The Association between Overnight Fasting and Body Mass Index in Older Adults: The Interaction between Duration and Timing

Abstract

Background

Circadian rhythms play an important role in the regulation of eating and fasting, and mistimed dietary intakes may be detrimental to metabolic health. Extended overnight fasting has been proposed as a strategy to better align the eating-fasting cycle with the internal circadian clock, and both observational and experimental studies have linked longer overnight fasting with lower body weight. However, it remains unclear if the timing of overnight fasting modifies the relationship between fasting duration and weight outcomes.

Methods

The current study included 495 men and 499 women age 50–74 years. Dietary intake over 12 months was assessed by 24-hour dietary recalls every two months, and body-mass index was measured at the beginning, middle and end of the study. Logistic regression was used to estimate the relationship between overnight fasting duration and the likelihood of being overweight or obesity adjusted for multiple confounders, and assessed whether the relationship was modified by the timing of overnight fasting, measured as the midpoint of the fasting period.

Results

Among participants with early overnight fasting (midpoint < 02:19 am), a longer fasting duration was associated with lower odds of overweight and obesity; while among those with late fasting (≥ 02:19 am), longer fasting was associated with higher odds of overweight and obesity. Specifically, when compared to the shortest quintile of overnight fasting duration, the longest quintile was associated with a 53% reduction in the odds of overweight and obesity in the early fasting group (OR=0.47, 95% CI=0.23, 0.97), but a 2.36-fold increase in the late fasting group (OR=3.36, 95% CI=1.48, 7.62). Additionally adjusting for dietary intakes during morning and late evening periods did not affect the observed associations.

Conclusions

Longer overnight fasting was associated with a reduced likelihood of being overweight or obese, but only among those with an early timing of fasting.

Introduction

Eating and fasting diurnal behaviors are regulated by the internal circadian timing system. Growing evidence demonstrates that circadian organization of the eating-fasting cycle, which is influenced by numerous environmental and individual factors, associates with cardiometabolic health.¹ In particular, recent research has linked prolonged overnight fasting with indicators of favorable metabolic health, including a lower body weight and/or fat mass.^2–13 For example, a long overnight fasting period (>18 hours) has been associated with a relative decrease in body-mass index (BMI).² However, a study among women in the 2009–2010 U.S. National Health and Nutrition Examination Survey (NHANES) showed a significant bivariate relationship between longer overnight fasting and higher BMI,³ and a similar finding of a correlation between a shorter diet intake window (i.e. longer fasting) and higher waist circumference was reported in the Cancer Prevention Study-3 Diet Assessment sub-study.¹⁴ Moreover, multiple pilot intervention studies examined the effects of time-restricted feeding, otherwise known as time-restricted eating, or restricting caloric intake to a consistent window of <12 hours per day, on weight loss. Restricting eating duration to <12 hours over several weeks led to modest weight or fat loss.^4–13

The physiological responses to eating and fasting are also regulated by the circadian clock, and vary within a 24-hour period: Eating breakfast shortly after waking lowers odds of overweight and obesity and reduces risk for weight gain,^15–17 whereas eating at night associates with higher BMI and obesity.^17–19 Moreover, weight loss via bariatric surgery or caloric restriction is diminished if caloric distribution is higher in the evening.^{20, 21} These findings suggest that the relationship between overnight fasting duration and BMI may differ by the timing of fasting, as an early window of overnight fasting characterized by breakfast consumption and early evening meals may have different metabolic consequences when compared with a late fasting window characterized by skipping breakfast and nighttime snacking. Indeed, a recent randomized trial compared the effects of early (8 AM to 5 PM) vs. late (12 PM to 9 PM) feeding protocol for one week and reported that an early, but not late, feeding schedule led to lower fasting glucose.²² However, due to the short length of intervention (one week), meaningful differences in weight outcomes were not observed.

In the current study, we examined the associations of overnight fasting duration and timing with BMI status among older American men and women participating in the Interactive Diet and Activity Tracking in AARP (IDATA) study. We hypothesized that longer overnight fasting would be associated with lower likelihood of being overweight and obesity, and the association would be stronger among those with an earlier fasting timing. We also hypothesized that the association between overnight fasting duration and BMI status could be partially mediated by dietary intakes in the morning and at night.

Methods

Study Population

We conducted analysis among participants of the IDATA (ClinicalTrials.gov ID: NCT03268577; study details are available on the study website (https://biometry.nci.nih.gov/cdas/idata/). Briefly, IDATA was designed to evaluate the validity of internet-based and conventional self-report for assessing diet and physical activity among older adults aged 50–74 years. Eligibility criteria included being able to speak and read English, BMI ≥ 18.5 and < 40 kg/m², having access to high-speed internet, being mobile, and reporting no history of diabetes, renal failure, congestive heart failure, or conditions involving disturbances in fluid balance or digestion. People who were on a weight loss diet were excluded from the study. Between January 2012 and December 2013, a total of 1,082 people who were eligible and consented to the study, and of these, 994 had self-reported dietary intake data and objectively measured BMI data available and were included in the current analysis. The study was approved by the National Cancer Institute Special Studies Institutional Review Board.

