several sensitivity analyses. First, to minimize the potential impact of reverse causality, we excluded participants who died within the first two years of the follow-up (n = 312) from our main analysis. Second, to assess the reliability of our findings regarding missing data handling, we performed multiple imputations with five rounds of imputations for all covariates with missing values. The imputation model included demographic and socioeconomic characteristics, anthropometric measures, and lifestyle factors as predictors. Third, we further adjusted our Model 3 for employment status using the variable ‘Type of work done last week’ (‘Have work’ or ‘Not working/looking for a job’) as a surrogate measure for employment condition and potential shift work. Fourth, we repeated group-based trajectory modelling after excluding participants who had extreme energy intake (≥ 50% of total daily energy intake) from 00:00–05:59 to assess the impact of nocturnal eating on our findings. All analyses were performed using SAS version 9.4 (SAS Institute, Inc.). Two-sided P values < 0.05 were considered statistically significant.

Trajectory groups of within-day temporal patterns of energy intake

Using group-based trajectory models, we identified four distinct trajectory groups of temporal patterns of within-day energy intake (Figure 1, panels A to B): (1) Midday-evening Balance: this group included 38.8% of participants, who distributed their daily energy intake almost evenly between midday and evening., (2) Skewed to Morning: this group comprised 17.5% of participants, who consumed majority of their daily energy intake before noon (12:00 PM), (3) Skewed to Midday: included 22.1% of participants, who consumed majority of their daily energy intake in the midday between 12:00 and 17:59, and (4) Skewed to Evening: comprised 21.6% of participants, who consumed majority of their daily energy intake in the evening after 18:00 PM (Figure 1, panel A). Figure 1, panel B shows participants in the same energy intake trajectory group, whose temporal energy intake patterns throughout the day were concentrated around the group’s average, with relatively low variability.

Figure 1, panel C shows the within-day temporal patterns of energy intake by trajectory groups. In Midday-evening Balance group, the average proportion of daily energy intake was 41% during midday (12:00–17:59) and 38% during evening (18:00–23:59). Participants in the Skewed to Morning trajectory had the highest proportion of daily energy intake (45% kcal/d) in the morning (06:00–11:59) while consumed their remaining daily energy intake (26–28% kcal/d) in the midday or evening. Those in the Skewed to Midday group consumed the majority of daily energy intake (62%) during midday, with smaller proportions of their energy intake (15–23%) was consumed in the morning or evening. Finally, participants in the Skewed to Evening trajectory group consumed the highest proportion of their daily energy intake (57%) in the evening, with only 20–21% consumed in the morning or midday (Figure. 1, panel C).

Participants characteristics by trajectory groups of daily energy intake

Table 1 shows the baseline characteristics of participants according to trajectory groups based on temporal patterns of within-day energy intake. Participants in the Skewed to Midday group were generally older, women, had lower education and income levels, and had more chronic conditions compared to the other groups. Participants in the Midday-evening balance group had higher education and income levels, lower BMI, slightly higher daily energy intake, were more physically active and healthier than the other groups. Participants in the Skewed to Morning group had the earliest first intake time, while those in the Skewed to Evening group had the latest last intake time. The shortest eating window and lowest eating frequency were observed in the Skewed to Midday group (Table 1).

Associations between within-day temporal patterns of energy intake and all-cause mortality

We used survey-weighted Cox proportional hazards regression to determine the associations between within-day temporal patterns of energy intake obtained from our group-based trajectory models and all-cause mortality, using the Midday-evening Balance group as the reference category (Figure 2). In the unadjusted model (Model 1), all skewed temporal patterns of energy intake (i.e., Skewed to Morning, Midday, and Evening) were associated with higher all-cause mortality risk compared to Midday-evening Balance reference group (all p-values ≤0.01). After adjusting our models for demographics, lifestyle factors, BMI, sleep duration, self-reported health condition, daily energy intake, diet quality, and food security in Model 2, only Skewed to Morning (p=0.037) and Skewed to Evening (p=0.023) remained significantly associated with higher all-cause mortality, compared to Midday-evening Balance reference group. In the fully adjusted Model 3, which additionally accounted for perceived body weight, marital status, and recall days of the week, these associations remained significant for both Skewed to Morning (HR=1.18, 95% CI, 1.01–1.38) and Skewed to Evening (HR=1.23, 95% CI, 1.03–1.47) (Figure 2).

