Alignment Between 24-h Light-Dark and Activity-Rest Rhythms Is Associated With Diabetes and Glucose Metabolism in a Nationally Representative Sample of American Adults

The circadian timing system plays a central role in orchestrating behavioral and physiological processes critically involved in glucose metabolism (1). Circadian disruption has been shown to decrease insulin sensitivity in human subjects (2,3). Moreover, epidemiological studies have demonstrated that both severe circadian misalignment (e.g., shift work) and more subtle alterations in the diurnal activity pattern (e.g., decreased amplitude, overall rhythmicity) are associated with a higher risk of type 2 diabetes (4,5).

A key feature of the circadian timing system is its ability to respond and synchronize to rhythmic environmental signals, a process known as entrainment. Different circadian clocks may be entrained by different environmental signals: The master clock in the suprachiasmatic nucleus is primarily entrained by the light-dark pattern (6), while peripheral clocks can be entrained by a wide range of other stimuli, including nutrient intake and physical activities (7,8). Optimal entrainment is manifested as proper alignment among environmental stimuli, behavioral cycles, and physiological processes and is critical to maintaining healthy circadian function (9). To date, most of the epidemiological studies that examined the metabolic effects of circadian rhythms focused on either the behavioral cycle (e.g., activity-rest rhythm) or environmental exposure (e.g., light-dark exposure) alone (5,10–12), while there has been little investigation aimed at studying the unique role of activity-light alignment in metabolic health.

In a large and nationally representative sample of American adults with multiday, 24-h measurements of activity and light, we applied phasor analysis to quantify two important aspects of the alignment between activity-rest and light-dark cycles: the strength of coupling (i.e., phasor magnitude) and relative difference in timing (i.e., phasor angle). We tested the hypothesis that diabetes and impaired glucose metabolism are associated with weakened activity-light alignment as measured by 1) reduced coupling strength and 2) a larger phase difference, specifically a more delayed activity-rest cycle relative to the light-dark cycle. Moreover, we further examined these associations in subpopulations of different age, sex, race and ethnicity, and obesity status because previous research reported differences in both diabetes (13) and circadian/diurnal patterns (14–16) across these subgroups.

The National Health and Nutrition Examination Survey (NHANES) is a nationally representative cross-sectional survey designed to assess the health and nutritional status of adults and children in the U.S. (17). The current analysis used NHANES data from the 2011–2012 and 2013–2014 cycles, when the Physical Activity Monitor component was performed to measure light exposure and activity levels. The NHANES is approved by the National Center for Health Statistics ethics review board.

The NHANES 2011–2014 included a total of 19,931 participants. Of these participants, we excluded those aged ≤20 years (n = 8,829), lacking light or activity data (n = 1,880), with <4 valid days of actigraphy data (see below for definition of valid days, n = 1,908), and with missing HbA_1c values (n = 243), as well as those who were pregnant at the time of the study (n = 58). The remaining 7,013 adults were included in our main analysis focusing on diabetes. In our secondary analysis that focused on glycemic markers, we further excluded participants taking diabetes medications (n = 924) and missing the biomarker of interest. The analytic samples for fasting glucose, fasting insulin, HOMA of insulin resistance (HOMA-IR), and 2-h oral glucose tolerance test (OGTT) results included 2,796, 2,704, 2,704, and 2,603 participants, respectively. The process for sample selection is depicted in Supplementary Fig. 1.

Activity and light data were collected using the ActiGraph GT3X+ (ActiGraph, Pensacola, FL) (1). Briefly, all participants aged ≥6 years were instructed to wear the device on their nondominant wrist for 7 consecutive days. The device recorded triaxial acceleration at 80 Hz and ambient light levels at 1 Hz, which were processed to derive per-minute measures of light (in lux) and activity (the Monitor-Independent Movement Summary [MIMS]). Signal patterns deemed unlikely to reflect human movement were flagged as invalid measurements by quality reviews (18). Moreover, the per-minute data were categorized as wake wear, sleep wear, nonwear, or unknown (19). On the basis of these data, we determined a minute of recording as valid when 1) no quality flag was present and 2) the minute was categorized as either wake wear or sleep wear. We then defined a valid day as having at least 20 h of valid recordings. In our sample, 79.2% participants had 7 days of valid data.

