Longitudinal timing of physical activity and associated cardiometabolic and behavioral health outcomes in young adults

Abstract

Background

This is the first study to examine longitudinal associations between self-selected timing of moderate-to-vigorous physical activity (MVPA) and health outcomes in young adults over 18 months.

Methods

Young adults (N = 434, M_age = 23.9, SD_age = 4.6 years) enrolled in a weight management trial recorded 4-7 days of ActiGraph wear time at ≥1 time point (baseline, months 6, 12, and 18). Time-of-day categories were based on quartiles of the temporal distribution of MVPA min/h at baseline: morning (06:00-11:59), afternoon (12:00-15:59), evening (16:00-18:59), and night (19:00-00:59). The proportion of weekly MVPA accumulated during each time category was the predictor in longitudinal linear mixed-effects models predicting body mass index (BMI) and total weekly MVPA. Longitudinal quasibinomial generalized estimating equations models predicted cardiometabolic risk. Interactions were tested, and marginal trend estimates were generated for sex and age subgroups.

Results

The analytic sample was 79% female and 49% non-Hispanic White, with a mean (±SD) weekly MVPA of 311 ± 167 min at baseline. In adjusted models, there were no associations with BMI. Morning MVPA was inversely associated with cardiometabolic risk (OR [95% CI]: 0.99 [0.98-0.99]) for both sex and age groups. Evening MVPA was inversely associated with cardiometabolic risk for 26-35 year olds (0.98 [0.97-0.99]). Morning MVPA was associated with greater total MVPA across subgroups, and afternoon MVPA was associated with less total MVPA in women.

Conclusions

Over 18 months, incremental health benefits may accrue with optimal activity timing in young adults. Activity-based interventions designed to improve cardiometabolic and behavioral health outcomes in young adults may be optimized by tailoring timing recommendations to demographic factors.

Morning activity was associated with lower cardiometabolic risk and more total weekly physical activity in young adults with overweight/obesity

Introduction

To prevent and manage chronic disease, national guidelines call for a minimum of 150 min of moderate-to-vigorous intensity physical activity (MVPA) per week.¹ Yet, little is known about what time of day might be best for performing this activity.² To date, studies in adult populations suggest that the timing of physical activity may be associated with weight,³ cardiometabolic risk,⁴ and total MVPA.⁵ However, few longitudinal studies on this topic exist, and several systematic reviews have reported that, overall, the evidence remains inconclusive due to methodological limitations, such as cross-sectional designs and small samples.^6–8 Furthermore, few studies on this topic focus on young adult populations, who are biologically and behaviorally different from older adult counterparts.^9–11 Of note, “young adulthood” generally refers to ages 18-35 years in the literature (and in this article), although experts further categorize this group into 2 distinct developmental periods: 18-25 years (typically referred to as “emerging adulthood”) and 26-35 years.¹²

The timing of physical activity may be particularly important for mitigating health risks in young adults who are vulnerable to social jetlag (ie, discrepancy between biological rhythms and sleep/wake behaviors), placing them at increased risk for chronic diseases.^9,10,13,14 Research indicates that morning activity produces phase-advance shifts (ie, earlier sleep-wake cycles) measured by dim light melatonin onset,^15,16 probably driven partially by morning light exposure, suggesting that morning activity may be protective against circadian misalignment and related cardiometabolic risk in young adults. Thus, properly timed physical activity could be one behavioral strategy incorporated into lifestyle interventions tailored to young adults for chronic disease prevention and management.

Despite this potential, physical activity timing and health outcomes are particularly understudied in young adult populations. In a secondary analysis of the Midwest Exercise Trial 2, young adults (N = 88, 18-35 years) with overweight/obesity who chose to complete endurance exercise in the morning (07:00-11:59) lost significantly more body weight (2.1%) compared with those who chose late exercise (15:00-19:00) over 10 months of follow-up.¹⁷ While these participants self-selected their activity time, they did so from predetermined time windows. Conversely, in a randomized cross-over trial of young men (N = 22, 19-25 years), it was reported that a single evening exercise session (starting at 18:00) was associated with greater fat oxidation and decreased body fat mass, regardless of obesity status, compared with the same exercise in the morning (starting at 08:00).¹⁸ Lastly, 1 study reported that, in a sample of young men (N = 12, M_age: 21.8 ± 0.2 years), late afternoon endurance exercise (16:00-18:00) accumulated in a laboratory setting over the course of 3 nonconsecutive weekdays was more effective at regulating 24-hour glucose and triglyceride levels compared with morning exercise (09:00-11:00).¹⁹ However, given the established sex differences in circadian rhythms and metabolism,²⁰ small studies excluding female participants lack generalizability, particularly in the context of cardiometabolic disease, and representative studies of young adult women are needed to further explore questions of activity timing and health outcomes.

The above studies administered exercise in supervised laboratory environments with predetermined exercise time windows, precluding participants from self-selecting their activity timing. Thus, it is still unknown when community-dwelling (ie, “free-living”) young adults choose to exercise throughout the 24-hour day. To date, only 1 study has reported data regarding young adults’ self-selected exercise time. Bailey et al. demonstrated, in a small, cross-sectional university cohort (N = 31, M_age: 23 at baseline), that the most popular self-reported time to exercise over the previous 7-day period was in the evening (17:00-03:59; n = 11) followed by varied times (n = 10).²¹ Self-reported total weekly MVPA was highest among self-reported evening exercisers (265 ± 123 min) compared with daytime (04:00-16:59; 144 ± 78 min) or varied times (109 ± 94 min, P = .009). Additionally, body mass index (BMI) was lowest among self-reported evening exercisers (28.6 ± 4.0 kg/m²), compared with daytime (32.8 ± 4.0 kg/m²) or varied times (35.2 ± 11.8 kg/m², P = .091).²¹ This study was not powered to explore differences by sex.

