Temporal Patterns of Diet and Physical Activity and of Diet Alone Have More Numerous Relationships with Health and Disease Status Indicators Compared to Temporal Patterns of Physical Activity Alone

Abstract

Background:

Daily temporal patterns of energy intake (temporal dietary patterns, TDPs) and physical activity (temporal physical activity patterns, TPAPs) have been independently and jointly (temporal dietary and physical activity patterns, TDPAPs) associated with health and disease status indicators.

Objective:

The aim of this study was to compare the number and strength of association between clusters of daily TDPs, TPAPs, and TDPAPs and multiple health and disease status indicators.

Design:

This cross-sectional study used one reliable weekday dietary recall and a random weekday of accelerometer data to partition to create clusters of participants representing the three temporal patterns. Four clusters were created via kernel-k means clustering algorithm of the same constrained dynamic time warping distance computed over the time series for each temporal pattern.

Participants/setting:

From the National Health and Nutrition Examination Survey (years 2003-2006), 1,836 U.S. adults ages 20-65 years who were not pregnant that had valid diet, physical activity, sociodemographic, anthropometric, questionnaire, and health and disease status indicator data were included.

Main outcome measures:

Health status indicators used as outcome measures were body mass index (BMI), waist circumference (WC), fasting plasma glucose, hemoglobin A1c, triglycerides, high-density lipoprotein cholesterol, total cholesterol, systolic and diastolic blood pressure and disease status indicators included obesity, type 2 diabetes mellitus, and metabolic syndrome.

Statistical analyses performed:

Multivariate regression models determined associations between the clusters representing each pattern and health and disease status indicators, controlling for confounders and adjusting for multiple comparisons. The number of significant differences among clusters and adjusted R²/the Akaike information criterion compared the strength of associations between clusters of patterns and continuous/categorical health and disease status indicators.

Results:

TDPAPs showed 21 significant associations with health and disease status indicators including BMI, WC, obesity, and type 2 diabetes; while TDPs showed 19 significant associations, and TPAPs showed 8 significant associations.

Conclusion:

TDPAPs and TDPs had stronger and more numerous associations with health and disease status indicators compared with TPAPs. Patterns representing the integration of daily dietary habits hold promise for early detection of obesity.

INTRODUCTION

Dietary intake and physical activity (PA) are independent modifiable risk factors for obesity and chronic diseases ^1,2. These daily behaviors are also expected to be interrelated, with potential synergistic associations to health ^3,4. A previous intervention⁵ and reviews^6,7 compared the combined effect of diet and PA with the independent effects of each of these behaviors on health and found their combination resulted in higher odds of experiencing health improvements than diet or PA alone including weight loss ⁵, glycemic control⁶, and bone mass⁷. Furthermore, there is a growing interest in when (i.e., time of day) people eat and engage in PA in addition to the amount of energy consumed and intensity of the activity ^8-10. Timing of eating plays an important role in body weight management, for example delayed or nighttime eating is significantly associated with increased body mass index (BMI) and weight gain ¹¹. Timing of exercise may be important for weight management with optimal exercise time varying based on the individual ¹². Both eating and PA behaviors occur within a time frame or “temporal pattern” that is associated with daily rhythms ¹³. “Temporal pattern” refers to a lifestyle behavioral pattern that emphasizes the energy intake relative to dietary intake or intensity relative to PA, and the timing, frequency, and regularity of these behaviors throughout the day ^14-16. The temporal pattern may be important to health because the circadian system has an important role in the regulation of metabolism, physiology, and behavior¹⁷. Diet and PA are behavioral factors that are also impacted by the sleep-wake cycle^18,19. The misalignment of behavior with the circadian system, such as in eating late at night, may impair metabolism and could lead to dysfunction ^20-25.

The authors have created new machine-learning-based clustering methodology to discover temporal dietary patterns (TDPs), temporal PA patterns (TPAPs), and temporal dietary and PA patterns (TDPAPs) using the National Health and Nutrition Examination Survey (NHANES) 2003-2006 ^26-29. The data-driven partitioning methods generated clusters based on the timing and amounts (energy intake and/or PA counts) of the behaviors between every two participants in the sample without additional inputs or influence of predefined standards that could mask the latent characteristics that play a potential role in health and disease status indicators. These studies showed that a TDP with evenly spaced (6:00-10:00, 12:00-15:00, and 18:00-22:00), energy balanced eating occasions, a TPAP with higher PA counts either early (8:00-11:00) or later (16:00-21:00), and a TDPAP with 2 evenly spaced (11:00-13:00 and 17:00-20:00) eating occasions and the highest PA counts from 8:00-20:00 had statistically significantly better indicators of health and disease including higher diet quality ²⁶, lower mean BMI ^27-29, waist circumference (WC) ^27-29, total cholesterol²⁹, and odds of obesity ^27-29 compared to other TDPs, TPAPs, and TDPAPs. However, no studies have compared the temporal patterns of single and multiple behaviors with health and disease status indicators to determine these patterns’ utility at differentiating groups. Evaluating the comparative strength of the various temporal behavioral patterns relationships to health could determine the temporal lifestyle behavior pattern or their combination that is optimal for health. Comparison to multiple health and disease status indicators, both short and long-term, may provide an overall evaluation of pattern utility. Therefore, the objective of this study was to compare number and strength (R² and Akaike information criterion (AIC)) of significant associations of TDP, TPAP, and TDPAP, respectively, with 12 health and disease status indicators: BMI, WC, fasting plasma glucose, hemoglobin A1c, triglycerides, high-density lipoprotein cholesterol (HDL-C), total cholesterol, systolic and diastolic blood pressure and also obesity, type 2 diabetes mellitus (T2DM) and metabolic syndrome (MetS) among U.S. adults aged 20-65 years old using the NHANES 2003-2006. Obesity-related indicators were of special focus as they are risk factors for other health and disease status indicators and previous studies showed their significant links to temporal patterns^27-29. It was hypothesized that the TDPAP clusters would have more significant associations with health and disease status indicators than clusters of TDPs and TPAPs.

METHODS

Participants and Data Collection

NHANES is a cross-sectional survey carried out by the National Center for Health Statistics (NCHS) of the U.S. Centers for Disease Control and Prevention ³⁰. NHANES participants are selected through a complex and multistage sampling design. During the interview in each participant’s household, an in-depth questionnaire is used to collect self-reported sociodemographic information including age, sex, race and ethnicity, and poverty-to-income ratio (PIR). In the physical health examination, a 24-hour dietary recall, anthropometric measures, laboratory tests, and recruitment for the PA assessment component are completed. During the phone follow-up interview 3 to 10 days after the health examination, the second 24-hour dietary recall is collected ³⁰. All participants provided written informed consent to participate in the survey and methods are approved by NCHS Research Ethics Review Board ³¹. The study is exempt from Institutional Review as participants of this public-use data were deidentified before public release of data on NHANES the website³⁰ and are not considered to be human subjects data.