Assessment of Physical Activity and sleep and Chronotype Calculation

Participants wore the activPAL^™ physical activity monitor for seven consecutive days at two time-points six months apart.²³ The monitor was attached to the right thigh and the participants were asked to wear the device continuously and only remove it when showering, bathing or swimming. For calculating physical activity variables, we excluded monitored days with < 10 hours of monitor wear time according to conventional practice reported before.²⁴ We calculated the average total steps per day to measure physical activity levels and the average minutes per day spent in a sitting position to measure sedentary time.

Participants were asked to complete a physical activity log for two 7-day periods, six months apart, when they wore accelerometer devices to record daily physical activity. In the activity log, they reported the times when they got out of bed and went to bed each day, from which we calculated average total time in bed and the midpoint of time in bed for weekdays and the weekend, separately. We also calculated chronotype (HH:MM), or the propensity to engage in diurnal behaviors such as sleep and physical activity at certain times of the day, based on the Munich Chronotype Questionnaire algorithm:¹²

$$\text{chronotype}={\text{M}}_{\text{weekend}}{\text{ if TIB}}_{\text{weekend}}>{\text{TIB}}_{\text{weekday}},$$

and

$$\text{chronotype}={\text{M}}_{\text{weekend}}-\left({\text{TIB}}_{\text{weekend}}-{\text{TIB}}_{\text{weekday}}\right)/2{\text{ if TIB}}_{\text{weekend}}\ge {\text{TIB}}_{\text{weekday}},$$

where M_weekend is the midpoint of time in bed on weekends and TIB_weekend and TIB_weekday represent total time in bed on weekends and weekdays, respectively. HH:MM represents 24-hour clock time.

Dietary Recall and Assessment of Overnight Fasting Duration and Midpoint

During the one-year study period, participants completed six 24-hour dietary recalls, once every two months, using the Automated Self-Administered 24-Hour Dietary Assessment Tool (ASA24).²⁵ The ASA24 is an online platform that asks participants to self-report all foods, beverages and dietary supplements consumed in the previous 24-hours from midnight to midnight. Participants reported detailed information for each eating episode, including the specific foods and beverages consumed, portion size for each food and beverage, name of the eating episode (e.g. breakfast, lunch, dinner, snack) and time of eating episode in 15-minute blocks. In the current analytic cohort, 94% participants had at least three recalls and 67% had all six recalls.

The duration of overnight fasting has been previously calculated based on the clock time of the first and last meal within the midnight-to-midnight recall period.^{3, 26} However, this approach is flawed for individuals who eat before sleep after midnight, because the first late night meal after midnight will be mistaken as the first meal of the day. Because earlier research showed that overnight fasting window often parallels time in bed,⁴ we therefore defined the first and last meal as the first eating episode after and the last eating episode before the midpoint of time in bed, respectively. Since sleep and eating schedules vary between weekdays and weekends, we calculated the timing of the first and last meal according to midpoint of time in bed separately for weekdays and weekend days. Because sleep measurement from physical activity logs and dietary recalls were performed on different days throughout the study period, we used the yearly average of weekday and weekend sleep midpoints for each participant to calculate their overnight fasting duration and midpoint for each recall day and derived the mean and standard deviation for the entire study period.

Meal Timing and Caloric Intake Distribution Variables

Because both overnight fasting and eating patterns during waking hours are regulated by the circadian rhythm, we calculated multiple variables related to meal timing and daily distribution of caloric intakes to describe diurnal eating patterns, control for confounding and examine potential mediation by daily energy consumption. Meal timing variables included the time of the first and last meals, both based on the clock time and relative to the sleep cycle, as measured by the time between midpoint of time in bed and the time of first and last meals. We also calculated the difference between midpoint of overnight fasting and chronotype, which serves as a proxy measure of the internal circadian time. On average, the midpoint of overnight fasting was earlier than the circadian time (i.e., chronotype, HH:MM) in this population, and a larger value indicated a more advanced or earlier timing of the overnight fasting window relative to the circadian time.

To calculate caloric intake distributions, we estimated total daily energy intake using the USDA Food and Nutrient Database for Dietary Studies (version 4.1) and calculated the percent of daily energy intake for breakfast, lunch and dinner. Because our earlier study found that dietary intakes within two hours after waking and before bedtime had a stronger association with BMI than intakes for conventional meals (breakfast, lunch and dinner) and intakes based on clock time,¹⁷ we calculated the percent of daily caloric intake in these two time windows as previously described and examined their role as potential mediators of the association between overnight fasting duration and BMI. For the two-hour time window after waking, 79% of eating episodes were reported as breakfast; and 78% of the eating episodes that occurred within two hours before bedtime were reported as snack.