Phasor analysis characterizes the relationship between the 24-h light-dark cycle (i.e., the input to the circadian clock) and the activity-rest cycle (i.e., the output) in terms of angle and magnitude (20,21). Phasor magnitude is the normalized cross covariance of the two cycles and represents the strength of the light-dark and activity-rest coupling exhibited by each individual, with greater values (i.e., longer phasor magnitude) representing better alignment. Phasor angle represents the temporal relationship between the 24-h light-dark and the activity-rest cycles and is similar, but not identical, to the difference between the activity acrophase and the light acrophase. A more positive phasor angle indicates a more delayed activity-rest cycle relative to the light-dark cycle, while a more negative phasor angle signifies a more advanced activity-rest cycle. Light exposure affects the human circadian system in a nonlinear manner (22). While spectral power distribution data were not available for the light these participants experienced, we applied a logistic transformation of the measured light. Furthermore, the circadian system has a saturation level (23). Here, we assumed a saturation level of 10,000 lux, and recorded light levels exceeding this threshold were set to 10,000 lux. These two variables were divided into quintiles, with the groups presumed as having the best alignment serving as the reference (Q5 for phasor magnitude and Q1 for phasor angle). We present 24-h average profiles of light and activity for three participants with varied phasor magnitude and phasor angle in Supplementary Fig. 2.

Markers of diabetes and glycemic control included HbA_1c, fasting plasma glucose, fasting serum insulin, and 2-h OGTT. Of these, HbA_1c was measured in the full sample, while the others were measured in participants who attended the morning examination session only. HbA_1c was measured by the Tosoh Automated Glycohemoglobin Analyzer HLC-723G8 (Tosoh Bioscience, South San Francisco, CA). Blood glucose was measured by the hexokinase method. Insulin was determined by ELISA.

Diabetes (primary outcome) was determined as HbA_1c ≥6.5%, self-reported diagnosis of diabetes, or use of diabetes medication. We further categorized participants without diabetes into low (<5%), normal (5–5.6%), and prediabetic (5.7–6.4%) groups (24). Secondary outcomes included fasting glucose, fasting insulin, HOMA-IR, and OGTT results. HOMA-IR was calculated as fasting glucose (mg/dL) × fasting insulin (μU/mL)/405 (25). OGTT results were based on plasma glucose levels measured 2 h after ingesting 75 g of glucose. We applied log transformation to fasting glucose and insulin, HOMA-IR, and OGTT to improve data normality.

Key covariates included sociodemographic variables (age, sex, race and ethnicity, education, household income, and marital status), lifestyle factors (smoking, alcohol consumption, total caloric intake, sleep duration, and total physical activity), and BMI. All sociodemographic variables, and smoking and alcohol intake were measured by in-person interviews. Total caloric intake was estimated based on two 24-h dietary recalls. Sleep duration was determined as the daily average of total sleep wear minutes categorized as sleep wear, and total physical activity as the daily average of the sum of per-minute MIMS using actigraphy data. BMI was calculated from weight and height measured at the Mobile Examination Center.

Descriptive statistics are presented as median and interquartile range (IQR) for continuous variables and percentages for categorical variables. The odds ratios (ORs) and 95% CIs of quintiles of phasor variables with HbA_1c categories were determined by multinomial logistic regression, with the normal HbA_1c level serving as the reference. Model 1 was adjusted for age and sex. Model 2 (main model) was adjusted for additional confounders, including race and ethnicity, education, household income, marital status, smoking, alcohol consumption, and total energy intake. In sensitivity analyses (model 3), we additionally adjusted for BMI because BMI may act both as a confounder and a mediator of the association between phasor variables and diabetes. Finally, in a second sensitivity analysis model (model 4), we adjusted for BMI, sleep duration, and total physical activity. Sleep and total physical activity were included in this model separately because they were intrinsic components of the 24-h activity cycle, and alterations in sleep and activity patterns may be both drivers and consequences of misalignment between light and activity cycles, thus making them both potential confounders and mediators at the same time. The purpose of models 3 and 4 was to assess to what degree the relationship between phasor variables and diabetes were explained by obesity, sleep, and physical activity. Missing covariates were generally <5% and were imputed as the mode for categorical variables and median for continuous variables.

In secondary analyses, we used multiple linear regression adjusted for covariates in model 2 to determine the association with fasting glucose and insulin, HOMA-IR, and OGTT results. We used model-derived least squares means to calculate back-transformed–predicted geometric means with 95% CIs for each quintile of the phasor variable. We also performed subgroup analyses stratified by age, sex, race and ethnicity, and BMI using multiple logistic regression in which the binary diabetes status served as the outcome. P values for trend were determined by modeling quintiles of phasor variables as continuous (i.e., 1–5 for Q1–Q5). P values for interaction were calculated using the likelihood ratio test, comparing a model with the cross-product term to one without. J.D., who did not perform the main analysis, reproduced the main statistical model and performed k-fold cross-validation, with 10 folds and 1,000 replicates for each phasor variable. The results from the validation analyses were stable (data not shown), with 100% of replicates maintaining a significant estimate of the directional trend. All analyses were performed using SAS 9.4 (SAS Institute, Cary, NC), Matlab R2023a (MathWorks, Natick, MA), and R 4.3.0 (R Foundation for Statistical Computing, Vienna, Austria) statistical software.