Given current knowledge gaps, ecologically valid longitudinal evidence regarding when young adults (including 18-25 and 26-35 years, men and women) choose to be active, along with potential physical and behavioral health correlates, is needed to inform future randomized trials. Furthermore, studies that consider all the clinically relevant physical activity accrued within the 24-hour day (ie, MVPA), as opposed to solely assessing planned and structured “exercise” sessions, may demonstrate stronger correlations with long-term health outcomes.¹

Finally, there are both behavioral and biological mechanisms that might theoretically link the timing of physical activity to improved weight and cardiometabolic health in young adults. First, the timing of physical activity may be associated with total volume of MVPA,^5,21–23 and it is well established that increases in MVPA lead to improvements in weight and cardiometabolic health.¹ Second, optimally timed physical activity may lead to improvements in insulin-related circadian processes, such as insulin-dependent glucose uptake by muscle cells,²⁴ which could lead to improved weight and cardiometabolic health. Furthermore, insulin sensitivity peaks in the morning, which, coupled with regular physical activity, could improve fat substrate utilization and metabolic outcomes. This is particularly salient for young adults experiencing social jetlag, a phenomenon known to disrupt metabolic control.²⁵ However, to the best of the authors' knowledge, no longitudinal studies have examined MVPA or insulin as mediators of the relationship between physical activity timing and health outcomes.

Thus, knowledge gaps remain regarding when young adults self-select to be active, whether self-selected activity timing is associated with health outcomes, and whether MVPA or insulin mediates the proposed relationship. This exploratory secondary data analysis aimed to address these gaps using a longitudinal observational design in a cohort of young adults enrolled in a randomized weight management trial. We hypothesized that morning activity would be associated with lower BMI and lower cardiometabolic clustering score (CCS; cardiometabolic outcomes) and greater weekly MVPA (behavioral outcome) across time points. Further exploratory analyses were conducted to test whether proposed relationships were (1) moderated by sex or age group and (2) mediated by total weekly MVPA or insulin. Finally, we described changes in young adults’ self-selected activity timing across measurement time points in the overall sample and at the group level, as this has not been previously described.

Methods

Participants

The Healthy Body Healthy U (HBHU) trial was a randomized controlled weight management intervention designed for young adults with overweight/obesity.^26,27 At baseline, the trial enrolled 459 participants between the ages of 18 and 35 years. The trial recruited participants from 2015 to 2018. See ref.²⁶ for details regarding eligibility and recruitment. In brief, eligibility criteria were 18-35 years old; BMI of 25-45 kg/m²; enrolled in a university in the greater DC or Boston areas; Facebook user; English fluency; and access to text messaging. Exclusion criteria were major medical conditions and participation in other weight-related programs.

HBHU was designed to test the effectiveness of a Tailored (personalized) vs. Targeted (generic) weight management messaging program delivered via Facebook and text messaging, compared with a contact Control group that received non–weight-related wellness messaging via the same channels. Briefly, the Tailored intervention consisted of personalized messages based on participants’ self-reported weight loss barriers, while the Targeted and Control interventions consisted of generic weight-related and generic health/wellness-related messages, respectively. The intervention spanned 18 months, including 4 data collection time points (baseline and months 6, 12, and 18). Procedures were approved by the Institutional Review Boards of both study sites.

Measures

Body mass index

BMI (in kg/m²) was calculated from average height and weight measurements at all 4 time points. Height was measured using a standard portable stadiometer to the nearest 0.1 cm, and weight was measured via digital scale (Seca Model 769) to the nearest 0.2 kg. Height and weight measurements were recorded in duplicate and averaged.

Cardiometabolic risk factors

A CCS (range 0-5), previously developed for use in young adults,²⁸ was created based on whether a standard clinical value was exceeded (1 = exceeded, 0 = not exceeded) for each of the 5 risk factors: abdominal circumference (men: >102 cm, women: >88 cm), systolic blood pressure (≥130 mmHg), diastolic blood pressure (≥85 mmHg), high-density lipoprotein-cholesterol (HDL-C; men: <40 mg/dL, women: <50 mg/dL), and hemoglobin A1c (HbA1c; ≥5.7%). The 5 cardiometabolic risk factors were assessed by trained research staff. Abdominal circumference and blood pressure were measured in triplicate at each time point. Abdominal circumference was measured at the navel with a cloth tape measure, and blood pressure was taken using a digital blood pressure monitor (OMRON HEM-907XL). HDL-C and HbA1c were obtained from blood samples after an 8-hour overnight fast at baseline, month 6, and month 18. The CCS outcome score was generated for baseline, month 6, and month 18 due to the absence of venous blood sampling at month 12.

Insulin, a proposed mediator in this secondary analysis was also collected by trained research staff via blood samples after an 8-hour overnight fast at baseline, month 6, and month 18.

Total weekly MVPA

Participants were instructed to wear an ActiGraph accelerometer for 7 days during waking hours at baseline (prior to randomization) and at months 6, 12, and 18. Valid wear data was defined as 10 or more hours per day of recording on 4 or more days.²⁹ Non-wear was defined as a period longer than 1 hour with consecutive zeros in vector magnitude, allowing for a spike tolerance of up to 2 minutes to account for artificial movements. Total time spent in MVPA was calculated using Freedson cutpoints (ie, ≥1952 counts per minute³⁰) in ActiLife software. Daily average minutes of MVPA on valid wear days multiplied by 7 were used to represent the total weekly MVPA at each time point.³¹ Supplementary analyses of bouted MVPA (ie, MVPA in bouts of ≥ 10 minutes) are available in Tables S1 and S2.

