Poor dietary habits and a sedentary lifestyle are prominent risk factors for non-communicable diseases, including type 2 diabetes, CVD and cancers(Reference Danaei, Ding and Mozaffarian1). American adults who adhere to national recommendations for a healthy diet and regular physical activity tend to have significantly lower rates of cardiovascular morbidity and mortality compared with those who do not(Reference Loprinzi, Smit and Mahoney2,Reference Patnode, Redmond and Iacocca3) . Hence, adopting a healthy lifestyle, such as maintaining healthy dietary habits and engaging in regular physical activity, may effectively decelerate the progression of these non-communicable diseases, curtail mortality and thereby mitigate the socio-economic burden associated with these ailments(Reference Ding, Lawson and Kolbe-Alexander4,Reference White, Greene and Kivimaki5) .
In the realm of chrononutrition, the intricate interplay among food intake, circadian rhythms and metabolism is a focal point, underscoring the significance of suitable eating schedules for human health(Reference Asher and Sassone-Corsi6). Factors such as meal frequency, regularity, the temporal distribution of energy or nutrient intake and the timing of daily eating periods are considered in chrononutrition(Reference O’Connor, O’Connor and Higgins7). Recently, the influence of eating duration on the development of non-communicable diseases has garnered increasing attention. As a promising method for ameliorating metabolic disorders, time-restricted eating denotes a dietary strategy that restricts eating duration to a specific window of 8–12 h or less within a 24-h period(Reference Mihaylova, Chaix and Delibegovic8,Reference Di Francesco, Di Germanio and Bernier9) . Recent human studies have demonstrated that time-restricted eating may confer diverse metabolic benefits, primarily in body weight loss and adipose tissue reduction(Reference Peeke, Greenway and Billes10–Reference Wilkinson, Manoogian and Zadourian12). Additionally, experiments on Drosophila and mice further revealed that reducing feeding duration can have a long-term impact on health, such as extending lifespan, especially when combined with restricted energy intake within a specific time frame(Reference Acosta-Rodríguez, Rijo-Ferreira and Izumo13–Reference Hodge, Meyerhof and Katewa15). Nonetheless, the evidence from human studies with extended durations and larger sample sizes remains rare to definitively ascertain the long-term impact of eating duration on all-cause mortality.
Aside from eating duration, the daily level of physical activity is also intricately linked to health. A sedentary lifestyle is regarded as the fourth major risk factor for global mortality, following only hypertension, smoking and elevated blood glucose levels(Reference Park, Moon and Kim16). There is compelling evidence indicating that a sedentary lifestyle could increase susceptibility to numerous chronic ailments and mortality(Reference Lee, Shiroma and Lobelo17,Reference Biswas, Oh and Faulkner18) . However, consistent results have not been observed across previous studies regarding the association between physical activity and the risk of mortality(Reference Del Pozo Cruz, Ahmadi and Lee19–Reference Wen, Wai and Tsai24). Multiple studies have demonstrated that participating in vigorous physical activity, regardless of its intensity, duration or type, is associated with a dose-dependent reduction in the risk of mortality(Reference Lee, Shiroma and Kamada20–Reference Ekelund, Tarp and Steene-Johannessen22). However, a 5-year human trial from Norway found no association between high- or moderate-intensity exercise and decreased all-cause mortality among the elderly(Reference Slomski23). Furthermore, the understanding of how eating duration and physical activity jointly contribute to the risk of all-cause mortality is limited, with only a few prospective studies exploring the combined effects of diet and physical activity on this outcome(Reference Alvarez-Alvarez, Zazpe and Pérez de Rojas25–Reference Kazemi, Sasani and Mokhtari27). The findings indicated that following either a physically active lifestyle or a healthy diet could reduce the risk of all-cause mortality, but the greatest potential for risk reduction lies in combining both healthy behaviours(Reference Kazemi, Sasani and Mokhtari27). Notably, it is still largely undetected whether different degrees of physical activity have a diverse impact on the effect of eating duration on all-cause mortality.