Body mass index

Height and weight were measured at three clinic visits in the 1^st, 7^th and 12^th month of the study. Body mass index (weight (kg)/height(m)²) was calculated using average weight and height over the three measures and dichotomized into normal weight (18.5≤BMI<25 kg/m²) and overweight and obese (25≤BMI<40 kg/m²).²⁷

Statistical Analyses

Descriptive statistics are shown as means and standard deviations for continuous variables, and frequencies and percentages for categorical variables. We used multivariable logistic regression to estimate odds ratios (OR) and 95% confidence intervals (CI) for the association between duration of overnight fasting and odds of being overweight or obese. We compared quintiles of overnight fasting duration with the lowest (shortest quintile) as the reference group. For all our regression analyses, we considered a series of models: the base model controlled for sex (men, women) and age (continuous); the second model, which was considered as the main model in the current work, additionally controlled for variables that are likely confounders, including race/ethnicity (black, white, other), total time in bed (< 7, 7–8 and 9+ hours), chronotype (quintiles), total steps per day (quintiles), total sitting time per day (quintiles), and total daily caloric intake (continuous); the third model additionally controlled for meal timing variables that may mediate the relationship between overnight fasting and overweight and obesity, including the percent of daily energy intakes within two hours after wake up and the percent of daily energy intakes within two hours before bedtime, both as continuous variables. Linear trend was assessed by assigning the median value to each quintile and modeling as a continuous exposure. We also assessed p-values for interactions using the likelihood ratio test comparing a model with a cross-product term between overnight fasting duration and potential modifiers of its association with overweight and obesity to one without. To test the main hypothesis that the timing of overnight fasting modifies the relationship between overnight fasting duration and overweight and obesity, we conducted stratified analysis by the midpoint of overnight fasting using a median split (02:19 am). Finally, because we found no evidence supporting an interaction between overnight fasting duration and sex (p value=0.79), we combined men and women in the current analyses.

Code availability

All programs used in this study for data analysis can be obtained by communicating with the corresponding author.

Results

The average duration of overnight fasting in the current study sample was 11.4 ± 1.7 hours. Table 1 presents study characteristics according to quintiles of overnight fasting duration, stratified by early (< median 02:19 am) or late (≥ 02:19 am) midpoint for overnight fasting. Study characteristics for the overall study sample can be found in Supplementary Table 1. For both early and late fasting groups, those who reported longer overnight fasting also reported longer time in bed on both weekdays and weekends and a lower total energy intake. In addition, they also had less sedentary time, a later midpoint of time in bed on both weekdays and weekends, and a later chronotype among the early fasting group, but these patterns were not observed among the late fasting group. Sex differences were apparent in duration of overnight fasting time, with a higher proportion of women having longer fasting times.

Table 1.

Study characteristics according to quintiles^a of duration of overnight fasting among people with early and late midpoint of overnight fasting^b.

	Duration of overnight fasting, quintiles
	Q1 (shortest)	Q2	Q3	Q4	Q5 (longest)
Overnight fasting duration, hour, median (IQR)	9.5 (9.0, 9.9)	10.6 (10.4, 10.8)	11.3 (11.2, 11.6)	12.1 (12.0, 12.3)	13.4 (13.0, 13.9)	p-value c
Early overnight fasting (midpoint < 02:19 am)
N	92	81	97	108	119
Female, %	32.6	49.4	48.5	57.4	52.1	0.01
White, %	96.7	96.3	94.9	92.6	91.6	0.35
Age, year, mean (SD)	61.8 (5.4)	63.0 (6.1)	61.8 (6.5)	62.8 (6.2)	64.5 (5.3)	0.004
Steps, count/day, mean (SD)	7774.7 (3157.8)	7683.1 (2914.1)	7820.4 (2998.5)	7268.7 (3097.8)	7417.6 (3079.8)	0.51
Sedentary time, min, mean (SD)	523.1 (149.1)	509.8 (122.5)	494.6 (132.9)	474.0 (119.4)	477.4 (107.3)	0.05
Time in bed, weekend, hour, mean (SD)	8.1 (1.2)	8.5 (1.1)	8.6 (1.0)	8.6 (1.2)	8.9 (1.1)	<.0001
Time in bed, weekday, hour, mean (SD)	7.3 (1.1)	7.9 (0.9)	7.9 (0.9)	9.1 (0.9)	8.4 (1.0)	<.0001
Midpoint of time in bed, weekend, HH:MM, mean (SD)	02:42 (00:50)	02:54 (00:46)	02:57 (00:47)	03:04 (00:47)	03:11 (00:50)	0.002
Midpoint of time in bed, weekday, HH:MM, mean (SD)	02:08 (00:41)	02:28 (00:43)	02:36 (00:40)	02:37 (00:46)	02:52 (00:45)	<.0001
Chronotype, e HH:MM, mean (SD)	02:14 (00:46)	02:32 (00:42)	02:34 (00:44)	02:43 (00:52)	02:51 (00:56)	<.0001
Calorie, kcal/day, mean (SD)	2348.1 (771.7)	2116.4 (641.3)	2131.6 (524.4)	1965.2 (564.1)	1795.2 (572.0)	<.0001
Late overnight fasting (midpoint ≥ 02:19 am)
N	109	116	99	93	80
Female, %	45.9	46.6	51.5	65.6	52.5	0.04
White, %	92.7	94.0	87.9	83.9	85.0	0.18
Age, year, mean (SD)	63.1 (6.7)	64.3 (5.7)	62.4 (5.7)	62.8 (5.6)	63.0 (5.8)	0.12
Steps, count/day, mean (SD)	7093.8 (3154.6)	6682.1 (2896.0)	6835.9 (2854.4)	6368.6 (2561.0)	6198.1 (2412.8)	0.38
Sedentary time, min, mean (SD)	493.4 (131.1)	475.2 (114.9)	488.9 (128.1)	486.9 (129.7)	489.9 (122.5)	0.76
Time in bed, weekend, hour, mean (SD)	8.3 (1.2)	8.5 (1.2)	8.5 (1.2)	8.8 (1.4)	9.0 (1.4)	0.0004
Time in bed, weekday, hour, mean (SD)	7.6 (1.2)	7.9 (0.8)	8.2 (1.4)	8.4 (1.2)	8.7 (1.2)	<.0001
Midpoint of time in bed, weekend, HH:MM, mean (SD)	03:55 (00:59)	03:56 (00:50)	03:47 (00:49)	03:59 (00:53)	04:02 (01:01)	0.53
Midpoint of time in bed, weekday, HH:MM, mean (SD)	03:30 (00:53)	03:36 (00:54)	03:37 (00:54)	03:47 (00:55)	03:40 (01:00)	0.18
Chronotype, e HH:MM, mean (SD)	03:29 (01:01)	03:32 (00:52)	03:30 (00:53)	03:39 (01:02)	03:44 (01:01)	0.37
Calorie, kcal/day, mean (SD)	2398.3 (752.8)	2279.5 (573.8)	2245.3 (578.0)	2030.5 (547.0)	2008.8 (532.1)	<.0001