Analytic Sample

This analysis included data from NHANES 2003-2006 due to the availability of PA accelerometer data when the study was initiated. The analytic sample included U.S. adults ages 20-65 years who were not pregnant because the dietary and PA behaviors of pregnant women and participants out of the age range were expected to entail unique life stage patterns. Participants must have had at least one valid 24-hour weekday dietary recall and one random weekday of valid PA accelerometer data to be included. Participants with missing related sociodemographic, anthropometric, questionnaire, and health and disease status indicator data were excluded. Therefore, the study sample n=1,836 (Supplemental Figure 1) was used to generate TDPs, TPAPs, and TDPAPs. Adjustments for the complex survey design (including oversampling of certain groups), survey non-response (including fasting subsample weights considering the additional probability of selection into the subsample component, as well as an additional adjustment for component nonresponse), and post-stratification adjustment to match total population counts from the Census Bureau, were made for representativeness of the U.S. civilian noninstitutionalized resident population.

Dietary Data Assessment

One valid 24-hour weekday dietary recall from each participant with non-zero energy intake and reliable dietary recall status was used to determine the energy consumed throughout the day. One 24-hour dietary recall may be considered representative to estimate the overall dietary pattern on a population level ³². Previous research has revealed differences between weekdays and weekend days of dietary intake ³³, so to maintain the largest possible sample, the first weekday dietary recall was prioritized for selection to generate TDPs and dietary patterns for TDPAPs, otherwise the second weekday recall was used. Both dietary recalls were collected using the U.S. Department of Agriculture (USDA) Automated Multiple-Pass Method ^34,35. The energy intakes of all reported foods and beverages for each participant were determined by the USDA Food and Nutrient Database for Dietary Studies (FNDDS) for 2003-2004 data (USDA FNDDS, version 2.0, Beltsville, MD) ³⁶ and 2005-2006 data (USDA FNDDS, version 3.0, Beltsville, MD) ³⁷. Duration of eating occasions was not available in NHANES, but based on a previous study ³⁸, 15 minutes per occasion was applied such that energy intake reported at each time was divided by 15 minutes and used the reported mealtime as starting time to determine energy per minute. The total energy intake throughout the entire day for each participant was also calculated.

Physical Activity Assessment

PA monitors, ActiGraph model-7164 accelerometers, were worn on the hips of participants and recorded the PA intensity of participants throughout the day for up to 7 consecutive days. The recorded vertical accelerations or “counts per minute” represent the relative measure of changes in momentum that are assigned to one-minute time intervals (epochs) and which estimate PA intensity ^39-41. A valid day of accelerometer wear was defined as at least ten hours of wear time during 24 hours ⁴². In this study, the mean wear time of participants was 15.9 hours. Previous research showed differences between weekday and weekend day PA ⁴³, so to maintain the largest possible sample and consistency with dietary data, a random weekday of accelerometer data was chosen to generate TPAPs and PA patterns for TDPAPs. Even though multiple PA days of accelerometer data may better identify participants’ usual daily total PA, one valid random day of PA data may be considered sufficient at the population level ⁴² and can maintain specific timing and the largest possible sample. The total waking PA counts of the entire 24-hour day selected for each participant were also calculated.

Creating TDPs, TPAPs, and TDPAPs

Detailed descriptions of the methodology used to determine the TDPs, TPAPs, and TDPAPs were previously described ^27-29. Briefly, distance-based clustering analysis with a dynamic time warping (DTW)-type distance measure using energy intake or PA counts or both at minute epochs were used to derive TDPs, TPAPs and TDPAPs ⁴⁴. Constrained DTW (CDTW) determined the optimal matching path such that the sum of the squared differences between the matched eating event or PA or entries that considered both behaviors for each pair of participants were minimized with a Sakoe-Chiba band ⁴⁵ constraint on the maximum temporal difference to avoid pathological warping (e.g., matching activities in the morning to activities in the evening). To generalize CDTW to multi-dimensional time series such as the diet and PA time series, two commonly used methods ⁴⁶, i.e., Independent Multivariate CDTW (based on separately matching diet and PA data, CDTW_I) and Dependent Multivariate CDTW (based on jointly matching diet and PA data, CDTW_D) were adopted. Band constraints ranging from 60 to 720 minutes (in 60-minute increments) were explored. CDTW measures were determined for diet and PA data independently and jointly, and then input to several clustering algorithms including kernel k-means ^47,48, spectral ⁴⁹, and hierarchical ^50,51 clustering to divide individuals into different clusters according to their temporal lifestyle behavior. These clusters and their characteristic features constitute the desired TDPs, TPAPs and TDPAP patterns. Ultimately, CDTW_I with bandwidths (PA: 120 minutes; diet: 480 minutes) and kernel k-means clustering (k=4 based on internal evaluations including Silhouette ⁵² and Dunn Index ⁵³ and external criteria of visualization, time and energy differences among the clusters, and health and disease status indicator analysis of the clusters as described in the section Statistical Analysis) yielded TDPAPs with the most numerous significant relationships with health and disease status indicators. Similar methods with the same corresponding bandwidths were applied to generate TDPs and TPAPs. This approach facilitated a fair methodological comparison where differences among the pattern relationships with health and disease status indicators should be attributed to the behavioral patterns rather than the methods.

Anthropometric Assessment and Laboratory Tests

Anthropometric assessment data and information from the laboratory tests were used as health status indicators in the analysis. Weight ⁵⁴, WC ⁵⁴, BMI, hemoglobin A1c (high performance liquid chromatography using primus CLC 330 and Primus CLC 385 (Primus Corporation, Kansas City, MO) in 2003-2004) and Tosoh A1c 2.2 Plus Glycohemoglobin Analyzer (Tosoh Medics, Inc., San Francisco, CA) in 2005-2006) ^55,56, total cholesterol (measured enzymatically) ^57,58, HDL-C (direct immunoassay method) ^57,59, diastolic and systolic blood pressure (up to 4 measures using mercury sphygmomanometer, 1^st measure used if only one, otherwise measures were averaged) ⁶⁰ were measured or calculated during the physical health examination. After 8.5 to 24 hours fasting, blood samples from participants were drawn ^61,62 to measure triglycerides (measured enzymatically) ^63,64 and fasting plasma glucose (using a hexokinase method with Roche/Hitachi 911 (2003-2004) or Roche Cobas Mira (2005-2006)) ^65,66. The 9 health status indicators included were selected based on their role as common risk factors for major lifestyle-linked chronic diseases in the U.S including heart disease and diabetes⁶⁷. BMI is the priority indicator in this study because it has significant associations with temporal patterns based on previous studies^27-29 and is highly related to other indicators including WC⁶⁸.