To date, most human studies on the impact of eating duration on health are of relatively limited sample sizes and/or short follow-up periods(Reference Mihaylova, Chaix and Delibegovic8), leaving us with a knowledge gap on the long-term impact of eating duration on the risk of all-cause mortality. In this study, a representative sample of American adults from the National Health and Nutrition Examination Survey (NHANES) was utilised to investigate the hypothesis that eating duration may have long-term effects on the risk of all-cause mortality, as well as the potential impact of different levels of energy intake and physical activity on these effects.
Materials and methods
Study population
This study was conducted utilising data from NHANES, a nationally representative survey conducted using a stratified and multistage sampling method, with detailed descriptions available elsewhere(Reference Hou, Han and Sun28). The survey was approved by the Ethics Review Board of National Center for Health Statistics, and all participants provided written informed consent(Reference Frank, Jaacks and Adair29). Study participants underwent standardised face-to-face interviews, followed by physical examinations and laboratory tests, to obtain data on socio-demographics, dietary intake, health-related inquiries and physiological measurements and laboratory results. The procedures were conducted in a mobile examination centre under the supervision of trained medical personnel.
A total of 101 316 participants were enrolled in this study, including NHANES data from 1999 to 2018. To ensure the stability of the results, exclude those with missing follow-up data on all-cause mortality status (n 42 252), missing 24-h dietary recall data (n 6662), unreliable total energy consumption (<600 or >5000 kcal/d, n 1606)(Reference Frank, Jaacks and Adair29–Reference Lan, Wang and Ehrhardt31), unusual eating duration (<1 h or >23 h, n 221), currently pregnant (n 1473) and missing one or more covariates data (n 24 618). The final number of participants included in our study was 24 484 (Fig. 1).
Fig. 1. Flow chart of participants inclusion from the National Health and Nutrition Examination Survey 1999–2018.
Collection of dietary information
From 1999 to 2002, only one 24-h dietary review interview was held, while a second dietary review interview was held about 3–10 d after the first dietary review interview between 2003 and 2018. During the interview, participants were asked to report the type, quantity and intake time of each food. To ensure the consistency of the data from the two investigation periods, data from the first dietary review interview were adopted in this study(Reference Wang, Martínez Steele and Du32).
As the primary exposure variable, eating duration was defined as the duration spanning from the initial time of the first meal to the finished time of the last meal(Reference Meth, Egmond and Moulin33,Reference Lu, Cang and Liu34) . To obtain the eating duration, we simply subtracted the finished time of the last food from the initial time of the first food. For instance, if the first food was consumed at 8:00 and the last food was consumed at 20:00, the final eating duration would be 12 h. To maintain uniformity of the analysis, all times were converted into hours (i.e. intake at 7:30 h was recorded as 7·5 h). The study focused on eating duration within a single wake-up period, so we established the beginning of a behavioural day at 5.00 hours.
Main outcomes
The primary outcome variable of this study was all-cause mortality, which was derived from the National Death Index by 31 December 2019. The National Death Index is a centralised database managed by the National Center for Health Statistics and contains death record information from across the country, utilising identifying details such as social security numbers to ensure accuracy. The link between NHANES and the National Death Index was established through a rigorous process that involved probability matching and a comprehensive review of death certificates.