^aQuintiles are based on the overall population.

^bEarly vs. late midpoint for overnight fasting is determined by the median split (02:19 am).

^cp-values for categorical variables were derived from Chi-square test and p-values for continuous variables was derived from Kruskal–Wallis test.

^dChronotype is measured as the midpoint of time in bed on weekends (M_weekend), if total time in bed on weekends (TIB_weekend) ≤ total time in bed on weekdays (TIB_weekday); measured as the M_weekend-(TIB_weekend - TIB_weekday)/2, if TIB_weekend > TIB_weekday.

Abbreviation: AARP, formerly American Association of Retired Persons; SD, standard deviation; TEI, total energy intake.

Table 2 presents characteristics of diurnal eating patterns (i.e. variables of meal timing and caloric intake distributions) according to both the duration and timing of overnight fasting. Among both early and late fasting groups, a longer fasting duration was associated with an earlier last meal and a later first meal. Among the early fasting group, participants with longer overnight fasting also showed a significantly earlier, or more advanced, circadian time of fasting midpoint relative to the chronotype (p=<.0001). In contrast, among the late fasting group, there was no relationship between overnight fasting duration and the circadian timing of fasting midpoint (p=0.32). For caloric intake distributions, longer overnight fasting was associated with larger intakes at dinner and smaller intakes within 2 hours before bedtime among both the early and late fasting group. On the other hand, longer fasting was associated with larger breakfasts only among the early fasting group, and there was no clear dose-dependent relationship between fasting duration and breakfast intake among the late fasting group.

Table 2.

Meal timing and caloric intake distributions according to quintiles^a of duration of overnight fasting among people with early and late midpoint of overnight fasting^b.