Activity time-of-day

The temporal distribution of baseline accelerometry data was used to derive 4 activity time-of-day categories, morning, afternoon, evening, and night. This data-driven method has previously been used to describe activity timing among adults with type 2 diabetes and provides a rigorous alternative to selecting arbitrary time windows for analysis.⁴ Briefly, the 24-hour temporal distribution of MVPA minutes in the baseline sample was generated by summing the average MVPA minutes per clock hour per participant. Clock hours were grouped into quartiles so that each time-of-day category included similar amounts of MVPA after dropping the 5 hours with the fewest counts (~1% of the data) to account for non-wear time.⁴ Quartile-based time categories were morning (06:00-11:59), afternoon (12:00-15:59), evening (16:00-18:59), and night (19:00-00:59). For consistency, these categories were used across measurement time points. The proportion (in %) of weekly MVPA occurring in each time category was calculated for each participant by

$${\displaystyle \begin{array}{l}{\displaystyle \frac{\mathrm{a}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{d}\mathrm{a}\mathrm{i}\mathrm{l}\mathrm{y}\mathrm{M}\mathrm{V}\mathrm{P}\mathrm{A}\mathrm{m}\mathrm{i}\mathrm{n}\mathrm{u}\mathrm{t}\mathrm{e}\mathrm{s}\times 7}{\mathrm{t}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{w}\mathrm{e}\mathrm{e}\mathrm{k}\mathrm{l}\mathrm{y}\mathrm{M}\mathrm{V}\mathrm{P}\mathrm{A}}\times 100}\end{array}}$$

This was conducted for each time point.

Moderators and covariables

Additional variables of interest were measured via survey at baseline. These variables included the proposed moderators of sex and age category (18-25 years vs. 26-35 years) as well as covariables race/ethnicity and housing status. Race/ethnicity categories were non-Hispanic White, African American/Black, Hispanic, Asian/Pacific Islander, or multi-racial/other. Housing status categories were apartment, dorm, house with classmates, house with family, or other. Additionally, a time-varying covariate, average daily kilocalories consumed, was added to models predicting BMI. Average daily kilocalorie intake was calculated via ASA24 software from up to 3 recorded 24-hour recalls at each time point.³² Study group (Tailored, Targeted, Control) was inlcuded as a covariable in all models.

Statistical analysis

Descriptive statistics

Baseline descriptive statistics were generated using means and standard deviations for continuous variables and proportions for categorical variables. Change from baseline statistics (mean [SD]) in weekly MVPA (%) per time category for the full sample and by sex, age group, and study group was generated. All analyses were performed in R (version 2023.03.1).³³ An alpha level of <0.05 was determined a priori.

Body mass index

To model the relationship between BMI and the proportion of weekly MVPA in each time category over the 4 measurement time points, longitudinal linear mixed effects models with random intercepts were fitted. Models were run separately for each time-varying predictor (ie, proportion of weekly MVPA occurring in the morning, afternoon, evening, and night) to avoid potential multicollinearity. Models adjusted for time-invariant covariates age category, sex, race/ethnicity, housing status, and study group, as well as total weekly MVPA (time-varying) and daily kilocalorie intake (time-varying). Likelihood ratio tests were used to identify significant interaction terms between each time-of-day predictor with sex, age category, study group, and measurement time point, including 3-way interactions. Estimation proceeded via maximum likelihood using the R package “lme4.”³⁴ Likelihood ratio tests and visual assessment of autocorrelation confirmed that the data satisfied compound symmetry assumptions.

Cardiometabolic risk

To model the relationship between cardiometabolic disease risk and proportion of weekly MVPA in each time category over the 3 study time points with available CCS scores (baseline, month 6, and month 18), quasibinomial logistic regression models using generalized estimating equations (GEE) with robust sandwich estimation were fitted using the “gee” package.³⁵ As described above, models were run separately for each time-varying predictor to avoid potential multicollinearity. Models adjusted for all time-invariant covariates listed above and total weekly MVPA (time-varying). Quasibinomial model estimates were exponentiated to produce odds ratios, comparing the odds of participants having an elevated risk factor for cardiometabolic disease with respect to covariate patterns.

Total weekly MVPA

To model the relationship between total weekly MVPA and proportion of weekly MVPA in each time category, longitudinal linear mixed effects models with random intercepts were fitted as described for BMI. Models adjusted for all time-invariant covariates listed above.

Marginal trends

Marginal trends representing the effect of a 1 percentage-point increase in MVPA per time category on each health outcome were generated for sex and age subgroups. Marginal trends were estimated using the “emmeans” package.³⁶

Mediation

Mediation analyses were conducted for significant relationships between MVPA timing and health outcomes using the “mediation” package.³⁷ In brief, this consisted of fitting models to assess the total effect of the predictor on the outcome variable, the effect of the predictor on the mediator variable, and the effect of the mediator on the outcome variable.³⁸ The “mediation” package produces estimates and 95% confidence intervals for the average causal mediation effects, the average direct effects, the total effects, and the proportion of the effect that is mediated. Inference was conducted using bootstrapping with 1000 resamples. Mediation models were adjusted for the same covariates and interaction terms described above, including all time-invariant as well as applicable time-varying covariates for each outcome.