Covariates
A standardised household interview questionnaire was used to collect socio-demographic and lifestyle information and potential covariates including age, sex, race, education, family income:poverty ratio, intake day, energy intake, physical activity, drinking status and smoking status. Race was divided into five groups: Mexican American, other Hispanic, non-Hispanic white, non-Hispanic black and other races. Education was categorised into five groups: less than 9th grade, 9–11th grade, high school grade or general educational development, some college or associate of arts degree and college graduate or above. Family income uses the poverty ratio to classify as <1·3, 1·3–2·4, ≥2·4. According to the question: ‘Intake day of the week?’, intake day is divided into weekends and workdays. According to the calorie level ranges that are estimated based on sex and age, energy intake was classified into three groups: low, reference and high energy intake levels(Reference Phelan, Joyce and Bode35). Physical activity was categorised into three levels based on tertiles: ‘lightly active’ for less than 600 metabolic equivalents (MET)/week, ‘moderately active’ for 600–2400 MET/week and ‘highly active’ for more than 2400 MET/week. According to the question: ‘Had at least 12 alcoholic drinks a year?’, participants were divided into drinkers and non-drinkers. According to the question: ‘Smoked at least 100 cigarettes in life?’, participants were classified as smokers and non-smokers. The professionals measured participants’ standing body height and body weight at the mobile examination centre. BMI was calculated as body weight (kg) divided by height squared (m2).
Chronic metabolic diseases such as diabetes, hyperlipidaemia, hypertension and obesity were defined by medicinal history, relative physiological measurements and laboratory data.
Statistical analyses
Data are presented as the medians (IQR) for the continuous variables and unweighted numbers (weighted percentages) for the categorical variables. Continuous baseline characteristics were compared using general linear models, while x 2 tests were employed to assess categorical variables. Cox proportional hazards regression models, accounting for complex weighted designs, were utilised to estimate hazard ratios (HR) and their corresponding 95 % CI for all-cause mortality across quartiles of eating duration. Three statistical models were fitted to comprehensively describe the association between eating duration and all-cause mortality. In model 1, we adjusted for age, sex and race. In model 2, we incrementally adjusted for BMI, education level, family income:poverty ratio, drinking status and smoking status. In model 3, we further adjusted for intake day, energy intake and physical activity level. Moreover, a restricted cubic spline regression with five knots was utilised to examine the potential non-linear relationship between eating duration and all-cause mortality based on Cox proportional hazards regression analysis. Stratified analyses were conducted to estimate the relationship between eating duration and all-cause mortality by energy intake and physical activity levels. The statistical significance of interactions was assessed by examining the P-values associated with the product terms between eating duration and the stratification variables.
Sensitivity analyses were conducted to evaluate the robustness of the results. First, participants with <2 years of follow-up were excluded to mitigate potential bias from reverse causality. Second, early-morning eaters (eating duration >20 h) and participants with unusual first mealtimes (<6.00 hours or >11.00 hours) were excluded. Third, given the potential influence of other eating times, we further adjusted for the first and last mealtime to account for their potential confounding effects on the association of interest. Finally, we verified the results among participants with any of the following metabolic diseases: diabetes, hyperlipidaemia, hypertension and obesity.
All statistical analyses were conducted using R software (version 4·3·2). A two-tailed P-value < 0·05 was considered statistically significant for all analyses.
Results
The median (IQR) eating duration for the final analytic participants was 12·5 (11·0, 14·0) h. Table 1 provides a summary of the baseline characteristics of the final adult participants (n 24 484) sorted according to quartiles of eating duration. Participants with a shorter eating duration tended to be younger, female and of Mexican American descent and consumed lower amounts of dietary energy. On the other hand, those with a longer eating duration were predominantly non-Hispanic white Americans, who tended to have higher levels of education and family income and were fond of physical activity, drinking and smoking. Additionally, participants were more likely to have shorter dietary periods on weekends. Notably, no significant differences were observed in BMI and drinking status across the four groups.
Table 1. Baseline characteristics of participants according to eating duration quartiles in the National Health and Nutrition Examination Survey (1999–2018)†
†Data are presented as the medians (IQR) for the continuous variables and unweighted numbers (weighted percentages) for the categorical variables.
*P-values were derived from general linear models for continuous variables and x 2 tests for categorical variables.
GED, general educational development; AA, associate of arts; MET, metabolic equivalent.