In our analytic sample, the weighted average of age and BMI were 50.2 years and 29.2 kg/m², respectively, and a weighted 52% and 68% were female and non-Hispanic White, respectively. The average phasor magnitude and phasor angle were 0.29 (SD 0.11) and 0.93 (1.34), respectively, and the two variables were weakly correlated (Pearson correlation coefficient −0.13). We present descriptive statistics for selected study characteristics by quintiles of phasor magnitude and phasor angle in Table 1. Compared with participants in lower quintiles of phasor magnitude, those in Q5 (i.e., strong light-activity coupling) were more likely to be non-Hispanic White and married but less likely to report a household income of ≤$20,000, be a current smoker, or have a BMI of ≥30 kg/m² and had a higher level of physical activity. They were also more likely to report alcohol consumption of ≥1 drinks/day, sleeping <7 h, and having a higher total energy intake. For phasor angle, the lowest quintile (i.e., small temporal gap between light and activity) was associated with older age, a higher percentage of non-Hispanic White, being married, and having a sleep duration between 7 and 9 h, but a lower percentage reported having a household income ≤$20,000, being a current smoker, or having obesity.

Table 2 presents associations between phasor variables and diabetes. Results from model 1 showed a strong association between both shorter phasor magnitude and larger phasor angle and a higher odds of diabetes. Adjusting for the full set of confounders in model 2 attenuated the association, but the results remained statistically significant. Specifically, compared with the reference groups (Q5 for phasor magnitude and Q1 for phasor angle), participants with the shortest phasor magnitude and the highest phasor angle were 76% (OR 1.76 [95% CI 1.39, 2.24]) and 73% (1.73 [1.34, 2.25]) more likely to have diabetes, respectively (P for trend < 0.0001). Adjusting for BMI in model 3 further attenuated the associations. After adjusting for sleep duration and total physical activity in model 4, the association between phasor magnitude and diabetes became null, but phasor angle remained significantly and positively associated with diabetes. Results from the models with additional adjustment for total light exposure during the 24-h period were largely similar to those without such adjustment (data not shown). No phasor variable was associated with having prediabetes (Supplementary Table 1) or low HbA_1c status (Supplementary Table 2).

Associations between phasor variables and additional markers of glucose metabolism and insulin resistance are presented in Table 3. A shorter phasor magnitude was associated with higher levels of fasting insulin (P for trend = 0.001), HOMA-IR (P for trend = 0.002), and OGTT results (P for trend = 0.0007), while a larger phasor angle was associated with higher levels of fasting insulin (P for trend = 0.02) and HOMA-IR (P for trend = 0.02). Specifically, compared with the longest quintile, participants in the shortest quintile of phasor magnitude had an average of 2.03 μU/mL increase in fasting insulin (predicted geometric means 8.90 μU/mL for Q5 and 10.93 μU/mL for Q1) and a 0.49-unit increase in HOMA-IR (2.20 units for Q5 and 2.69 units for Q1). For phasor angle, the predicted mean differences between Q5 and the reference Q1 groups were 1.19 μU/mL for fasting insulin (9.25 μU/mL for Q1 and 10.64 μU/mL for Q5) and 0.37 units for HOMA-IR (2.27 units for Q1 and 2.64 units for Q5). Fasting glucose was not associated with either phasor variable.

We present the distributions of phasor magnitude and phasor angle by age, sex, race and ethnicity, and BMI subgroups in Supplementary Figs. 3 and 4 and subgroup-specific associations in Fig. 1 and Supplementary Tables 3–6. The associations appeared to be qualitatively similar across most of the subgroups (all P for interaction > 0.05), except for a nominally larger effect size for phasor magnitude among participants in the younger age-group (20–39 years) compared with older age-groups (P for interaction = 0.04).

In a representative sample of noninstitutionalized U.S. adults, our analysis suggested that the alignment between activity-rest and light-dark cycles is associated with glucose metabolism and diabetes. Specifically, people with reduced strength in the coupling of activity and light and a more delayed activity-rest cycle relative to the light-dark cycle were more likely to have diabetes and/or show impaired glucose metabolism as evident by elevated fasting insulin and HOMA-IR. Moreover, these findings were consistently observed across different sociodemographic and BMI groups.