	Duration of overnight fasting, quintiles
	Q1 (shortest)	Q2	Q3	Q4	Q5 (longest)
Overnight fasting duration, hour, median (IQR)	9.5 (9.0, 9.9)	10.6 (10.4, 10.8)	11.3 (11.2, 11.6)	12.1 (12.0, 12.3)	13.4 (13.0, 13.9)	p-value c
Early overnight fasting (midpoint < 02:19 am)
Midpoint of overnight fasting, HH:MM, mean (SD)	01:22 (00:49)	01:33 (00:34)	01:40 (00:27)	01:31 (00:31)	01:21 (00:41)	0.01
Time of first meal, HH:MM, mean (SD)	05:53 (01:14)	06:50 (00:36)	07:20 (00:27)	07:36 (00:32)	08:13 (00:44)	<.0001
Time between midpoint of time in bed and first meal, hour, mean (SD)	4.0 (0.7)	4.3 (0.6)	4.7 (0.6)	4.8 (6.5)	5.2 (0.8)	<.0001
Time of last meal, HH:MM, mean (SD)	20:50 (00:54)	20:14 (00:36)	20:00 (00:29)	19:25 (00:36)	18:31 (01:05)	<.0001
Time between midpoint of time in bed and last meal, hour, mean (SD)	5.4 (0.6)	6.3 (0.6)	6.7 (6.8)	7.3 (0.7)	8.4 (1.1)	<.0001
Time between midpoint of overnight fasting and chronotype,d hour, mean (SD)	0.7 (0.7)	1.0 (0.6)	0.9 (0.7)	1.2 (0.9)	1.5 (0.9)	<.0001
% TEI from breakfast, mean (SD)	16.4% (6.3%)	17.9% (6.6%)	17.8% (5.6%)	19.4% (7.8%)	21.9% (10.6%)	<.0001
% TEI from lunch, mean (SD)	20.9% (8.1%)	23.9% (9.1%)	23.8% (8.2%)	24.4% (9.6%)	23.0% (10.9%)	0.07
% TEI from dinner, mean (SD)	36.0% (9.5%)	36.6% (9.7%)	37.6% (9.8%)	41.6% (9.9%)	41.1% (11.7%)	0.0003
% TEI from within 2 hours after wake up, mean (SD)	13.0% (6.9%)	15.0% (7.4%)	15.3% (6.6%)	16.2% (8.6%)	18.1% (9.4%)	0.001
% TEI from within 2 hours before bedtime, mean (SD)	9.4% (7.5%)	6.3% (6.3%)	5.7% (9.5%)	4.5% (7.9%)	2.3% (3.7%)	<.0001
Late overnight fasting (midpoint ≥ 02:19 am)
Midpoint of overnight fasting, HH:MM, mean (SD)	03:11 (00:45)	03:07 (00:35)	03:03 (00:43)	03:06 (00:40)	03:13 (00:53)	0.35
Time of first meal, HH:MM, mean (SD)	07:52 (00:46)	08:25 (00:35)	08:43 (00:44)	09:10 (00:41)	10:06 (01:17)	<.0001
Time between midpoint of time in bed and first meal, hour, mean (SD)	4.3 (0.6)	4.8 (0.6)	5.0 (0.7)	5.3 (0.9)	6.3 (1.5)	<.0001
Time of last meal, HH:MM, mean (SD)	22:29 (00:48)	21:48 (00:38)	21:22 (00:43)	21:02 (00:41)	20:20 (00:58)	<.0001
Time between midpoint of time in bed and last meal, hour, mean (SD)	5.1 (0.8)	5.9 (0.6)	6.3 (0.8)	6.8 (0.8)	7.5 (1.1)	<.0001
Time between midpoint of overnight fasting and chronotype,d hour, mean (SD)	0.3 (0.9)	0.4 (0.7)	0.4 (0.7)	0.6 (1.0)	0.5 (1.2)	0.32
% TEI from breakfast, mean (SD)	16.5% (6.2%)	16.8% (7.3%)	16.5% (6.8%)	19.1% (8.9%)	16.0% (8.6%)	0.26
% TEI from lunch, mean (SD)	21.2% (8.1%)	20.7% (8.4%)	20.2% (8.9%)	20.4% (9.5%)	18.4% (9.6%)	0.20
% TEI from dinner, mean (SD)	35.6% (9.0%)	38.7% (10.1%)	38.2% (8.6%)	37.7% (10.5%)	44.7% (11.5%)	<.0001
% TEI from within 2 hours after wake up, mean (SD)	13.8% (6.1%)	13.8% (6.8%)	13.1% (7.4%)	15.1% (8.5%)	11.0% (8.8%)	0.02
% TEI from within 2 hours before bedtime, mean (SD)	8.2% (6.0%)	7.1% (6.1%)	6.3% (6.2%)	5.7% (6.4%)	5.0% (6.7%)	<.0001

^aQuintiles are based on the overall population.

^bEarly vs. late midpoint for overnight fasting is determined by the median split (02:19 am).

^cp-values for categorical variables were derived from Chi-square test and p-values for continuous variables was derived from Kruskal–Wallis test.

^dChronotype is measured as the midpoint of time in bed on weekends (M_weekend), if total time in bed on weekends (TIB_weekend) ≤ total time in bed on weekdays (TIB_weekday); measured as the M_weekend-(TIB_weekend - TIB_weekday)/2, if TIB_weekend > TIB_weekday.

Abbreviation: AARP, formerly American Association of Retired Persons; SD, standard deviation; TEI, total energy intake.