The median follow-up time was 125 (77, 177) months, during which 2896 death events were observed. Using Cox proportional hazards regression analyses in complex weighted designs, we found a significant association between longer eating duration and lower all-cause mortality (Table 2). For each additional hour of eating duration, there was a 2 % lower risk of all-cause mortality in the full multivariable-adjusted model (HR 0·98; 95 % CI 0·96, 1·00; P < 0·05). The full multivariable-adjusted HR (95 % CI) across quartiles of eating duration were 1·00 (reference), 0·83 (0·73, 0·94), 0·85 (0·74, 0·97) and 0·87 (0·75, 1·00) for all-cause mortality. Figure 2 shows the non-linear dose–response relationship between eating duration and all-cause mortality (P for non-linearity < 0·05). Within the eating duration range of 8·9 h to 15·3 h, a distinct U-shaped relationship was observed between HR and eating duration, with the lowest HR at 12·6 h. Beyond this range, the HR was likely to exhibit a negative correlation with eating duration. Furthermore, an interaction analysis was conducted to investigate the effect of different levels of energy intake and physical activity on the relationship between eating duration and all-cause mortality. However, the results of the interaction test were not significant for energy intake and physical activity levels (P for interaction > 0·05). To further explore this, stratified analyses were conducted. Figure 3 shows the dose–response relationship of eating duration with all-cause mortality in different energy intake and physical activity groups. The non-linear dose–response relationship can still be found among participants with high energy intake and lightly active physical activity levels (P for non-linearity < 0·05). The restricted cubic spline curve in the lightly active physical activity group was similar to that of the integrated population (Fig. 3(d)). When the eating duration was below 12·4 h, the HR was negatively correlated with eating duration. However, when the eating duration was between 12·4 h and 14·1 h, HR was positively correlated with eating duration and then negatively correlated again (Fig. 3(c)). In contrast, the associations between eating duration and the risk of all-cause mortality were not statistically significant among participants with low or reference energy intake levels, as well as moderately or highly active physical activity levels (P for non-linearity > 0·05).
Table 2. Hazard ratios (HR) (95 %CI) for all-cause mortality were estimated based on eating duration among participants
Model 1: adjusted for age, sex and race.
Model 2: adjusted (from model 1) for BMI, education level, family income:poverty ratio, drinking status and smoking status.
Model 3: adjusted (from model 2) for intake day, energy intake and physical activity levels.
Fig. 2. Associations between eating duration and all-cause mortality among participants. Hazard ratios were adjusted for age, sex, race, BMI, education level, family income:poverty ratio, drinking status, smoking status, intake day, energy intake and physical activity levels. P non-linearity < 0·05.
Fig. 3. Associations between eating duration and all-cause mortality in different levels of energy intake and physical activity. Low energy intake (a), reference energy intake (b) and high energy intake (c). Model adjusted for age, sex, race, education level, family income:poverty ratio, drinking status, smoking status, intake day and physical activity levels. Lightly active physical activity (d), moderately active physical activity (e) and highly active physical activity (f). Model adjusted for age, sex, race, education level, family income:poverty ratio, drinking status, smoking status, intake day and energy intake levels. Different levels of energy intake and physical activity.
In sensitivity analyses, the non-linear dose–response relationship of eating duration with all-cause mortality was more pronounced when excluding participants with a follow-up time of <2 years (online Supplementary Fig. 1). However, the dose–response relationship exhibited a U-shaped curve after excluding early-morning eaters (eating duration >20 h) (online Supplementary Fig. 2). The non-linear dose–response relationship remained even after excluding participants with unusual first mealtimes (<6.00 hours or >11.00 hours), although the shape of the curve was slightly altered (online Supplementary Fig. 3). Similar results were also observed when we further adjusted for the first mealtime or the last mealtime (online Supplementary Figs. 4–5). Finally, the non-linear dose–response relationship was more pronounced among a total of 19 659 participants who had any of the metabolic diseases (diabetes, hyperlipidaemia, hypertension and obesity), with the curve remaining largely unchanged compared with the entire population (online Supplementary Fig. 6).