We performed several sensitivity analyses to support the validity of our results (Table 2). First, after excluding participants who died within the first two years of follow-up, all three skewed temporal trajectory groups of energy intake remained significantly associated with higher all-cause mortality compared to the Midday-Evening Balance reference group in the unadjusted Model 1. After fully adjusting our models for all potential confounders in Model 3, Skewed to Evening group remained significantly associated with 28% higher risk of all-cause mortality (HR=1.28; 95% CI, 1.04–1.56; p=0.018), while the Skewed to Morning showed a marginal association with higher all-cause mortality risk (HR=1.18; 95% CI, 1.00–1.40; p=0.051). Second, we performed multiple imputation instead of mode imputation for covariates with missing values. Consistent with results from our main analyses, Skewed to Morning was associated with 18% (p=0.034) and Skewed to Evening was associated with 23% (p=0.021) higher all-cause mortality risk. Third, further adjusting our models for employment status as a surrogate measure for regular/irregular work schedules did not change the observed relationships. That is, Skewed to Morning and Skewed to Evening energy intake patterns were still associated with 17% and 22% greater mortality risk, respectively (Table 2). Finally, we identified a total of 23 participants (0.1%) with ≥ 50% of total daily energy intake during 00:00–05:59 in our analysis. Excluding those participants from our analysis did not alter the four temporal patterns of energy intake identified via group-based trajectory modeling (eFigure 2). In addition, Skewed to Morning and Skewed to Evening patterns remained significantly associated with 19% and 23% greater mortality risk, respectively, after fully adjustment (Table 2).

We also conducted subgroup analyses to assess whether the associations between within-day temporal patterns of energy intake and all-cause mortality are different by age, sex, and race/ethnicity population subgroups (Figure 3 and eTable 1). In older adults (aged ≥70 years), all three skewed temporal trajectory patterns of energy intake were significantly associated with higher all-cause mortality risk compared to Midday evening balance group in our unadjusted Model 1. In the fully adjusted Model 3, only the associations for Skewed to Morning (p=0.046) and Skewed to Midday (p=0.024) remained significant, while the association for Skewed to Evening was attenuated (p=0.096) (Figure 3 and eTable 1). In contrast, in young to middle-aged adults (aged 20–70 years), Skewed to Midday and Skewed to Evening were significantly associated with higher all-cause mortality in Model 1, but none of these associations remained significant in Model 3 (Figure 3 and eTable 1). Of note, to minimize the influence of non-diet-related factors such as accidents and injuries on mortality risk, we conducted a sensitivity analysis excluding participants aged <40 years. This exclusion did not alter the results, as no significant associations were observed between temporal patterns of within-day energy intake and all-cause mortality in adults aged 40–70 years (eTable 1).

In our sex-stratified analysis, in women all three skewed temporal patterns of energy intake (i.e., Skewed to Morning, Midday, and Evening) were associated with higher all-cause mortality risk compared to the Midday-evening Balance reference group in Model 1 (Figure 3 and eTable 1). These associations remained significant in fully adjusted models 2 and 3, with Skewed to Morning, Midday, and Evening were associated with 28%, 27%, and 59% higher mortality risk, respectively (all p<0.05). In contrast, for men, despite the significant association of Skewed to Morning and Midday with higher mortality in Model 1, none remained significant after adjusting for covariates (Figure 3 and eTable 1). When stratified by race/ethnicity groups, among Non-Hispanic White adults, all three skewed temporal patterns of energy intake were associated with higher all-cause mortality, compared with Midday-evening Balance group in Model 1. After adjusting our model for all the covariates in Model 3, only the association for Skewed to Evening (p=0.007) remained significant. In Non-White adults (i.e., Non-Hispanic Black, Hispanic, and Others), although Skewed to Morning and Midday were associated with higher mortality in Model 1, none remained significant in Model 3 (Figure 3 and eTable 1).

Associations between within-day temporal patterns of energy intake and cause-specific mortality

Table 2 shows the associations between temporal patterns of within-day energy intake distribution, CVD, and cancer mortality. For CVD mortality, we observed that Skewed to Morning and Skewed to Midday showed 86% and 105% higher mortality risks compared to Midday-evening Balance group in Model 1 (all p<0.001; Table 2). However, these associations were attenuated in fully adjustment Model 3, with none reaching significance. Notably, there was a suggestion of a positive association between the Skewed to Morning group and higher CVD mortality, with HR of 1.42 (95% CI: 0.96–2.10), but the p-value was only marginally significant (p=0.079).