Sensitivity analysis

There were missing values in variables related to MVPA (see Table S3), as some participants did not wear the device per protocol. Maximum likelihood estimation is robust to partial missingness under the missing at random (MAR) assumption when the marginal covariance structure is approximately correctly specified,³⁹ and GEEs are unbiased when data are missing completely at random.⁴⁰ Sensitivity analyses were conducted to assess robustness to missing data assumptions (ie, to test the assumption that missing wear days were similar in MVPA to non-missing wear days). The results of those analyses can be found in Tables S4–S7.

Results

Sample characteristics

The analytic sample consisted of 434 participants with valid accelerometry measurements from at least one time point. Demographic characteristics (eg, age, sex, race/ethnicity) did not differ between those with and without valid accelerometry data. A subset of data (n = 388) were available for analysis in the cardiometabolic risk models, due to missing data on one or more cardiometabolic risk factors across time points. Individuals with missing cardiometabolic data did not differ from those with full cardiometabolic data in terms of age, sex, or race/ethnicity. At baseline, participants in the full analytic sample were 23.9 ± 4.6 years of age, 79% female, 49% non-Hispanic White, and 45% reported living in an apartment (non-campus housing). The average baseline BMI was 31.2 ± 4.4 kg/m². See Table 1 for additional cardiometabolic characteristics across study time points.

Table 1.

Characteristics of the analytic sample.

	Baseline	Month 6	Month 12	Month 18
	(n = 434)	(n = 306)	(n = 257)	(n = 203)
Study group (%)
Control	141 (32.49)	–	–	–
Tailored	145 (33.41)	–	–	–
Targeted	148 (34.10)	–	–	–
Age (years)	23.86 (4.57)	–	–	–
Age category (%)
18-25 years	306 (70.51)	–	–	–
26-35 years	128 (29.49)	–	–	–
Female (%)	342 (78.80)	–	–	–
Race/Ethnicity (%)
Non-Hispanic White	214 (49.31)	–	–	–
African American/Black	86 (19.82)	–	–	–
Asian/Pacific Islander	39 (8.99)	–	–	–
Hispanic	58 (13.36)	–	–	–
Multi-racial/other	37 (8.53)	–	–	–
Housing status (%)
Apartment	194 (44.70)	–	–	–
Dormitory	84 (19.35)	–	–	–
House	14 (3.23)	–	–	–
House with family	122 (28.11)	–	–	–
Other	20 (4.61)	–	–	–
Weekly MVPA (minutes)	311.45 (166.64)	314.07 (163.67)	300.73 (151.76)	295.72 (173.59)
Morning MVPA (% of weekly total)	28.38 (15.31)	23.09 (15.31)	20.13 (14.91)	30.99 (14.24)
Afternoon MVPA (% of weekly total)	28.55 (11.83)	22.96 (13.37)	20.73 (13.10)	27.06 (13.61)
Evening MVPA (% of weekly total)	22.86 (11.12)	17.64 (11.37)	16.68 (11.64)	22.36 (11.23)
Night MVPA (% of weekly total)	20.42 (13.13)	15.34 (12.15)	12.76 (9.97)	20.48 (13.96)
BMI (kg/m2)	31.18 (4.41)	30.82 (4.49)	30.82 (4.78)	30.94 (4.78)
Weight (kg)	86.46 (15.59)	85.81 (15.22)	85.36 (16.20)	86.00 (16.06)
Abdominal circumference (cm)	99.57 (11.58)	98.23 (11.54)	97.75 (12.06)	97.38 (12.36)
Systolic blood pressure (mmHg)	114.23 (11.31)	113.06 (10.36)	115.41 (10.79)	113.87 (11.48)
Diastolic blood pressure (mmHg)	72.63 (8.70)	73.77 (8.66)	73.75 (8.03)	73.74 (8.26)
HDL-C (mg/dL)	48.63 (11.09)	49.15 (11.02)	–	50.41 (11.54)
HbA1c (%)	5.30 (0.45)	5.26 (0.44)	–	5.10 (0.57)
Insulin (mIU/mL)	11.88 (7.64)	12.03 (7.93)	–	12.20 (7.88)
CCS (0-5)	1.65 (0.93)	1.48 (0.99)	–	1.29 (1.02)

Abbreviations: BMI, body mass index; CCS, cardiometabolic clustering score; HbA1c, hemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; MVPA, moderate-to-vigorous physical activity. See Table S3 for missing data per variable.

The total weekly MVPA at baseline was 311.5 ± 166.6 minutes. The average proportion of weekly MVPA engaged in during the morning (06:00-11:59), afternoon (12:00-15:59), evening (16:00-18:59), and night (19:00-00:59) hours at baseline were 28% ± 15%, 29% ± 12%, 23% ± 11%, and 20% ± 13%, respectively (see Table 1 for proportions at months 6, 12, and 18). Across all time categories, MVPA appeared to decline slightly at months 6 and 12, and increase roughly back to baseline values at month 18, although these changes were not statistically significant (Figure 1). The average change from baseline in weekly MVPA per time category at months 6, 12, and 18 did not differ by sex, age group, or study group, demonstrating consistent weekly activity patterns between measurement timepoints at the group level (Table 2).

Figure 1.

Average change from baseline in weekly MVPA (%) accrued in each time category across study time points. Across all time categories, MVPA declined at months 6 and 12, and increased roughly back to baseline values at month 18. Changes from baseline were not statistically significant at any time point.

Table 2.

Change from baseline (mean [SD]) in weekly MVPA (%) per time category by sex, age group, and study group.