Disease Status Indicators Classification

Three disease status indicators were selected for analysis: obesity, defined as BMI ≥30 kg/m^{2 69}, T2DM, and MetS. Number of participants with T2DM was calculated by subtracting out those reporting type 1 diabetes from those reporting a diabetes diagnosis by a doctor, having diabetes, taking glucose-lowering medications, or having fasting plasma glucose ≥126 mg/dL (to convert to mmol/L, multiply by 0.0555), or hemoglobin A1c ≥6.5% ⁷⁰. Type 1 diabetes was classified when a participant reported being diagnosed with diabetes before 30 years old and reporting continuous insulin use since diagnosis ⁷¹. MetS was defined based on the presence of three or more of the following risk factors: 1. WC >102 cm for men or >88 cm for women; 2. triglycerides ≥150 mg/dL(to convert to mmol/L, multiply by 0.0113); 3. HDL-C<40 mg/dL(to convert to mmol/L, multiply by 0.0259) in men or <50 mg/dL(to convert to mmol/L, multiply by 0.0259) in women; 4. hypertension (systolic blood pressure ≥130 / diastolic blood pressure≥85 mmHg); and 5. impaired fasting glucose >110 mg/dL(to convert to mmol/L, multiply by 0.0555)⁷².

Measures for Covariates

Demographic variables included survey year (2003-2004 and 2005-2006), sex (male and female), race and ethnicity (Mexican American, other Hispanic, non-Hispanic white, non-Hispanic black, and other including multi-race), age group (20-34, 35-49, and 50-65 years), PIR (the ratio of reported household income to the federal poverty guideline for household income, 0-0.99 indicating PIR below the guideline, 1-1.99, 2-2.99, 3-3.99, 4-4.99, and 5 or more) ⁷³. BMI was classified as underweight and normal weight (≤24.9 kg/m²), overweight (25.0-29.9 kg/m²), and obese (≥30.0 kg/m²) ⁶⁹. Energy misreporting was adjusted to mitigate the inclusion of implausible value/outliers, and was calculated as the ratio of total energy intake to estimated energy requirement (EER) ^74-76. EER can be calculated using the Dietary Reference Intake equations for adults based on age, sex, weight, height, and PA level ⁷⁷. A low PA level was applied to determine the EER based on mean PA counts and previous studies ^78,79. Clustering and stratification were accounted for as NCHS guidelines⁸⁰ direct to obtain reliable estimates. Survey weights were applied to match the survey population at the midpoint of 2003-2006 to produce unbiased nationally representative estimates ^81-83.

Statistical Analysis

Residual plots and outliers were checked for all continuous health status indicators; suspected outliers for triglycerides, hemoglobin A1c, and fasting plasma glucose did not influence the results and were considered biologically plausible, so they were retained. The receiver operating characteristic curves were used to evaluate the models including categorical disease status indicators and showed that the area under the curves were all more than 0.8.

Analysis of variance (ANOVA) and Kruskal-Wallis test were used to evaluate each set of temporal pattern clusters with continuous health status indicators. Multiple linear regression models compared clusters of the three temporal patterns on continuous health status indicators while multiple logistic regression models were used for categorical disease status indicators. BMI, WC, and obesity models were adjusted for survey year, sex, race and ethnicity, age group, PIR, total PA counts and energy misreporting and multiple comparisons (Tukey-Kramer adjustment) while other health and disease status indicators were additionally adjusted for BMI. Significance of comparisons between clusters was set at adjusted P≤0.05. The strength between health status indicators and temporal pattern were evaluated through the number of significant differences among clusters and adjusted R² where a larger R² indicates that more of the variation in the health status indicator was explained by the temporal lifestyle behavior pattern in the linear regression model. Similarly, the AIC is an indicator for strength of the relationship for logistic regression models where a lower AIC means that the temporal lifestyle behavior pattern explains a greater amount of the variation in the disease status indicators ^84-86. All analysis was completed using SAS (version 9.4)⁸⁷.

RESULTS

Visualization

The visualization (Figure 2) shows the distribution of non-zero eating occasions, non-zero PA counts, or both, in each cluster of TDPs (Figure 2a), TPAPs (Figure 2b), and TDPAPs (Figure 2c & 2d) using heat maps. Eating or activity occasions in the heat map are marked by time (x-axis: 00:00-24:00) and absolute amount of energy intake ranging from 0 kcal-4,000 kcal or PA counts ranging from 0 to > 1.2× 10⁵ (truncated) (y-axis). The proportion of individuals in each cluster of TDPs (e.g. D1 represents Cluster 1 of TDPs), TPAPs (e.g. P1 represents Cluster 1 of TPAPs), and TDPAPs (e.g. DP1 represents Cluster 1 of TDPAPs) reporting energy intake, PA, or both, is indicated by the shading and ranged from 0.0% to 30.6% (TDPs), 0.0% to 13.8% (TPAPs), 0.0% to 30.8% (TDPAPs’ energy intake), 0.0% to 14.0% (TDPAPs’ PA) of each cluster in the three temporal patterns, respectively. Darker shading represents a greater proportion of the cluster reporting the amount of energy intake or PA counts at that specific hour.

Figure 2a.

Energy intake heat maps representing four distinct Temporal dietary patterns (TDPs) for U.S. adults 20-65 years drawn from National Health and Nutrition Examination Survey 2003-2006 (Cluster 1 (D1): n=521, Cluster 2 (D2): n=338, Cluster 3 (D3): n=767, Cluster 4 (D4): n=210). Distribution of 1836 participants’ energy intake of 4 TDP clusters is shown using heat maps. The absolute energy intake ranging from 0 to 4,000 kcal (y-axis) ranging 00:00-24:00 hrs at hourly level (x-axis) is depicted. The proportion of participants in each cluster reporting energy intake is represented through shading ranging from 0.0% to 30.6% of participants of 4 TDP clusters. The darker shading represented that a greater percentage of participants in the cluster reported the same amount of energy intake at that time.

^aD1, D2, D3, D4: Cluster 1, 2, 3, and 4 of temporal dietary pattern, respectively.

Figure 2b.

PA heat maps representing four distinct temporal physical activity patterns (TPAPs) for U.S. adults 20-65 years drawn from National Health and Nutrition Examination Survey 2003-2006 (Cluster 1 (P1): n=211, Cluster 2 (P2): n=686, Cluster 3 (P3): n=321, Cluster 4 (P4): n=618). Distribution of physical activity (PA) counts of 4 TPAP clusters is shown using heat maps. PA counts ranging from 0 count per hour to 1.2×10⁵ counts per hour (y-axis, truncated) ranging 00:00-24:00 hrs (x-axis) is depicted. The proportion of participants in each cluster reporting PA is represented through shading ranging from 0.0% to 13.8% of participants of 4 TPAP clusters. The darker shading represented that a greater percentage of participants in the cluster reported the same number of PA counts at that time.