For cancer mortality, in Model 1, only Skewed to Midday pattern of energy intake was significantly associated with 56% higher mortality risk compared to the Midday-Evening Balance group (p<0.001). However, the association was attenuated and no longer significant in Model 2 and 3. We did not observe significant associations between Skewed to Morning or Evening and cancer mortality risk in either unadjusted or multivariable models (Table 2).

Daily energy intake Skewed to Evening was associated with higher mortality risk

In this study, we observed that the skewed daily energy intake to evening was consistently associated with higher risk of all-cause mortality compared to the Midday-evening Balance trajectory. Consistent with our finding, a large-scale prospective cohort study also suggested that a higher energy intake consumed at dinner was associated with higher all-cause mortality among diabetes patients, but the reference group was high energy intake at breakfast [53]. Other studies reported that having energy consumption skewed towards the later than earlier times of the day was associated with higher blood pressure [54], blood triglycerides [55], systemic inflammation [56], and higher risk of obesity [23, 57, 58] and metabolic syndrome [7]. However, we did not observe a significant association between the Skewed to Evening pattern and CVD mortality. It is possible that the relatively small proportion of participants in the Skewed to Evening group (2.4%; 148 events among 6,152 individuals) limited statistical power to detect such associations. Additionally, our study population was relatively young (mean age around 46 years, with 36–38% aged under 40) compared to other studies, which may have limited the ability to capture CVD mortality events that typically occur later in life.

Furthermore, our study demonstrated that a balanced daily energy intake during midday and evening was associated with the lowest all-cause mortality risk, compared to patterns skewed toward morning or evening. Mechanistically, our body’s metabolic rhythms are optimized for more efficient nutrient digestion and absorption at certain periods of the day [59]. For example, glucose tolerance declines throughout waking hours and reaches its lowest point during sleep [60]. This might suggest distributing energy intake evenly throughout the day, rather than concentrating it in the evening, may better align with our body’s natural metabolic rhythms and improve glucose regulation [59]. On the other hand, evidence has shown that having excessive energy intake towards evening and night may disrupt the rhythm, potentially leading to metabolic disturbances and increased risk of various chronic diseases [61, 62].

Daily energy intake Skewed to Morning was associated with higher mortality risk

We observed that energy intake skewed toward morning was also associated with higher all-cause mortality risk compared to the Midday-evening Balance energy intake pattern, which might be potentially driven by CVD mortality. To our knowledge, no prior study has directly assessed this association; however, some studies have explored its relationship with CVD risk factors. Supporting our findings, a longitudinal study in overweight/obese adults (N=383) aged 55–75 found that consuming large breakfasts (>30% of daily energy intake) was associated with higher BMI, waist circumference, and triglyceride levels over 36 months compared to regular-sized breakfasts (20–30% of daily calories) [63].

In contrast, short-term small-scale randomized trials among overweight/obese women reported benefits of large breakfasts (50–70% of daily calories) over large dinners, including greater weight loss, improved glucose and triglyceride levels, and enhanced insulin sensitivity [12, 64]. These discrepancies may stem from differences in study populations, as prior studies largely involved overweight or obese individuals, whose dietary needs differ from the general population. In addition, recent trials of eTRE, where most energy intake is shifted toward earlier hours, typically from morning to early afternoon, have demonstrated beneficial metabolic effects, including improvements in fasting glucose, reductions in blood pressure, and weight loss [51, 65]. However, inconsistencies remain, as some studies reported no benefits in fasting glucose [19, 20, 66] and diastolic blood pressure [19, 66, 67], or lipid measures [19, 66]. These findings suggest the need for further research to clarify the impact of morning-skewed energy intake on long-term health outcomes.

We observed stronger associations between energy intake patterns skewed toward morning or evening and mortality risk among older adults, women, and White adults. These differences in associations across population subgroups may reflect physiological and behavioral mechanisms specific to these groups. In older adults, for example, maintaining a steady energy supply throughout the day might be particularly crucial due to their reduced physiological reserves and decreased muscle mass [68]. Women’s heightened sensitivity to skewed daily energy intake patterns might be attributed to sex-specific differences in glucose homeostasis, insulin signaling, and hormonal cycles [69]. Among White individuals, lifestyle factors, such as structured occupational routines or earlier sleep onset time [70] may further amplify the adverse effects of an evening-skewed intake. These findings underscore the need for more personalized dietary recommendations to promote longevity. Additionally, it is important to note that we did not observe significant associations between within-day temporal patterns of energy intake and mortality in other population subgroups, such as men. This lack of significant findings highlights the need for further studies to explore the potential differences in dietary impacts across diverse demographic profiles.