		Full sample	Sex		P-value		Age group		P-value	Study group			P-value
			Men (n = 92)	Women (n = 342)			18-25 years (n = 306)	16-35 years (n = 128)		Tailored (n =  141)	Targeted (n = 145)	Control (n = 148)
Month 6	Morning	−3.38 (19.71)	−3.84 (19.83)	−3.25 (19.72)	0.842	−3.85 (19.31)		−2.32 (20.68)	0.575	−3.02 (19.93)	−1.74 (21.48)	−5.30 (17.68)	0.486
	Afternoon	−5.78 (17.64)	−7.48 (16.60)	−5.21 (17.95)	0.391	−5.47 (16.97)		−6.34 (19.16)	0.719	−5.61 (16.71)	−5.63 (19.33)	−5.94 (16.89)	0.991
	Evening	−5.58 (15.09)	−5.21 (15.04)	−5.69 (15.14)	0.833	−5.43 (13.73)		−5.91 (17.90)	0.820	−5.72 (14.62)	−7.48 (14.79)	−3.61 (15.69)	0.241
	Night	−7.68 (15.38)	−5.18 (14.76)	−5.23 (15.60)	0.982	−5.29 (16.33)		−5.06 (13.07)	0.914	−5.33 (14.66)	−6.61 (16.90)	−3.78 (14.46)	0.481
Month 12	Morning	−7.68 (17.68)	−9.20 (16.94)	−7.34 (17.88)	0.538	−8.12 (16.06)		−6.83 (20.57)	0.602	−7.37 (16.96)	−7.71 (17.67)	−7.91 (18.47)	0.983
	Afternoon	−9.26 (17.00)	−8.01 (16.96)	−9.54 (17.04)	0.599	−9.08 (17.13)		−9.59 (16.84)	0.831	−8.01 (17.90)	−9.13 (18.16)	−10.39 (15.17)	0.695
	Evening	−5.70 (15.80)	−6.00 (13.40)	−5.63 (16.32)	0.893	−6.37 (14.35)		−4.39 (18.32)	0.372	−5.16 (13.46)	−5.38 (16.31)	−6.42 (17.20)	0.870
	Night	−7.39 (14.88)	−7.09 (16.94)	−7.45 (14.43)	0.885	−6.83 (16.26)		−8.47 (11.78)	0.432	−8.59 (15.44)	−5.31 (15.30)	−8.30 (13.99)	0.329
Month 18	Morning	4.47 (18.79)	3.59 (19.06)	4.73 (18.77)	0.726	5.20 (18.92)		2.93 (18.58)	0.437	3.67 (21.94)	4.81 (19.17)	4.86 (15.56)	0.925
	Afternoon	−2.27 (17.85)	−1.53 (17.30)	−2.49 (18.06)	0.758	−2.27 (16.36)		−2.27 (20.83)	0.998	−4.11 (18.93)	−1.29 (16.48)	−1.56 (18.45)	0.630
	Evening	−0.42 (15.10)	−1.12 (11.37)	−0.22 (16.04)	0.733	0.05 (14.53)		−1.43 (16.32)	0.528	1.03 (14.28)	−1.50 (15.74)	−0.71 (15.32)	0.645
	Night	−0.83 (17.99)	−0.24 (18.31)	−1.00 (17.96)	0.808	−1.24 (18.30)		0.05 (17.43)	0.645	−0.16 (17.63)	−1.22 (17.69)	−1.05 (18.78)	0.940

P-values generated using ANOVA tests.

BMI and MVPA timing

The percentage of weekly MVPA accumulated during morning, afternoon, evening, and night hours, respectively, was not associated with BMI across study time points, adjusting for total weekly MVPA, kilocalories, and other covariables (Tables 3 and S8–S10). See Table S2 for estimates based on bouted MVPA.

Table 3.

Marginal trends for the relationship between time-based MVPA and health outcomes for each sex and age group. Estimates indicate the change in outcome (column) for a one percentage-point increase in weekly MVPA per time category (quadrant) and demographic group (row).

	BMI	CCS	Total weekly MVPA	BMI	CCS	Total weekly MVPA
	β (95% CI)	OR (95% CI)	β (95% CI)	β (95% CI)	OR (95% CI)	β (95% CI)
Morning MVPA				Afternoon MVPA
Female	0.004 (−0.002, 0.01)	0.99 (0.98, 0.99)	M0: 0.91 (0.10, 1.72) M6: 0.58 (0.02, 1.14) M12: 0.25 (−0.39, 0.89) M18: −0.08 (−1.05, 0.89)	0.005 (−0.003, 0.01)	0.99 (0.99, 1.01)	M0: −1.69 (−2.69, −0.69) M6: −0.91 (−1.56, −0.25) M12: −0.12 (−0.84, 0.60) M18: 0.67 (−0.46, 1.80)
Male			M0: −0.21 (−2.00, 1.58) M6: 1.08 (−0.09, 2.26) M12: 2.37 (1.05, 3.69) M18: 3.66 (1.59, 5.73)	−0.01 (−0.03, 0.001)		M0: 1.03 (−0.90, 2.96) M6: 0.44 (−0.82, 1.69) M12: −0.16 (−1.52, 1.21) M18: −0.75 (−2.89, 1.39)
18-25 years			0.93 (0.31, 1.56)	−0.004 (−0.01, 0.004)		−0.21 (−0.90, 0.46)
26-35 years
Evening MVPA				Night MVPA
Female	0.0003 (−0.01, 0.01)	0.99 (0.99, 1.00)	0.02 (−0.62, 0.66)	0.004 (−0.004, 0.01)	1.00 (0.99, 1.01)	0.27 (−0.34, 0.88)
Male
18-25 years		1.01 (0.99, 1.01)
26-35 years		0.98 (0.97, 0.99)

Table quadrants represent marginal trends for a one percentage-point increase in MVPA in the morning (top left), afternoon (top right), evening (bottom left), and night (bottom right), respectively, for each demographic subgroup (row). Estimates were generated from the final selected models (see Tables S10–S12). Unique subgroup estimates are only available where significant interactions were identified; otherwise, estimates were based on all included study participants. Significant estimates and 95% confidence intervals are presented in bold. Linear mixed effects model estimates are shown for BMI and total weekly MVPA outcomes. Odds ratios are shown for CCS outcomes.