Although our study is the first to examine the activity-light alignment in relation to metabolic health, previous research has suggested that alterations in either activity-rest rhythms or light-dark exposure patterns may be contributing factors to metabolic dysfunction. For example, our previous study of the NHANES data reported that weakened activity-rest rhythms (i.e., lower amplitude, delayed acrophase, reduced overall rhythmicity) were associated with impaired glucose metabolism and diabetes (10). In a cohort of older men, we linked similarly impaired profiles of activity-rest rhythms with higher risk of developing diabetes over 3–5 years of follow-up (5). In experimental studies, human subjects who underwent circadian misalignment protocols also showed impaired metabolic health as evidenced by increases in insulin and glucose (2). Light exposure has also been directly linked with metabolic dysfunction, including diabetes, with light at night associated with detrimental effects and light early in the day showing protective effects. For example, a recent study reported that nighttime light exposure was associated with an 82% increase in the odds of obesity (OR 1.82 [95% CI 1.26, 2.65]) and 100% increase in diabetes (2.00 [1.19, 3.43]) in older adults (11). In another study, a 3-week morning bright light treatment led to reduced weight and body fat in overweight women (12). Our study expanded the literature by showing a unique and important role of the alignment between activity-rest and light-dark, thus highlighting the importance of considering the interrelationship between and combined effects of both behavioral cycles and environmental exposure rhythms in metabolic health. Notably, the light-dark and activity-rest cycles of an individual are interdependent, and intervening on one may affect the other and influence the alignment between the two. Given the growing evidence consistently linking characteristics of these diurnal cycles with metabolic health, it would be of interest for studies to develop and evaluate intervention strategies aimed at manipulating light exposure and activity patterns and/or strengthening their alignment for improving metabolic health. Indeed, clinical trials have shown that enhancing sleep and physical activity, which may improve both activity-rest and light-dark patterns, can lead to better metabolic health (26–28). However, most of such studies did not explicitly focus on improving circadian function, an important topic that warrants future investigation.

Previous studies also have suggested that studying light and activity alone may not fully capture the degree of circadian disruption experienced in humans. For example, on the basis of light-dark and activity-rest data collected from dayshift and rotating nightshift nurses, Miller et al. (20) reported that both the average activity levels and the types of light exposure sources (i.e., natural, electric) were similar between the two groups of nurses, although it is well known that rotating nightshift workers are at a higher risk for diabetes and obesity (29). When using phasor analyses, the authors were able to demonstrate that the alignment between the light-dark exposures and the activity-rest rhythms was much stronger for the dayshift nurses than for the rotating nightshift nurses, as reflected by a significantly longer phasor magnitude in the former group (mean 0.50 [SD 0.11] for dayshift nurses and 0.33 [0.10] for rotating nightshift nurses, P < 0.0001). These data suggest that phasor magnitude is a better metric to describe circadian disruption than either light or activity alone.

The associations presented here are consistent with results from Figueiro et al. (30), who showed that circadian disruption, as measured using the phasor magnitude, was linked to glucose tolerance in mice. In their study, they measured circadian disruption based on phasor analysis and glucose tolerance in mice exposed to light-dark stimulus patterns simulating those that humans would experience while working dayshift or rotating nightshift schedules. They showed that mice exposed to nightshift schedules had significantly shorter phasor magnitude and a higher glucose area under the curve compared with those exposed to dayshift conditions, suggesting that nightshift work may induce both circadian disruptions and impaired glucose metabolism.

Our study also showed a relationship between a more delayed activity-rest rhythm relative to the light-dark exposure and diabetes and impaired glucose metabolism. Previous studies have linked a late chronotype, or delayed sleep window, with metabolic dysfunction. For example, a recent meta-analysis of observational studies found that an evening chronotype was associated with higher levels of HbA_1c, blood glucose, and diabetes risk (31). However, these studies measured chronotype based on conventional definitions of time (e.g., evening, morning, clock time) that may not accurately reflect the phase relationship between the behavioral (i.e., sleep) and the zeitgeber or time cue of the internal circadian clock (i.e., light). Future studies should investigate whether the phase of behavioral cycles relative to light pattern is a better predictor of health risk than the absolute phase measured against clock time.