Discussion
To our knowledge, rare prior studies have explored the association between eating duration and all-cause mortality among US adults. Using data from NHANES and the National Death Index, our hypothesis was confirmed by demonstrating a negative association between eating duration and all-cause mortality. Furthermore, a non-linear dose–response association was observed between eating duration and all-cause mortality, and stratified analysis revealed differences in this association among populations with varying levels of energy intake and physical activity. Interestingly, the non-linear dose–response association between eating duration and all-cause mortality was observed only among participants with high energy intake or lightly active physical activity.
Previous studies have suggested that a shorter eating duration is associated with a lower risk of cardiometabolic diseases(Reference Anton, Lee and Donahoo36–Reference Moro, Tinsley and Bianco38). There is compelling evidence demonstrating that time-restricted eating could extend the healthy lifespan of animals(Reference Acosta-Rodríguez, Rijo-Ferreira and Izumo13,Reference Ulgherait, Midoun and Park14) , but sparse evidence exists regarding the associations between eating duration and all-cause mortality in humans. This nationally representative study, to the best of our knowledge, is the first to explore the associations between eating duration and all-cause mortality among adults. Therefore, there are no directly comparable studies available to provide corroborative assessments. In this study, Cox proportional hazards regression analyses revealed a negative correlation between eating duration and the risk of all-cause mortality. Some possible reasons have been speculated to explain this finding. First, individuals with a shorter eating duration might lack synchronisation for biological and social times. A vital study reported that the lifespan-extending effect of intermittent time-restricted feeding on Drosophila was closely related to enhanced transcription of specific genes regulated by circadian. It was shown that the induction of autophagy specifically during nighttime is essential for the lifespan extension effect of intermittent time-restricted feeding in flies, but this beneficial effect was absent in autophagy specifically induced during daytime(Reference Ulgherait, Midoun and Park14). Second, many individuals shortened their eating duration by reducing the amount of food they consumed and shortening intervals between meals; the above two dietary behaviours were reported to be associated with a higher risk of all-cause mortality(Reference Sun, Rong and Liu39). Additionally, it was observed that individuals with shorter eating duration consumed relatively less daily energy in our study. Third, in an observational study, a shorter eating duration was often accompanied by a delayed first mealtime (after 8.00 hours), which was associated with an increased risk of all-cause mortality(Reference Kant and Graubard40,Reference Fishbein, Knutson and Zee41) . A randomised controlled trial study also showed that early time-restricted eating had greater benefits for metabolic health than midday fasting(Reference Xie, Sun and Ye42). Hence, a longer eating duration may be linked to a reduced risk of all-cause mortality in natural conditions, which needs to be confirmed by additional studies. Moreover, a U-shaped dose–response relationship between eating duration and the risk of all-cause mortality was observed, with the lowest HR being observed at 12·6 h within the range of 8·9 h to 15·3 h. Previous studies on the dietary regimes of Americans revealed that the typical daily eating duration was between 9 h and 14 h, with only a small percentage extending it to 16 h or longer(Reference Kant43–Reference O’Connor, Reedy and Graubard45). Therefore, the U-shaped relationship observed in our study provided valuable insight into the optimal range of eating duration for minimising the risk of all-cause mortality.
Several cohort studies have shown a correlation between consuming energy intake beyond an individual’s needs and an elevated risk of mortality(Reference Nagai, Ohkubo and Miura46,Reference Lassale, Hernáez and Toledo47) . Similarly, our findings demonstrated a significant association between consuming a high-energy diet within a short duration and an elevated risk of all-cause mortality. Furthermore, the non-linear association between eating duration and all-cause mortality was solely found in individuals with a high energy intake. Therefore, our study highlighted the importance of avoiding excessive energy intake within a short eating window. Given that physical activity is widely recognised as one of the most important strategies for individuals to improve their health(Reference Ding, Mutrie and Bauman48–Reference Bull, Al-Ansari and Biddle50), stratified analyses were conducted to assess the correlation between eating duration and all-cause mortality by physical activity level. Ample evidence demonstrates that increased physical activity, both recreational and non-recreational, is linked to a reduced risk of mortality(Reference Lear, Hu and Rangarajan51,Reference Kraus, Powell and Haskell52) . The physical activity guidelines in the USA show that as individuals increase their levels of physical activity, there is a corresponding decrease in the relative risk of all-cause mortality(Reference Piercy, Troiano and Ballard49). In this study, the association between shorter eating duration and increased risk of all-cause mortality was only found in individuals engaged in light physical activity. This finding could potentially offer additional evidence supporting the significant and positive impact of regular physical activity in reducing the risk of mortality. However, further studies are required to elucidate the underlying mechanisms.