Abbreviations: BMI, body mass index; CCS, cardiometabolic clustering score; CI, confidence interval; MVPA, moderate-to-vigorous physical activity; OR, odds ratio

Cardiometabolic risk and MVPA timing

The percentage of weekly MVPA accumulated during morning and evening, respectively, was associated with cardiometabolic risk across time points, adjusting for total weekly MVPA and other covariables (Tables 3 and S11). Specifically, morning MVPA was inversely associated with cardiometabolic risk (OR for a 1 percentage-point increase in morning MVPA: 0.99 [0.98, 0.99]) for both sex and age groups. Evening MVPA was inversely associated with cardiometabolic risk in 26-35 year olds only (OR for a 1 percentage-point increase in evening MVPA: 0.98 [0.97, 0.99]; interaction P = .017). Afternoon and night MVPA were not associated with cardiometabolic risk. See Tables S8 and S9 for estimates based on MVPA increases of 20 percentage-points and 50 percentage-points, respectively. See Table S2 for estimates based on bouted MVPA.

Total weekly MVPA and MVPA timing

The percentage of weekly MVPA accumulated during morning and afternoon hours, but not evening or night, was associated with total weekly MVPA across study time points in adjusted models (Tables 3 and S12). Time-varying morning MVPA was generally associated with more total weekly MVPA. Specifically, morning MVPA was associated with more total weekly MVPA for women at baseline (beta estimates for a 1 percentage-point increase in morning MVPA: 0.91 [0.10, 1.72] min) and month 6 (0.58 [0.02, 1.14] min), men at months 12 (2.37 [1.05, 3.69] min) and 18 (3.66 [1.59, 5.73] min), and both age groups (0.93 [0.31, 1.56] min). Afternoon MVPA was associated with less total weekly MVPA for women at baseline (beta estimates for a 1 percentage-point increase in afternoon MVPA: −1.69 [−2.69, −0.69] min) and month 6 (−0.91 [−1.56, −0.25] min). See Table S12 for interaction P-values, including 3-way interactions with month of measurement. See Tables S8 and S9 for estimates based on MVPA increases of 20 percentage-points and 50 percentage-points, respectively. See Table S2 for estimates based on bouted MVPA.

Mediation

Total weekly MVPA and insulin did not mediate identified relationships between MVPA in morning and evening hours, respectively, and cardiometabolic risk (Table 4).

Table 4.

Total weekly MVPA and insulin as mediators of longitudinal relationships between activity timing and cardiometabolic clustering score (CCS).

Relationship (IV → DV)	ACME	ADE	Total effect	Proportion mediated
Mediator: Total weekly MVPA
Morning MVPA % → CCS	0 (0, 0)	0.76 (−0.01, 1.53)	0.76 (−0.01, 1.53)	0 (0, 0)
Evening MVPA‡ % → CCS	0 (0, 0)	0.42 (−0.48, 1.37)	0.42 (−0.52, 1.48)	0 (0, 0)
Mediator: Insulin †
Morning MVPA % → CCS	0 (0, 0)	−0.02 (−0.07, 0.03)	−0.02 (−0.07, 0.03)	0 (0, 0)
Evening MVPA‡ % → CCS	0 (0, 0)	−0.04 (−0.10, 0.02)	−0.04 (−0.10, 0.02)	0 (0, 0)

Total effect = ADE + ACME, proportion mediated = ACME/total effect; estimates <0.001 are rounded to 0; models adjusted for age, sex, race/ethnicity, month, study group, housing status, and total weekly MVPA; insulin adjusted for in insulin-mediated models only; ^†N = 231, ^‡models adjusted for age interaction as described in Table S11 (Model 3).

Abbreviations: ACME, average causal mediation effects; ADE, average direct effects; DV, dependent variable;

IV, independent variable; MVPA, moderate-to-vigorous physical activity.

Discussion

This study is the first longitudinal observational study of young adult activity timing and cardiometabolic outcomes. Over the course of 18 months, we found that morning MVPA was associated with a lower probability of being at elevated cardiometabolic risk for all participants. While this association was small, modest increases in morning MVPA could have a clinically relevant impact. For example, a 20 percentage-point increase in MVPA accumulated during the morning (roughly 9 additional minutes per day for the average participant) was associated with 12% lower odds (OR: 0.88; 95% CI: 0.80-0.97) of cardiometabolic risk (Table S8). A more substantial increase in morning MVPA of 50 percentage-points (an additional 22-min per day for the average participant) was associated with 26% lower odds (OR: 0.74; 95% CI: 0.58-0.93) of cardiometabolic risk (Table S9). Importantly, this finding did not hold across our sensitivity analyses exploring departures from the MAR assumption. So, while morning activity appears to be inversely associated with cardiometabolic risk assuming activity data are comparable on missing and non-missing wear days, this may not be the case if participants selectively did not wear the ActiGraph on low activity days. The finding also did not hold for our bouted MVPA analyses, suggesting that the health benefits of morning activity are not dependent on performing longer, sustained bouts; operationalizing MVPA in bouts may overlook some of the beneficial, cumulative effects of shorter activities.