^aP1, P2, P3, P4: Cluster 1, 2, 3, and 4 of temporal physical activity pattern, respectively.

Figure 2c.

Energy intake heat maps representing four distinct temporal dietary and physical activity patterns (TDPAPs) for U.S. adults 20-65 years drawn from National Health and Nutrition Examination Survey 2003-2006 (Cluster 1 (DP1): n=382, Cluster 2 (DP2): n=586, Cluster 3 (DP3): n=591, Cluster 4 (DP4): n=277). Distribution of 1836 participants’ energy intake of 4 TDPAP clusters is shown using heat maps. The absolute energy intake ranging from 0 to 4,000 kcal (y-axis) ranging 00:00-24:00 hrs at hourly level (x-axis) is depicted. The proportion of participants in each cluster reporting energy intake is represented through shading ranging from 0.0% to 30.8% of participants of 4 TDPAP clusters. The darker shading represented that a greater percentage of participants in the cluster reported the same amount of energy intake at that time.

^aDP1, DP2, DP3, DP4: Cluster 1, 2, 3, and 4 of temporal dietary and physical activity pattern, respectively.

Figure 2d.

PA heat maps representing four distinct temporal dietary and physical activity patterns (TDPAPs) for U.S. adults 20-65 years drawn from National Health and Nutrition Examination Survey 2003-2006 (Cluster 1 (DP1): n=382, Cluster 2 (DP2): n=586, Cluster 3 (DP3): n=591, Cluster 4 (DP4): n=277). Distribution of physical activity (PA) counts of 4 TDPAP clusters is shown using heat maps. PA counts ranging from 0 count per hour to 1.2×10⁵ counts per hour (y-axis, truncated) ranging 00:00-24:00 hrs (x-axis) is depicted. The proportion of participants in each cluster reporting PA is represented through shading ranging from 0.0% to 14.0% of participants of 4 TDPAP clusters. The darker shading represented that a greater percentage of participants in the cluster reported the same number of PA counts at that time.

^aDP1, DP2, DP3, DP4: Cluster 1, 2, 3, and 4 of temporal dietary and physical activity pattern, respectively.

Temporal Characteristics of TDPs, TPAPs, and TDPAPs

Characteristics of TDPs, TPAPs, and TDPAPs are summarized in Table 1 based on the cluster visualizations in Figure 2. For TDPs, the energy intake timing was similar among 4 clusters (11:00-13:00 and 18:00-21:00), but the amount was different, D3 energy was evenly distributed in 3 eating occasions (7:00-10:00, 11:00-14:00, and 17:00-21:00) and was the lowest throughout the day compared to other 3 clusters. D2 and D4 displayed more energy intake in the evening, while D1 displayed more energy intake at noon. For TPAPs, timing and intensity of PA were different among the 4 clusters. P1 demonstrated the highest PA counts from 6:00-18:00, P2 demonstrated the lowest from 7:00-18:00, and P3 and P4 demonstrated similar activity counts in the middle among the 4 clusters. P4 displayed high intensity of activity in the morning (6:00-13:00), while P3 displayed high intensity in the afternoon (13:00-20:00). For TDPAPs, with respect to dietary intake, timing of energy intake was similar with peaks at 12:00 and 19:00 but the amount varied among the 4 clusters. DP4 had the highest energy intake (11:00-13:00 and 18:00-21:00) and DP3 had the lowest among the 4 clusters from 7:00-10:00, 11:00-14:00 and 17:00-21:00, while DP1 and DP2 were in the middle. With respect to PA, DP1 demonstrated the highest activity counts, while DP3 demonstrated the lowest activity counts among 4 clusters. The intensity of activity in DP1 had a peak around 10:00, however, DP4 had an activity peak around 17:00, and DP2 and DP3 had comparatively consistent activity counts from 10:00-18:00.

Table 1.

Qualitative description of separate energy and physical activity pattern clusters representing temporal dietary patterns, temporal physical activity patterns, and temporal dietary and physical activity patterns of U.S. adults 20-65 years as drawn from the National Health and Nutrition Examination Survey, 2003-2006 (n=1836).

	Clusters	N(%)	Characteristics of diet patterns
Temporal Dietary Patterns	D1a	521 (28.4)	Peaks in energy intake at two main occasions reaching up to 1,400 and 1,200 kcal from 11:00 to 13:00 and 18:00 to 21:00 with 51.4% and 72.4% of the cluster consuming 400-1,000 kcal and 0-600 kcal, respectively
	D2a	338 (18.4)	Peaks in energy intake at two main occasions reaching up to 1,600 kcal and 1,800 kcal from 11:00 to 13:00 and 17:00 to 20:00 with 53.9% and 50.0% of the cluster consuming 400-800 kcal and 0-600 kcal, respectively
	D3a	767 (41.8)	Proportionally equivalent peaks in energy intake at three main occasions reaching up to 800 kcal from 7:00-10:00, 11:00- 14:00, and 17:00 to 21:00 with 72.0%, 69.6% and 80.5% of the cluster consuming 0-400 kcal, respectively
	D4a	210 (11.4)	Peaks in energy intake at two main occasions reaching up to 2,000 and 2,800 kcal from 11:00 to 13:00 and 18:00 to 21:00 with 59.1% and 68.6% of the cluster consuming 200-1,000 kcal and 200-1,200 kcal, respectively
Temporal Dietary and Physical Activity Patterns	DP1 b	382 (20.8)	Peaks in energy intake at two main occasions reaching up to 1,200 and 1,000 kcal from 11:00 to 13:00 and 17:00 to 20:00 with 66.5% and 80.4% of the cluster consuming 0-600 kcal, respectively
	DP2 b	586 (31.9)	Proportionally equivalent peaks in energy intake at two main occasions reaching up to 1,400 kcal from 11:00 to 14:00 and 18:00 to 21:00 with 57.0% and 59.9% of the cluster consuming 200-800 kcal, respectively
	DP3 b	591 (32.2)	Proportionally equivalent peaks in energy intake at three main occasions reaching up to 800 kcal from 7:00 to 10:00, 11:00 to 14:00 and 17:00 to 21:00 with 71.9%, 69.0%, and 80.4% of the cluster consuming 0-400 kcal, respectively
	DP4 b	277 (15.1)	Peaks in energy intake at two main occasions reaching up to 2,400 and 2,800 kcal from 11:00 to 13:00 and 18:00 to 21:00 with 48.7% and 47.7% of the cluster consuming 200-1,000 kcal and 400-1,200 kcal, respectively
	Clusters	N(%)	Characteristics of PAc patterns
Temporal Physical Activity Patterns	P1 d	211 (11.5)	Highest PAc counts reaching >1.2×105 cphe with 82.0% of the cluster engaged in PAc counts between 3.3×104 −5.1×104 cphe from 6:00-18:00
	P2 d	686 (37.4)	Lowest PAc counts reaching up to 9.9×104 cphe with 95.7% of the cluster engaged in PAc counts between 0.3×104-1.2×104 cphe from 7:00-18:00
	P3 d	321 (17.5)	High PAc counts reaching >1.2×105 cphe with 91.6% of the cluster engaged in PAc counts between 0.9×104 −2.4×104 cphe from 13:00-20:00
	P4 d	618 (33.6)	High PAc counts reaching >1.2×105 cphe with 95.3% of the cluster engaged in PAc counts between 0.9×104 −2.4×104 cphe from 6:00-13:00
Temporal Dietary and Physical Activity Patterns	DP1 b	382 (20.8)	Highest PAc counts reaching >1.2×105 cphe with 91.6% of the cluster engaged in PAc counts between 2.4×104 −4.5×104 cphe from 9:00-19:00
	DP2 b	586 (31.9)	Low PAc counts reaching >1.2×105 cphe with 94.9% of the cluster engaged in PAc counts between 0.6×104-2.1×104 cphe from 10:00-18:00
	DP3 b	591 (32.2)	Lowest PAc counts reaching >1.2×105 cphe with 97.5% of the cluster engaged in PAc counts between 0.3×104 −1.8×104 cphe from 10:00-18:00
	DP4 b	277 (15.1)	High PAc counts reaching >1.2×105 cphe with 91.7% of the cluster engaged in PAc counts between 0.9×104 −2.7×104 cphe from 11:00-19:00