Table 3 and Figure 1 present the association between the duration of overnight fasting and overweight and obesity in the overall population and by the midpoint of overnight fasting. We found that this association differed drastically by the midpoint of overnight fasting (p-interaction = 0.003): a longer overnight fasting duration was associated with lower odds of overweight and obesity among those with an early midpoint of overnight fasting, but higher odds among those with a later midpoint. Specifically, when compared to the shortest quintile, the longest quintile of overnight fasting was associated with a 53% reduction in the odds of overweight and obesity in the early fasting group (OR (95% CI), 0.47 (0.23, 0.97)), but a 2.36-fold increase in odds in the late fasting group (3.36 (1.48, 7.62)). This interaction attenuated the associations in the overall analysis, which showed no relationship between overnight fasting duration and overweight and obesity. Additional adjustment of the percent of energy intake within 2 hours after waking and before bedtime had only a small impact on the results and slightly strengthened the association among the early fasting group (Model 3, Table 3). Adjusting for caloric intakes at breakfast and dinner, and the timing of first and last meals also did not meaningfully alter the results (data not shown). Finally, midpoint of overnight fasting itself was not significantly associated with overweight and obesity in the study (Supplementary Table 2).

Table 3.

Associations between the duration of overnight fasting and the odds of being overweight or obese in the Interactive Diet and Activity Tracking in AARP Study.

	Duration of overnight fasting, quintiles						per-quintile change
	Q1 (shortest)	Q2	Q3	Q4	Q5 (longest)	p-trend
Early overnight fasting (midpoint < 02:19 am)
No. of obese and overweight subjects (%)	73 (79.4)	59 (72.8)	62 (63.9)	71 (65.7)	76 (63.9)
OR (95% CI), Model 1	ref	0.79 (0.39, 1.62)	0.50 (0.26, 0.97)	0.59 (0.31, 1.13)	0.55 (0.29, 1.05)	0.06	0.87 (0.76, 1.00)
OR (95% CI), Model 2	ref	0.75 (0.35, 1.59)	0.50 (0.24, 1.01)	0.48 (0.24, 1.00)	0.47 (0.23, 0.97)	0.02	0.83 (0.71, 0.98)
OR (95% CI), Model 3	ref	0.67 (0.29, 1.57)	0.41 (0.18, 0.89)	0.42 (0.18, 0.96)	0.34 (0.14, 0.81)	0.01	0.78 (0.64, 0.94)
Late overnight fasting (midpoint ≥ 02:19 am)
No. of obese and overweight subjects (%)	74 (67.9)	93 (80.2)	80 (80.8)	72 (77.4)	66 (82.5)
OR (95% CI), Model 1	ref	1.97 (1.06, 3.67)	2.16 (1.12, 4.15)	1.97 (1.03, 3.78)	2.44 (1.19, 5.00)	0.02	1.22 (1.04, 1.43)
OR (95% CI), Model 2	ref	2.07 (1.03, 4.15)	2.16 (1.05, 4.44)	2.35 (1.11, 4.98)	3.36 (1.48, 7.62)	0.005	1.30 (1.08, 1.56)
OR (95% CI), Model 3	ref	1.90 (0.90, 4.05)	1.69 (0.77, 3.68)	2.07 (0.93, 4.61)	3.47 (1.36, 8.84)	0.02	1.28 (1.05, 1.57)
Overall
No. of obese and overweight subjects (%)	147 (73.1)	152 (77.2)	142 (72.5)	143 (71.1)	142 (71.4)
OR (95% CI), Model 1	ref	1.35 (0.85, 2.14)	1.03 (0.66, 1.62)	1.06 (0.68, 1.65)	1.02 (0.65, 1.60)	0.70	0.98 (0.89, 1.09)
OR (95% CI), Model 2	ref	1.29 (0.79, 2.10)	1.02 (0.63, 1.64)	0.99 (0.61, 1.60)	1.03 (0.63, 1.68)	0.70	0.97 (0.88, 1.09)
OR (95% CI), Model 3	ref	1.22 (0.71, 2.09)	0.87 (0.51, 1.48)	0.94 (0.55, 1.62)	0.88 (0.49, 1.55)	0.41	0.95 (0.83, 1.08)

Model 1: adjusted for sex and age.

Model 2: adjusted for variables in Model 1 and race/ethnicity, total time in bed, chronotype, total steps per day, duration of sedentary time and total daily energy intake.

Model 3: adjusted for variables in Model 2 and midpoint of overnight fasting, % of TEI consumed within 2 hours after wake up and % of TEI consumed within 2 hours before bedtime.

P-for-interaction between midpoint and duration of overnight fasting was 0.003.

Figure 1.

Associations between the duration of overnight fasting and the odds of being overweight or obese in the overall study population and as divided according to the midpoint of overnight fasting. An earlier midpoint was defined as earlier than the median (2:19 am), while a later midpoint was defined as later than the median. The model was adjusted for age, sex, race/ethnicity, total time in bed, chronotype, total steps per day, duration of sedentary time, total daily energy intake, midpoint of overnight fasting, % of TEI consumed within 2 hours after wake up and % of TEI consumed within 2 hours before bedtime. P-for-interaction between midpoint and duration of overnight fasting was 0.003. Abbreviations: CI, confidence interval; OR, odds ratio.

Discussion

Among older men and women in the US, we found that, consistent with our hypothesis, longer overnight fasting was associated with a reduced likelihood of being overweight or obese, but only among those with an early timing of overnight fasting. In contrast, among those with late overnight fasting timing, there was a positive association between the duration of overnight fasting and overweight and obesity. However, our findings did not support a role of dietary intakes shortly after waking and before bedtime as mediators of these associations.