Other studies have also reported that morning activity may yield cardiometabolic health benefits. A longitudinal study of UK Biobank participants (62 ± 8 years, 58% female) found that over 6 years of follow-up, participants who tended to be active in the late morning (activity peak between 09:00 and 10:00) had a lower risk of incident coronary artery disease and stroke compared with those with a later activity pattern.⁴¹ However, evidence remains mixed, as 2 studies in older adults (mean age: 62 and 58 years, respectively) suggest morning physical activity may actually increase the risk of cardiovascular disease-related mortality,^42,43 which could be explained by several biological phenomena occurring in the morning, including vascular function impairments, higher platelet activity, and elevated risk of cardiac arrythmia.^44–46 It is possible that the relationship between morning activity and risk of cardiovascular disease events increases with age; future studies could explore this hypothesis in samples that include both young adults and older adults.

This study demonstrated evening MVPA was inversely associated with cardiometabolic risk for 26-35 year olds, but not 18-25 year olds. Specifically, a 20 percentage-point increase in evening MVPA was associated with 19% lower odds (OR: 0.81; 95% CI: 0.66-0.99) of cardiometabolic risk, and a 50 percentage-point increase in evening MVPA was associated with 41% lower odds (OR: 0.59; 95% CI: 0.36-0.98) of cardiometabolic risk (Tables S8 and S9). This finding did not hold in our bouted MVPA analyses, which may indicate that cardiometabolic benefits of time-based MVPA are not contingent on performing sustained bouts. Previous research in adults has reported some benefits associated with afternoon and/or evening activity, such as improvements in glycemic control⁴⁷ and reductions in insulin resistance.⁴⁸ Our findings regarding differential association for evening activity by age group may be related to circadian phenotypes, such as chronotype, which were not measured in the parent trial and thus unable to be adjusted for in the present analyses. For example, a study of UK Biobank participants (58 ± 8 years) reported that cardiovascular disease mortality hazard was higher for individuals with activity times that were mismatched with their chronotype over 7 years of follow-up.⁴³

This study did not find any association between activity timing and BMI. Previous studies have also reported no relationship between activity timing and weight metrics, including both observational⁴⁹ and randomized trial designs.^50,51 In contrast, other studies (observational,⁵² quasi-experimental,¹⁷ and randomized controlled trials^53,54) have supported a link between morning activity and reduced weight in both adults and young adults, indicating more research on this topic is needed. One possible explanation for our results is that unobserved variables, such as chronotype, may have reduced the effect such that a benefit of morning activity was less pronounced (or even reversed) among participants with later chronotypes. Conversely, our findings may be due to limitations regarding the selection of time categories based on participant activity data, a behavioral proxy for circadian biology.

Comparable methods and data harmonization across studies are needed.⁸ A current barrier to harmonizing findings across studies is the lack of agreement regarding the best way to operationalize and analyze the timing of physical activity. The present study used a data-driven approach, as has been done previously.^4,47 The authors felt that a data-driven approach was more rigorous, compared with a priori time category selection, given that few studies have previously described physical activity timing in young adults. However, even a data-driven approach may introduce bias into the analysis by selecting cutoffs based on activity behavior rather than circadian biology. In order for the activity timing field to further mature, standardized approaches to analyzing physical activity timing will be needed.

In terms of behavioral health, morning MVPA was generally associated with more total weekly MVPA across subgroups, while afternoon MVPA was associated with less total weekly MVPA for women at baseline and month 6. For example, among men, our findings are roughly equivalent to a 47- and 73-minute increase in total weekly MVPA for an additional 20% of weekly MVPA accrued in the morning at months 12 and 18, respectively (Table S8). Among women, our findings are roughly equivalent to an 18- and 12-minute increase in total weekly MVPA for an additional 20% of weekly MVPA accrued in the morning at baseline and month 6, respectively, and a 34- and 18-minute decrease in total weekly MVPA for an additional 20% of weekly MVPA accrued in the afternoon at baseline and month 6, respectively (Table S8). Other studies have also reported links between morning activity and greater total activity. For example, one pre-posttest study found that, after 6-weeks of physical activity counseling, preoperative bariatric surgery candidates (46 ± years, 88% female) who performed their longest bout of activity in the morning (before 12:00 noon) were more likely to achieve public health guidelines (≥150 min of MVPA/week).²³ To the best of our knowledge, no other studies have reported an inverse relationship between afternoon activity and total MVPA. These relationships generally held across our bouted MVPA analyses.

Few studies have tested the relationship between accelerometer-measured activity timing and total weekly MVPA,^5,23 and none have previously done so in a young adult sample, making our findings a unique contribution to the literature. With regards to mechanisms of action, researchers have posited that consistent morning MVPA may lead to increased total weekly MVPA and related health outcomes via both biological and behavioral mechanisms. For example, consistent morning physical activity may lead to weight loss via alterations in circadian-based processes, such as reductions in energy intake, improvements in circadian-based physiological processes (eg, fat oxidation), and improvements in sleep.²² However, we found no evidence of mediation by total weekly MVPA (behavioral mediator) or insulin (biological mediator) in the relationship between timing of activity and cardiometabolic risk. It is possible that the data collection time points for our study (baseline, and months 6, 12, and 18) were temporally too sparse to capture the mediation effects that we were interested in, or otherwise temporally delimited (eg, insulin samples taken at 1 time within the 24-hour day). Future studies should consider testing the proposed mediators using more proximal time points, such as multiple measurements throughout the 24-hour day.