^aD1, D2, D3, D4: Cluster 1, 2, 3, and 4 of temporal dietary pattern, respectively.

^bDP1, DP2, DP3, DP4: Cluster 1, 2, 3, and 4 of temporal dietary and physical activity pattern, respectively.

^cPA: physical activity

^dP1, P2, P3, P4: Cluster 1, 2, 3, and 4 of temporal physical activity pattern, respectively.

^ecph: counts per hour

Demographic Characteristics of TDPs, TPAPs, and TDPAPs

The demographic characteristics among the clusters of the three temporal patterns all varied significantly by sex and age groups (all P<0.0001) (Table 2). TDPs additionally showed significant differences in race and ethnicity (P =0.003); TPAPs showed significant differences in household PIR (P =0.04) and BMI (P <0.0001); and TDPAPs showed significant differences in race and ethnicity (P =0.02), household PIR (P =0.004), and BMI (P <0.0001) (Table 2).

Table 2.

Characteristics of clusters representing temporal dietary patterns, temporal physical activity patterns, and temporal dietary and physical activity patterns of U.S. adults 20-65 years drawn from the National Health and Nutrition Examination Survey, 2003-2006 (n=1836).

		Temporal Dietary Patterns				Temporal Physical Activity Patterns				Temporal Dietary and Physical Activity Patterns
Characteristics	Total (n)	D1, n(%) a	D2, n(%) a	D3, n(%) a	D4, n(%) a	P1, n(%) b	P2, n(%) b	P3, n(%) b	P4, n(%) b	DP1, n(%) c	DP2, n(%) c	DP3, n(%) c	DP4, n(%) c
Total	1836	521 (28.4)	338 (18.4)	767 (41.8)	210 (11.4)	211 (11.5)	686 (37.4)	321 (17.5)	618 (33.6)	382 (20.8)	586 (31.9)	591 (32.2)	277 (15.1)
Survey year	n(%)	p-valued=0.42				p- valued =0.09				p- valued =0.88
2003-2004	895 (48.7)	258 (49.5)	153 (45.3)	373 (48.6)	111 (52.9)	108 (51.2)	348 (50.7)	165 (51.4)	274 (44.3)	182 (47.6)	285 (48.6)	287 (48.6)	141 (50.9)
2005-2006	941 (51.3)	263 (50.5)	185 (54.7)	394 (51.4)	99 (47.1)	103 (48.8)	338 (49.3)	156 (48.6)	344 (55.7)	200 (52.4)	301 (51.4)	304 (51.4)	136 (49.1)
Sex		p- valued <0.0001				p- valued <0.0001				p- valued <0.0001
Male	933 (50.8)	307 (58.9)	210 (62.1)	241 (31.4)	175 (83.3)	159 (75.4)	249 (36.3)	177 (55.1)	348 (56.3)	213 (55.8)	327 (55.8)	158 (26.7)	235 (84.8)
Female	903 (49.2)	214 (41.1)	128 (37.9)	526 (68.6)	35 (16.7)	52 (24.6)	437 (63.7)	144 (44.9)	270 (43.7)	169 (44.2)	259 (44.2)	433 (73.3)	42 (15.2)
Race and Ethnicity		p- valued =0.003				p- valued =0.08				p- valued =0.02
Mexican American	389(2 1.2)	132 (25.3)	51 (15.1)	176 (22.9)	30 (14.3)	46 (21.8)	131 (19.1)	56 (17.4)	156 (25.2)	98(25.7 )	117 (20.0)	131 (22.2)	43 (15.5)
Other Hispanic	57 (3.1)	16 (0.9)	7 (0.4)	27 (1.5)	7 (0.4)	8 (3.8)	16 (2.3)	11 (3.4)	22 (3.6)	19 (5.0)	17 (2.9)	14 (2.4)	7 (2.5)
Non-Hispanic white	910 (49.6)	244 (46.8)	192 (56.8)	365 (47.6)	109 (51.9)	108 (51.2)	346 (50.4)	162 (50.5)	294 (47.6)	185 (48.4)	298 (50.8)	284 (48.1)	143 (51.6)
Non-Hispanic black	385(2 1.0)	99 (19.0)	71 (21.0)	161 (21.0)	54 (25.7)	41 (19.4)	150 (21.9)	76 (23.7)	118 (19.1)	66 (17.3)	119 (20.3)	128 (21.7)	72 (26.0)
Othere	95 (5.2)	30 0.8)	17 (5.0)	38 (5.0)	10 (4.8)	8 (3.8)	43 (6.3)	16 (5.0)	28 (4.5)	14 (3.7)	35 (6.0)	34 (5.8)	12 (4.3)
Age group (year)		p- valued <0.0001				p- valued <0.0001				p- valued <0.0001
20-34	560 (30.5)	189 (36.3)	100 (29.6)	187 (24.4)	84 (40.0)	97 (46.0)	154 (22.4)	141 (43.9)	168 (27.2)	137 (35.9)	185 (31.6)	118 (20.0)	120 (43.3)
35-49	623 (33.9)	173 (33.2)	120 (35.5)	250 (32.6)	80 (38.1)	82 (38.8)	190 (27.7)	115 (35.8)	236 (38.2)	149 (39.0)	189 (32.2)	182 (30.8)	103 (37.2)
50-65	653 (35.6)	159 (30.5)	118 (34.9)	330 (43.0)	46 (21.9)	32 (15.2)	342 (49.9)	65 (20.3)	214 (34.6)	96 (25.1)	212 (36.2)	291 (49.2)	54 (19.5)
Household PIRf		p- valued =0.43				p- valued =0.04				p- valued =0.004
0-0.99	288 (15.7)	81 (15.6)	45 (13.3)	130 (17.0)	32 (15.2)	32 (15.2)	133 (19.4)	45 (14.0)	78 (12.6)	47 (12.3)	81 (13.8)	113 (19.1)	47 (17.0)
1.00-1.99	427 (23.3)	118 (22.7)	70 (20.7)	178 (23.2)	61 (29.1)	52 (24.6)	166 (24.2)	70 (21.8)	139 (22.5)	87 (22.8)	138 (23.5)	135 (22.8)	67 (24.2)
2.00-2.99	280 (15.2)	83 (15.9)	48 (14.2)	120 (15.6)	29 (13.8)	32 (15.2)	111 (16.2)	53 (16.5)	84 (13.6)	43 (11.3)	92 (15.7)	100 (16.9)	45 (16.2)
3.00-3.99	277 (15.1)	84 (16.1)	60 (17.8)	109 (14.2)	24 (11.4)	29 (13.7)	94 (13.7)	51 (15.9)	103 (16.7)	76 (19.9)	95 (16.2)	75 (12.7)	31 (11.2)
4.00-4.99	168 (9.1)	45 (8.6)	32 (9.5)	65 (8.5)	26 (12.4)	20 (9.5)	56 (8.1)	27 (8.4)	65 (10.5)	36 (9.4)	47 (8.0)	51 (8.6)	34 (12.3)
≥ 5.00	396 (21.6)	110 (21.1)	83 (24.5)	165 (21.5)	38 (18.1)	46 (21.8)	126 (18.4)	75 (23.4)	149 (24.1)	93 (24.3)	133 (22.7)	117 (19.8)	53 (19.1)
BMIg		p- valued =0.39				p- valued <0.0001				p- valued <0.0001
Underweight and Normal weight	543 (29.6)	154 (29.6)	86 (25.4)	239 (31.1)	64 (30.5)	88 (41.7)	174 (25.4)	112 (34.9)	169 (27.3)	142 (37.2)	130 (22.2)	175 (29.6)	96 (34.7)
Overweight	630 (34.3)	188 (36.1)	118 (34.9)	259 (33.8)	65 (30.9)	80 (37.9)	196 (28.6)	122 (38.0)	232 (37.5)	139 (36.4)	208 (35.5)	192 (32.5)	91 (32.9)
Obese	663 (36.1)	179 (34.3)	134 (39.7)	269 (35.1)	81 (38.6)	43 (20.4)	316 (46.1)	87 (27.1)	217 (35.1)	101 (26.4)	248 (42.3)	224 (37.9)	90 (32.5)