The main results of our study were verified by sensitivity analyses. However, it is worth noting that after excluding early-morning eaters, a clear U-shaped relationship between eating duration and all-cause mortality was observed. This finding further supports our previous notion that there may be an optimal eating duration that may significantly reduce the risk of all-cause mortality. Interestingly, a majority of the participants in our study had one or more metabolic diseases (diabetes, hyperlipidaemia, hypertension and obesity). Upon conducting a detailed analysis of these individuals, we found that the dose–response relationship between eating duration and all-cause mortality followed a similar pattern to that observed in the integrated analytical population. Several studies have indicated that metabolic disorders could profoundly influence dietary habits and preferences, necessitating significant changes in eating patterns to effectively manage their health(Reference Maștaleru, Cojocariu and Oancea53,Reference Tirfessa, Abebe and Darega54) . For example, individuals with diabetes will adjust their mealtimes and restrict carbohydrate intake to control postprandial blood glucose levels, resulting in natural changes in their eating duration. Therefore, it is still unclear whether the observed association is due to the changed dietary regime induced by metabolic diseases or whether the changes in eating duration fundamentally contribute to the development of these diseases and subsequently influence all-cause mortality. Thus, more comprehensive investigations and analyses are needed to better understand the relationship between metabolic diseases, eating duration and the risk of all-cause mortality.
This study has several strengths. First, a nationally representative sample of US adults with long-term follow-up data was used in our study, which may facilitate the generalisability of our findings to the broader population of US adults. Second, the employment of a standardised and comprehensive laboratory and physical examination protocol by NHANES successfully minimised measurement errors. Third, a wide range of potential confounders in the multivariable analyses were taken into account in the study, which enhanced the reliability and validity of the results. However, the following limitations should be fully considered when interpreting the results of the study. First, eating duration was calculated based on a single recall over the past 24 h, which could not fully represent the participant’s habitual dietary behaviour, and possible effects of random measurement errors were likely present. Second, despite our efforts to control major confounders, we cannot entirely dismiss the possibility of residual confounding from unknown or unmeasured variables in the correlation between eating duration and all-cause mortality. Lastly, it is crucial to recognise that the observational nature of our study limits our ability to establish causality, highlighting the need for future prospective large cohort studies to provide more robust evidence regarding the association between eating duration and all-cause mortality.
Conclusion
A negative association was identified between the risk of all-cause mortality and eating duration, and a non-linear dose–response relationship between eating duration and all-cause mortality was observed exclusively among participants with high energy intake or low levels of physical activity. These findings offer valuable insights for managing daily dietary habits, especially in terms of eating duration. However, further research is necessary to confirm and explore these results more comprehensively.
Acknowledgements
This work was supported by grants from the Natural Science Foundation of Fujian Province (no. 2022J01237) and the Fujian Medical University’s Research Foundation for Talented Scholars (no. XRCZX2017002).
F. G. and M. L. conceptualised the study question and designed the research; M. L. and J. H. analysed the data; S. D., J. C. and K. S. provided technical support; F. G. and M. L. wrote the paper. All authors contributed to the interpretation of the results, reviewed and approved the final manuscript.
The authors declare no conflict of interest.
Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S0007114524001739.