One of the unique strengths of our study is its large and diverse sample. We observed relatively similar results in multiple demographic subgroups, suggesting that altered alignment between activity-rest and light-dark cycles may be consistently associated with suboptimal metabolic health in many human populations. However, we also noted suggestive evidence for potential subgroup differences that warrant future investigations. For example, the association between phasor magnitude and diabetes was substantially larger among younger adults (age <40 years), among whom there was a fourfold increase in the likelihood of having diabetes in the lowest quintile of phasor magnitude compared with the highest. In our previous study, we also observed a stronger relationship between impaired activity-rest rhythms and diabetes in this age-group than in older adults (10). These findings suggest that circadian dysregulation may play a particularly important role in or share common causes with early-onset metabolic diseases. For example, a previous study showed that compared with participants with average-onset and late-onset diabetes, those with early-onset diabetes had a substantially higher prevalence of several risk factors, including severe obesity, physical inactivity, and smoking (32). These risk factors have also been linked to impaired circadian function (33–35), although the direction of and specific mechanism underlying such associations are complex and may not imply a simple, unidirectional causal relationship. Regardless of the nature of such associations, the consistent relationship between diurnal light and activity patterns and diabetes suggests that relevant behavioral metrics based on wearable data may serve as useful biomarkers for disease monitoring and risk prediction. Given that wearable devices often produce a large number of behavioral biomarkers that are often interrelated with one another, it is beneficial for future studies to use advanced statistical methods (e.g., machine learning) to systematically evaluate the predictive value of these biomarkers and develop algorithms for clinical applications.

Our study has several additional strengths. First, as mentioned above, phasor analysis is a novel way to quantify alignment between light-dark and activity-rest cycles, which likely offers better measures of circadian disruption than looking at light or activity alone. Second, we had multiple glycemic biomarkers that offered a comprehensive measure of glucose metabolism and insulin sensitivity.

Our study also has several limitations. First, the light-dark data were obtained with uncalibrated devices worn at the wrist, which could have resulted in over- or underestimation of light exposures. However, given that phasor analyses look at the cross correlation between light-dark and activity-rest patterns, rather than the absolute values, we do not believe that this had a major impact on our findings. Second, we did not have a biomarker of circadian phase (e.g., dim-light melatonin onset), although collecting such data in a large sample like NHANES would be highly challenging. Third, it has been suggested that not only average behavioral patterns but also variance in behavioral patterns over time may be an important predictor of health outcomes. However, the phasor analysis only produced measures that indicate average alignment patterns over the 7-day recording period, and there is a need for future studies with repeated recordings over an extended period to examine to what degree variability in the activity-rest and light-dark alignment may be associated with metabolic health. Fourth, seasonality or time of the year may influence rest-activity rhythms, light-dark exposure patterns, and metabolic health. However, we do not have information on the date, month, or season when data collection was performed and, thus, were not able to adjust for these variables in the analysis. Fifth, shift workers experience severe circadian misalignment and are at high risk of developing diabetes. However, we do not have information on shift work and, thus, were not able to evaluate how it may have affected our study findings and/or whether the reported associations differed between shift workers and non–shift workers. Sixth, it is possible that a high negative value of phasor angle (i.e., large phase advance of activity relative to light) may also reflect circadian misalignment and be associated with impaired metabolic health. However, only a small fraction (<3%) of our study sample had a phasor angle of −1 or lower, limiting our ability to examine the relationship between advanced activity cycle and metabolic health. Seventh, we did not consider the role of other diurnal behavior characteristics, such as meal timing, which has been previously linked to metabolic health (36,37). Finally, the NHANES is a cross-sectional survey, and we were not able to assess the temporality of the relationship, which is critical to establishing a causal relationship between circadian alignment and metabolic health. Future studies should leverage cohort studies with objectively measured light and activity data to examine the prospective association between the activity-light alignment patterns and health outcomes and disease risk.

In conclusion, our study contributes to the growing literature demonstrating a crucial role of circadian rhythms in metabolic health. Specifically, our study extends existing evidence by highlighting the importance of proper alignment between diurnal behavioral cycles and light exposure patterns, a novel finding that warrants further investigation, particularly in prospective cohorts. If confirmed, our findings suggest that measures of activity-light alignment, such as phasor magnitude and phasor angle, may serve as useful digital biomarkers for risk prediction and disease management. Moreover, future studies should also evaluate the effectiveness of interventions focusing on enhancing the alignment of multiple components of circadian rhythms on metabolic health.

Funding. This work was supported by National Heart, Lung, and Blood Institute grant R21HL165369 and National Institute of Diabetes and Digestive and Kidney Diseases grant R01 DK128972.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. Q.X. and J.D. performed to the statistical analysis. Q.X., J.D., C.B., C.H.C.Y., and M.G.F. critically revised the manuscript for important intellectual content. Q.X., J.D., and M.G.F. interpreted the data. Q.X. and M.G.F. contributed to the study concept and design. Q.X. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.