^aD1, D2, D3, D4: Cluster 1, 2, 3, and 4 of temporal dietary pattern, respectively. Values n(%) are unweighted.

^bP1, P2, P3, P4: Cluster 1, 2, 3, and 4 of temporal physical activity pattern, respectively. Values n(%) are unweighted.

^cDP1, DP2, DP3, DP4: Cluster 1, 2, 3, and 4 of temporal dietary and physical activity pattern, respectively. Values n(%) are unweighted.

^dRao Scott F adjusted χ² p-value is a goodness-of-fit, one-sided test; statistical significance is indicated when p≤0.05. Analyses were adjusted for clustering and stratification. Sample weights were constructed and applied to the analysis as directed by National Center for Health Statistics⁸⁰. Weights were rescaled so that the sum of the weights matched the survey population at the midpoint of 2003-2006.

^eOther indicates other race and ethnicity including multi-racial.

^fPIR: poverty income ratio

^gBMI: Body mass index, the categories were defined per the World Health Organization⁶⁹.

Associations of TDPs, TPAPs, and TDPAPs with Health and Disease Status Indicators

Clusters of participants representing the three temporal patterns were all significantly associated with BMI (Table 3), WC (Table 4), and odds of being obese (Table 5). D3 representing TDPs, had significantly lower mean BMI, odds of being obese, and smaller mean WC than the other 3 clusters (Table 3-5). Results of TPAPs showed P2 had significantly higher mean BMI compared to those in P1 and P3 (Table 3), and significantly larger mean WC compared to those in P3 (Table 4). P4 had significantly higher odds of being obese compared to those in P1 (Table 5). Results of the TDPAPs showed that DP3 had significantly lower mean BMI, odds of being obese and smaller mean WC than the other 3 clusters (Table 3-5). Participants in DP1 had significantly lower mean BMI, odds of being obese and smaller mean WC compared to those in DP2 (Table 3-5).

Table 3.

Adjusted regression model results for mean body mass index (kg/m²) with clusters representing temporal dietary patterns, temporal physical activity patterns, temporal dietary and physical activity patterns of U.S. adults 20-65 years as drawn from the National Health and Nutrition Examination Survey, 2003-2006 (n=1836).

Temporal Dietary Patterns
Adjusted models a	n (%)	BMI (kg/m2) , Mean (SEM) b	βc ± SEd compared to D2e	95% CI f	p-value	βc ± SEd compared to D3 e	95% CI f	p-value	βc± SEd compared to D4 e	95% CI f	p-value
D1 e	521 (28.4)	28.8 (0.3)	0.6± 0.4	−0.4, 1.6	0.11	5.1 ± 0.4	4.0, 6.1	<0.0001	−5.0 ± 0.6	−6.5, −3.5	<0.0001
D2 e	338 (18.4)	29.6 (0.4)				4.5 ± 0.4	3.5, 5.4	<0.0001	−5.6 ± 0.6	−7.3, −3.8	<0.0001
D3 e	767 (41.8)	28.7 (0.2)							−10.1 ± 0.7	−12.1, −8.1	<0.0001
D4 e	210 (11.4)	29.0 (0.5)
Temporal Physical Activity Patterns
Adjusted models a	n (%)	BMI (kg/m2), Mean (SEM) b	βc ± SEd compared to P2g	95% CI f	p-value	βc ± SEd compared to P3 g	95% CI f	p-value	βc± SEd compared to P4 g	95% CI f	p-value
P1 g	211 (11.5)	26.8 (0.3)	−2.2± 0.8	−4.3, −0.0	0.05	−0.3 ± 0.5	−1.6, 1.1	0.95	−1.4 ± 0.6	−3.1, 0.2	0.10
P2 g	686 (37.4)	30.2 (0.3)				1.9 ± 0.6	0.3, 3.6	0.02	0.7 ± 0.6	−0.8, 2.2	0.54
P3 g	321 (17.5)	27.6 (0.3)							−1.2 ± 0.5	−2.7, 0.3	0.16
P4 g	618 (33.6)	28.9 (0.3)
Temporal Dietary and Physical Activity Patterns
Adjusted models a	n (%)	BMI (kg/m2), Mean (SEM) b	βc ± SEd compared to DP2h	95% CI f	p-value	βc ± SEd compared to DP3 h	95% CI f	p-value	βc’ SEd compared to DP4 h	95% CI f	p-value
DP1 h	382 (20.8)	27.5 (0.3)	−2.9± 0.5	−4.2, −1.6	<0.0001	1.5 ± 0.5	0.1, 2.9	0.04	−5.3 ± 0.6	−6.8, −3.8	<0.0001
DP2 h	586 (31.9)	30.0 (0.3)				4.3 ± 0.4	3.3, 5.4	<0.0001	−2.4 ± 0.5	−3.9, −0.9	0.0007
DP3 h	591 (32.2)	29.1 (0.3)							−6.8 ± 0.6	−8.5, −5.1	<0.0001
DP4 h	277 (15.1)	29.1 (0.4)