In addition to generating novel findings regarding cardiometabolic and behavioral health outcomes, our study provides descriptive evidence of young adults’ preferred activity times, providing a foundation for the design of future time-based lifestyle interventions for young adults. Our study demonstrated that young adult activity timing remained relatively consistent—at the week level—over 18 months across demographic subgroups, although there may be individual variability not captured by our group-level analysis. At all time points, participants accrued a greater proportion of their total activity in the morning and afternoon versus evening and night. These data might allow for the identification and prescription of activity times that are feasible for young adult adherence in trial settings. Future research could develop and test such tailored interventions using randomized messaging designs that incorporate prompts to engage in physical activity prior to/during targeted time windows, and take advantage of opportunities for real-time data capture, such as passive-sensing data and ecological momentary assessment.

The present study provides evidence that incremental health benefits may occur when activity time-of-day is optimized and tailored to individual factors. For example, activity-based interventions designed to improve cardiometabolic health and/or promote weekly physical activity may be enhanced by tailoring timing recommendations to the age and sex of the target population. However, given the substantial evidence to date supporting the physical activity guidelines¹ and the low prevalence (<50%) of young adults meeting these guidelines,⁵⁵ engaging in MVPA at any time of day is still likely of the greatest importance for overall health. At this time, public-facing guidance should continue to emphasize accumulation of activity throughout the day or week, as the study of activity timing remains in its early stages, particularly with regard to recommendations for young adults. Future research in this population could examine whether a programmatic focus on clock-based activity targets (eg, exercising at the same time each day/week) may promote greater accumulation of total activity through habit formation.²²

Strengths and limitations

This study had several strengths. Timing categories were derived from an objective measure of physical activity at baseline, based on a previously established method.^4,47 We conceptualized the explanatory variables (timing of activity in morning, afternoon, evening, and night) as a proportion of the total weekly MVPA, allowing us to assess all lifestyle MVPA and not just planned exercise sessions. Furthermore, analyses adjusted for total weekly MVPA, ensuring that the identified relationships were independent of MVPA volume. Finally, missing data is an issue inherent to measuring behavior via actigraphy. We conducted supplementary sensitivity analyses to explore departures from the MAR assumption. The conclusions drawn are largely robust to departures from the MAR assumption.

There are also limitations to note. A major limitation of this study was that the parent trial did not collect information about participant sleep/wake timing or chronotype, preventing us from controlling for such circadian factors that may also be related to self-selected timing of activity. Chronotype, in particular, may represent unmeasured variability in our population, although our study did control for sex and age, which are known to be associated with chronotype.⁹ Future studies of activity timing in young adults should collect and control for sleep parameters. Similarly, it is important that future studies of this topic collect biological samples at the same clock time each measurement day to control for time-based intraindividual variations. The non-randomized, observational design may have introduced bias, such as confounding by unobserved variables (eg, psychosocial factors, work schedule), which could influence self-selected activity timing and health outcomes. The study sample consisted of young adults with overweight/obesity who were majority White, female, and attending school in an urban area. Thus, our results may not be generalizable to healthy weight populations and/or underrepresented groups, such as minority populations and those residing in rural areas. However, it should be noted that representation of young adult women is a strength of this study. Our study combined 5 cardiometabolic risk factors into a composite CCS, as has been previously established in young adults.²⁸ Future studies could consider testing associations with individual risk factors. The average change from baseline in the weekly MVPA per time category was operationalized at the week level; a description of the intra-week (ie, minute-by-minute) profiles was beyond the scope of this manuscript. We did not adjust for multiple comparisons due to the hypothesis-driven nature of the analyses, per recommendation by Rothman,⁵⁶ and this should be noted when interpreting the results.

Conclusions

In this longitudinal study of young adults with overweight/obesity, the timing of self-selected physical activity was not associated with BMI. However, the timing of physical activity was associated with cardiometabolic disease risk, adjusting for total weekly MVPA and demographic factors. Specifically, morning MVPA was inversely associated with cardiometabolic disease risk for all participants. Evening MVPA was inversely associated with cardiometabolic disease risk for 26-35 year olds only. Finally, morning MVPA was also generally associated with more total weekly MVPA in this sample. Engaging in MVPA at any time of day is likely of the greatest importance for overall health. However, the present study indicates that incremental health benefits may occur when activity time-of-day is optimized and tailored to individual factors.

Author contributions

Caitlin P. Bailey (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Software, Visualization, Writing—original draft, Writing—review & editing), Angelo Elmi (Conceptualization, Investigation, Methodology, Software, Supervision, Writing—review & editing), Jingyi Qian (Conceptualization, Investigation, Methodology, Software, Supervision, Writing—review & editing), Loretta DiPietro (Conceptualization, Funding acquisition, Investigation, Methodology, Resources, Supervision, Writing—review & editing), Mia S. Tackney (Conceptualization, Investigation, Methodology, Software, Supervision, Writing—review & editing), and Melissa A. Napolitano (Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Resources, Supervision Writing—review & editing)

Funding

Research reported in this publication was supported by the National Heart, Lung, And Blood Institute of the National Institutes of Health under Award Number F31HL167355 to CP Bailey, and by the National Institute of Diabetes and Digestive and Kidney Diseases of the National Institutes of Health under award number R01DK100916 to MA Napolitano. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflicts of interest

None declared.

Open science transparency statements

Study registration. This study was not formally registered. Analytic plan pre-registration. The analysis plan was not formally pre-registered.

Data availability

De-identified data from this study are not available in a public archive. De-identified data from this study will be made available (as allowable according to institutional IRB standards) by emailing the corresponding author. The analytic code used to conduct the analyses presented in this study is not available in a public archive. They may be available by emailing the corresponding author. The materials used to conduct the study are not publicly available. They may be available by emailing the corresponding author.