While we observed significant associations between energy intake patterns skewed to the morning or evening and higher all-cause mortality risk, we acknowledge that our cause-specific findings did not reach statistical significance in fully adjusted models. However, hazard ratios for CVD and cancer mortality remained above 1.0 across all skewed patterns, and the Skewed to Morning pattern showed a marginally significant 42% higher CVD mortality risk (p=0.079). These findings suggest that limited statistical power due to the lower incidence of CVD- or cancer-related deaths in our study, rather than absence of a true association, may explain the null cause-specific results.

Strengths and limitations

Our study presents several notable strengths. To the best of our knowledge, it is the first study to use a data-driven approach, specifically group-based trajectory modeling, to identify distinct within-day temporal patterns of natural energy intake distribution and assess their long-term association with mortality in US adults. The prospective study design and utilization of the large-scale nationally-representative data enhance the generalizability of our findings. The validation of our findings is further supported by multiple sensitivity analyses, which consistently produced similar results, suggesting the reliability of our conclusions.

Our study also has some limitations. First, the dietary assessment relied on 24-hour recall data, which may be subject to recall bias and might not fully capture long-term dietary patterns. However, the 24-h recall data in NHANES was collected by trained researchers using the USDA-valid method [29], which is considered one of the most reliable tools to precisely capture dietary intake. Importantly, our study builds on decades of analysis of food timing data from NHANES, with published results demonstrating the accuracy and reliability of 24-hour food recall data for assessing the timing of intake [71–73]. This underscores the relevance of our approach to investigating dietary behaviors. Also, the prospective design of our study ensured that within-day temporal patterns of energy intake distribution were identified before the occurrence of outcomes, minimizing the potential for reverse causation. However, because dietary data in NHANES were collected at a single time point, changes in dietary habits during the follow-up period could lead to an underestimation of the true effects on mortality. Second, 24-hour dietary recalls are prone to measurement error due to day-to-day variation in intake. To address this, we averaged the two non-consecutive recalls to estimate usual intake, and adjusted for recall day type (weekday vs. weekend) as a covariate. However, we acknowledge this does not fully capture within-person variability. Future research incorporating repeated dietary assessments over time and more frequent/consecutive dietary assessments are warranted to better characterize temporal patterns of energy intake and further validate and refine the findings.

Third, despite the comprehensive list of confounders we controlled in our analysis, including sleep duration, diet quality, intake day of the week, and various sociodemographic and lifestyle factors, we cannot entirely rule out residual confounding. For example, variations in occupation (e.g. night shift work) and chronotype may affect within-day temporal patterns of energy intake. However, the NHANES shift work variable (OCQ265) was only collected during 2005–2010 and discontinued thereafter, with substantial missing data (~45–49%) even within those cycles. Therefore, we adjusted our statistical models for employment status as a surrogate measure for shift work in the sensitivity analysis, and the findings remained consistent. Fourth, our outcome all-cause mortality may include deaths unrelated to diet, such as accidents. To partially address this, we excluded younger participants (aged 20–39 years) whose mortality may be more influenced by external and non-dietary factors from the analysis, but it has minimal impact on our results. We also assessed CVD and cancer, the leading causes of death, but their low incidence in our sample (2.1% for CVD and 1.9% for cancer) limits statistical power to observe significant results. Future studies with more comprehensive occupational (e.g. shift work) and circadian measures in a larger population with more adequate event cases would meaningfully strengthen the findings.

Conclusions

Overall, our findings revealed a significant relationship between within-day temporal patterns of energy intake and mortality risk in U.S. adults, with energy intake skewed toward morning or evening associating with higher all-cause mortality risk, compared to a balanced intake during midday and evening. These associations were independent of the content of the diet, and were more pronounced in older adults, women, and White individuals. Our study suggests the potential of considering the timing and temporal distribution patterns of daily energy intake into dietary recommendations, offering valuable public health insights into chrononutrition strategies to promote long-term health and longevity. The findings further suggest that tailoring dietary approaches to specific demographic subgroups could improve their effectiveness. However, more research is warranted to elucidate the causal relationships between within-day temporal patterns of energy intake and health outcomes through longitudinal and intervention studies, as well as investigating underlying biological mechanisms behind these relationships.