^aModels were adjusted for survey year, age group, sex, race/ethnicity, poverty to income ratio, energy misreporting, and physical activity count/day.

^bBMI: body mass index. Values are mean (standard error of the mean).

^cß represents the difference of mean body mass index between two compared clusters. Least square means were used to calculate the differences in mean body mass index.

^dSE: Standard error

^eD1, D2, D3, D4: Cluster 1, 2, 3, and 4 of temporal dietary pattern, respectively.

^fCI: Confidence Interval

^gP1, P2, P3, P4: Cluster 1, 2, 3, and 4 of temporal physical activity pattern, respectively.

^hDP1, DP2, DP3, DP4: Cluster 1, 2, 3, and 4 of temporal dietary and physical activity pattern, respectively.

Table 4.

Adjusted regression model results for mean waist circumference (WC) (cm) with clusters representing temporal dietary patterns, temporal physical activity patterns, and temporal dietary and physical activity patterns of U.S. adults 20-65 years as drawn from the National Health and Nutrition Examination Survey, 2003-2006 (n=1836).

Temporal Dietary Patterns
Adjusted Modelsa	n (%)	WC (cm), Mean (SEM) b	βc ± SEd compared to D2e	95% CI f	p-value	βc ± SEd compared to D3 e	95% CI f	p-value	βc± SEd compared to D4 e	95% CI f	p-value
D1 e	521 (28.4)	98.4 (0.6)	1.5 ± 1.0	−1.1, 4.2	0.12	13.3 ± 1.0	10.5, 16.0	<0.0001	−11.8 ± 1.4	−15.6, −8.1	<0.0001
D2 e	338 (18.4)	100.6 (0.9)				11.7 ± 0.8	9.6, 13.8	<0.0001	−13.4 ± 1.7	−18.0, −8.8	<0.0001
D3 e	767 (41.8)	96.3 (0.5)							−25.1± 1.9	−30.3, −19.9	<0.0001
D4 e	210 (11.4)	99.7 (1.1)
Temporal Physical Activity Patterns
Adjusted modelsa	n (%)	WC (cm), Mean (SEM) b	βc ± SEd compared to P2g	95% CI f	p-value	βc ± SEd compared to P3 g	95% CI f	p-value	βc± SEd compared to P4 g	95% CI f	p-value
P1 g	211 (11.5)	93.6 (0.9)	−5.0 ± 2.3	−11.2, 1.2	0.15	−0.1 ± 1.4	−3.9, 3.8	0.99	−3.0 ± 1.6	−7.3, 1.4	0.28
P2 g	686 (37.4)	101.2 (0.6)				4.9 ± 1.7	0.4, 9.4	0.03	2.0 ± 1.4	−1.7, 5.8	0.47
P3 g	321 (17.5)	94.4 (0.8)							−2.9 ± 1.3	−6.5, 0.8	0.17
P4 g	618 (33.6)	98.2 (0.6)
Temporal Dietary and Physical Activity Patterns
Adjusted modelsa	n (%)	WC (cm), Mean (SEM) b	βeSEd compared to DP2h	95% CI f	p-value	βc ± SEd compared to DP3 h	95% CI f	p-value	βc ± SEd compared to DP4 h	95% CI f	p-value
DP1 h	382 (20.8)	94.5 (0.7)	−7.2 ± 1.2	−10.4, −3.9	<0.0001	3.9 ± 1.2	0.5, 7.3	0.02	−13.2± 1.4	−16.9, −9.4	<0.0001
DP2 h	586 (31.9)	101.3 (0.7)				11.1 ± 1.1	8.2, 14.0	<0.0001	−6.0 ± 1.4	−9.9, −2.1	0.001
DP3 h	591 (32.2)	97.5 (0.6)							−17.1 ± 1.6	−21.5, −12.7	<0.0001
DP4 h	277 (15.1)	97.7 (0.9)

^aModels were adjusted for survey year, age group, sex, race/ethnicity, poverty income ratio, energy misreporting, and physical activity count/day.

^bWC: Waist circumference. Values are Mean (Standard Error of the Mean).

^cß represents the difference of mean WC between two compared clusters. Least square means were used to calculate the differences in mean WC.

^dSE: Standard error

^eD1, D2, D3, D4: Cluster 1, 2, 3, and 4 of temporal dietary pattern, respectively.

^fCI: Confidence Interval

^gP1, P2, P3, P4: Cluster 1, 2, 3, and 4 of temporal physical activity pattern, respectively.

^hDP1, DP2, DP3, DP4: Cluster 1, 2, 3, and 4 of temporal dietary and physical activity pattern, respectively.

Table 5.

Odds ratio (OR) of obesity relative to normal weight status and covariate-adjusted regression model results for clusters representing temporal dietary patterns, temporal physical activity patterns, and temporal dietary and physical activity patterns of U.S. adults 20-65 years as drawn from the National Health and Nutrition Examination Survey, 2003-2006 (n=1836).