Few observational studies have examined the relationship between overnight fasting duration and weight outcomes among a free-living population. In an analysis of over 50,000 adults in the Adventist Health Study 2,² Kahleova et al. found that a long overnight fast (≥18 hours) was associated with a lower relative gain in BMI per year (β (95% CI), −0.02 (−0.03, −0.004)) when compared with the reference group (12–17 hours), while a short fast (<12 hours) was associated a higher relative BMI gain (β (95% CI),0.02 (0.01, 0.03)). Like ours, this study also controlled for a wide range of potential confounders including sleep and physical activity. In contrast, in an earlier study by Marinac et al.,³ people in the longest tertile of overnight fasting (≥13.5 hours) appeared to have the highest BMI (29.1) compared with the lower tertiles (27.5 and 27.9 for the first and second tertile, respectively, p value=0.01), although the study did not present the multivariable-adjusted association between fasting duration and BMI. On the other hand, the study found that longer overnight fasting was associated with lower HbA1c and 2-hour glucose levels (measured by Oral Glucose Tolerance Test), indicating a relationship between longer fasting and better metabolic outcome. Neither of these studies examined whether the relationship between overnight fasting duration and weight and metabolic outcomes differed by the timing of the fasting window. Other human observational studies have been conducted among select populations that modify their fasting schedules for religious or work-related reasons. For example, a recent meta-analysis of 85 studies reported a significant weight reduction during Ramadan (−1.022 kg, 95% CI (−1.164 kg, −0.880 kg), and the effects appeared more pronounced with longer fasting duration.²⁸ Moreover, studies also reported beneficial changes such as reduced waist circumference and improved cardiometabolic biomarkers during Ramadan.^{29, 30} It is worth noting that the fasting pattern of Ramadan fasting is different from most of the intermittent fasting schemes: During Ramadan, Muslims are required to fast from sunrise to sunset, and have two main meals in early evening and pre-dawn. It is unclear whether and how the distinct diurnal pattern of fasting and eating in Ramadan contributes to the observed effects on weight and metabolic health.

There has been a growing interest in assessing time-restricted feeding as an intervention strategy to improve metabolic health, reviewed in detail previously by Patterson and Sears (2017).³¹ Several studies in animal models suggested that restricting feeding to defined periods of time over 24-hours may lead to lower weight gain on a high fat diet.^32–37 For example, a study by Hatori et al. randomized mice into four groups with different combinations of high-fat vs. normal diet and 8-hour restricted vs. ad lib feeding.³² Despite equivalent caloric intakes, mice on a restricted feeding schedule were protected against developing obesity induced by a high-fat diet. Similar effects were also observed in multiple studies in human subjects.^4–13 For example, in a pilot study, Gill and Panda recruited 8 individuals with BMI>25kg/m² and eating duration >14 hours/day and asked them to reduce eating duration to a self-selected window of 10–12 hours. After 16 weeks, subjects lost an average 3.27 kg.⁴ In another study among 23 obese subjects, a 12-week restricted feeding intervention (8-hour eating window, 10AM to 6PM) led to an average of 2.6% decrease in body weight.⁵ In a more recent study, ten overweight, sedentary adults aged 65 or older were instructed to choose a 16-hour fasting schedule that fit their lifestyle for a duration of 4-weeks. Mean weight loss post-intervention was 2.6 kg.⁶ Despite the small sample sizes, a common observation among existing intervention studies is modest reduction in total daily caloric intake, which could partially explain the effects of restricted feeding on weight loss. Indeed, in our study, we also observed that participants with longer overnight fasting had lower energy intake; however, the relationship between fasting duration and overweight and obesity remained after adjusting for total energy intake, suggesting that other mechanisms may play a role.

One mechanism by which time-restricted feeding, or extended overnight fasting, may lead to weight loss and improved metabolic health could be that it promotes a better alignment between the eating-fasting cycle and the circadian physiology of metabolism.³⁸ It has been suggested that disrupting the synchronization between eating behaviors and circadian physiology may lead to adverse metabolic consequences. For example, feeding nocturnal rodents during the light phase, which typically correlates with rest and inactivity, leads to a higher weight gain when compared with feeding during the dark phase.³⁹ In humans, shift work has been consistently associated with higher weight gain and impaired metabolic health and mistimed eating has been proposed as a potential mechanism that may explain these observed relationships.⁴⁰ While nightshift work represents a severe form of circadian misalignment, short overnight fasting is often a result of late night or early morning snacks, which may represent a subtler form of desynchronization between the eating-fasting cycle and internal circadian clock. Indeed, earlier studies have found that breakfast skipping^{15, 41} and high energy intake late in the evening^{18, 42–44} are associated with higher adiposity. For example, an earlier analysis in the IDATA participants found that a higher energy intake shortly after waking up and a lower energy intake before bedtime were associated with lower odds of overweight and obesity.¹⁷ Taken together, findings from experimental and observational studies support a role of mistimed eating in obesity.

There is a well-documented bidirectional relationship between metabolism and the circadian timing system.^{45, 46} On one hand, the circadian timing system, which consists of integrated central and peripheral molecular clocks, orchestrate multiple aspects of energy homeostasis and nutrient metabolism, influences hormones levels, and shapes the 24-hour cycle of food anticipation and eating behaviors. On the other hand, the actual eating behaviors and subsequent nutrient availability to numerous metabolic pathways can reset circadian gene expression and have wide-reaching effects on both the peripheral (e.g., muscle, liver, adipose tissue) and central clocks.^{45, 47} At a molecular level, the core clock genes that form the transcription-translation feedback loops in the circadian system regulate rhythmic expression of numerous enzymes involved in nutrient metabolism (e.g. AMPK, OGT, SIRTs); while subsequently these enzymes also modify the stability of core clock proteins and influence circadian function.⁴⁸ It has been reported that Ramadan fasting may have an impact on genes that are involved in metabolism and circadian clock regulation (SIRTs), supporting a role of fasting in influencing the link between circadian function and metabolism.⁴⁹

In our current study, a potentially important finding was that the relationship between overnight fasting and overweight and obesity depended on the timing of fasting. When comparing the early fasting group with the late group, we found multiple differences not only in their average dietary patterns, but also in how dietary patterns change as overnight fasting extends. For example, among the early fasting group, people with longer overnight fasting also had a larger gap between the midpoint of overnight fasting relative to chronotype). In other words, these people not only had a longer fasting period, but their fasting period occurred at an earlier phase of their internal circadian clock, which may impact nutrient responses independent of the effects of fasting duration. In contrast, such an association between the duration and circadian timing of overnight fasting was not observed among the later fasting group. These differences in diurnal dietary patterns (i.e., choice of when to start eating after waking or stop eating before bedtime) in relation to overnight fasting duration among early vs. late fasting groups may potentially explain the different patterns between fasting duration and overweight and obesity in these two groups. However, contrary to what we expected, dietary intakes within two hours after waking up and before bedtime did not appear to explain the observed relationships between fasting duration and overweight and obesity in either the early or late fasting group, nor did the intakes in these morning and evening time windows appear to account for the differential results between these two groups. One possibility could be that there are differences in internal circadian clocks between those with an early and late fasting timing that cannot be captured by measuring dietary intakes within these time windows. For example, the same 2-hour period after waking up or before bedtime may represent different circadian phases for individuals with different internal clocks (i.e., chronotype) and thus different preferences of sleep and eating schedules.⁵⁰ To further examine this possibility, future research should use more accurate measurements of circadian rhythms, such as the dim light melatonin onset and 24-hour core body temperature, to better characterize diurnal dietary patterns relative to the internal circadian timing.

Our study has several strengths. First, dietary intakes were measured by 24-hour recalls every two months for one year, and most subjects had at least three measurements, which enabled us to capture habitual dietary patterns and minimized the impact of seasonal fluctuations in dietary intakes. In addition, we derived a large number of variables related to various aspects of dietary intakes to characterize dietary patterns that may confound and/or mediate the associations between fasting and overweight and obesity. Also, we had detailed measures of sleep and physical activity that permitted robust control for additional lifestyle factors as potential confounders.

Our study also has some limitations. As mentioned above, circadian timing may play an important role in determining the effects of overnight fasting on weight status; however, unfortunately, we did not have measures of circadian oscillation biomarkers, such as melatonin, or hormones that may play an important role in the circadian regulation of metabolism and eating behaviors, such as leptin, ghrelin, and cortisol. In addition, we did not have information on other metabolic markers, such as lipids, glucose, insulin and adipokines, that would allow us to conduct a more comprehensive assessment of metabolic health. Moreover, although we controlled for sleep duration and timing in our analysis, we did not have information on sleep quality, which plays a distinct role in metabolic health and may impact our findings.^51–53 Also, dietary information was self-reported and thus subject to report error and recall bias. Finally, the current analytic sample included older predominantly white men and women with high education levels, and our results cannot be generalized to younger, or minority racial/ethnic groups and people with lower socioeconomic status that may have distinct dietary patterns when compared with affluent white populations. For example, breakfast skipping appeared to be more common among people with less-than-college education but less common in black populations.⁵⁴ Therefore, it would be important for future studies to include participants from more diverse backgrounds to examine how overnight fasting relates to BMI status in different subpopulations.

In conclusion, we found that the relationship between overnight fasting duration and BMI status was not universal, but instead determined by the timing of the overnight fasting window. Specifically, early and longer overnight fasting duration associated with lower odds of overweight and obesity. Our findings highlight the importance of considering individual differences in circadian rhythms when designing strategies that aim to improve weight outcomes and metabolic health via dietary interventions.