Temporal Dietary Patterns
Adjusted modelsa	n (%)	ORb compared to D2 c	95% CI d	p-value	ORb compared to D3 c	95% CI d	p-value	ORb compared to D4 c	95% CI d	p-value
D1 c	521 (28.4)	1.1	0.7, 1.9	0.93	8.4	6.1, 11.7	<0.0001	0.1	0.1, 0.2	<0.0001
D2 c	338 (18.4)				7.5	4.5, 12.4	<0.0001	0.1	0.1, 0.2	<0.0001
D3 c	767 (41.8)							0.0	0.0, 0.0	<0.0001
D4 c	210 (11.4)
Temporal Physical Activity Patterns
Adjusted modelsa	n (%)	ORb compared to P2 e	95% CI d	p-value	ORb compared to P3 e	95% CI d	p-value	ORb compared to P4 e	95% CId	p-value
P1e	211 (11.5)	0.4	0.2, 1.1	0.11	0.8	0.4, 1.5	0.75	0.5	0.2, 1.0	0.04
P2e	686 (37.4)				1.9	0.9, 3.9	0.13	1.1	0.6, 2.1	0.96
P3e	321 (17.5)							0.6	0.3, 1.1	0.12
P4e	618 (33.6)
Temporal Dietary and Physical Activity Patterns
Adjusted modelsa	n (%)	ORb compared to DP2f	95% CI d	p-value	ORb compared to DP3f	95% CI d	p-value	ORb compared to DP4f	95% CI d	p-value
DP1f	382 (20.8)	0.3	0.1, 0.5	<0.0001	1.8	1.0, 3.4	0.05	0.1	0.0, 0.2	<0.0001
DP2f	586 (31.9)				6.8	4.3, 10.9	<0.0001	0.4	0.2, 0.8	0.002
DP3f	591 (32.2)							0.1	0.0, 0.1	<0.0001
DP4f	277 (15.1)

^aModels were adjusted for survey year, age group, sex, race/ethnicity, poverty income ratio, energy misreporting, and physical activity count/day.

^bOR represents odds ratio of obesity to normal weight between two compared clusters.

^cD1, D2, D3, D4: Cluster 1, 2, 3, and 4 of temporal dietary pattern, respectively.

^dCI: Confidence Interval

^eP1, P2, P3, P4: Cluster 1, 2, 3, and 4 of temporal physical activity pattern, respectively.

^fDP1, DP2, DP3, DP4: Cluster 1, 2, 3, and 4 of temporal dietary and physical activity pattern, respectively.

The TDPs also had significant associations with T2DM and MetS (Supplementary Tables 6). TPAPs had significant associations with triglycerides, total cholesterol, and mean systolic blood pressure (Supplementary Tables 7) while the TDPAPs had significant associations with T2DM (Supplementary Tables 8).

Comparison of the Strength of Association between Clusters TDPs, TPAPs, and TDPAPs and Health and Disease Status Indicators

TDPAPs had the most significant differences among clusters and health and disease status indicators (21 significant associations) compared with TDPs (19 significant associations) and TPAPs (8 significant associations) (Table 3-5, Supplementary Tables 6-8). TDPAPs showed 6 significant associations with BMI, WC, and obesity, while TDPs showed 5 significant associations with BMI, WC, and obesity, and TPAPs showed 2 associations with BMI, and one with WC and obesity. However, the TDPs had the highest adjusted R² in the models of BMI and WC compared with adjusted R² in TDPAPs and TPAPs (For BMI: adjusted R²=0.20, 0.16, 0.10, respectively; and for WC: adjusted R²=0.27, 0.23, 0.17; respectively) (Table 9), which indicated TDPs fit the linear regression model of BMI and WC best compared to TDPAPs and TPAPs. TDPs also had the lowest AIC in the models of obesity compared with TDPAPs and TPAPs (AIC= 227,940,072; 230,737,435; 238,051,988; respectively) (Table 9), which indicated that TDPs fit the logistic regression model of obesity best compared to TDPAPs and TPAPs.

Table 9.

Number of significant differences, adjusted R² for body mass index and waist circumference models, and Akaike information criterion for odds of being obese models in temporal dietary patterns, temporal physical activity patterns, and temporal dietary and physical activity patterns of U.S. adults 20-65 years as drawn from the National Health and Nutrition Examination Survey, 2003-2006 (n=1836).

	Temporal Dietary Patterns a		Temporal Physical Activity Patterns a		Temporal Dietary and Physical Activity Patterns a
	# of significant differences	Adjusted R2	# of significant differences	Adjusted R2	# of significant differences	Adjusted R2
Body mass index	5	0.20	2	0.10	6	0.16
Waist circumference	5	0.27	1	0.17	6	0.23
	# of significant differences	AICb	# of significant differences	AICb	# of significant differences	AICb
Odds of obesity	5	227,940,072	1	238,051,988	6	230,737,435

^aModels were adjusted for survey year, age group, sex, race/ethnicity, poverty income ratio, energy misreporting, and physical activity count/day.

^bAIC: Akaike information criterion

DISCUSSION

This study objectively generated and compared TDPs, TPAPs, and TDPAPs emphasizing the timing of dietary intake and PA and their link to health and disease status indicators based on previous studies ^27-29. All pattern clusters, TDPs, TPAPs, and TDPAPs, exhibited significant associations with health and disease status indicators. Yet, TDPAPs and TDPs had more numerous significant differences and stronger relationships among all health and disease status indicators and also separately for obesity-related indicators: BMI, WC, and obesity models, compared with TPAPs. The R² values for all models were relatively small while the AIC values were large, likely because of several additional influential factors such as sleep patterns, environmental exposures, and genetics^85,88 that also play a role in the respective health and disease status indicators but were not included in the models due to lack of data. Furthermore, there may be other aspects of diet such as the dietary quality and micronutrients consumed, and aspects of PA such as the type of PA, that are important to health, but were not included in this analysis, potentially contributing to the low R² and large AIC. This study represents a starting point for future investigations where the multidimensionality of multiple lifestyle behavior patterns along with environment and genetics may be incorporated into patterns to determine relationships with health and where various components and combinations of the patterns relationships to health and disease status indicators may be evaluated for their strength and overall utility ^85,88.

Both TDPAPs and TDPs had more numerous significant differences and stronger relationships with health and disease status indicators compared with TPAPs. The TDP cluster with optimal links to health and disease status indicators exhibited a pattern of evenly spaced (7:00-10:00, 11:00-14:00, and 17:00-21:00), energy balanced eating occasions throughout the day and had significantly lower BMI, WC, and odds of being obese compared to the other TDP clusters. Similarly, the TDPAP cluster with proportionally equivalent energy consumed at three main eating occasions (7:00-10:00, 11:00- 14:00, and 17:00-21:00) and the lowest average energy intake with the lowest PA counts among 4 clusters from 10:00-18:00, was significantly associated with lower BMI, WC, and odds of obesity compared to the other three TDPAP clusters. Both TDPs and TDPAPs showed that participants with proportionally equivalent energy consumed at three main eating occasions were more healthful compared to their counterparts who had one energy intake peak at different times (either 11:00-14:00 or 17:00-21:00) throughout the day. In terms of PA, the TDPAP cluster with the lowest PA counts from 10:00-18:00 among the 4 TDPAPs clusters was associated with lower obesity-related indicators (Figure 2c and 2d) potentially due to low energy intake of participants, which suggests that the benefits of low energy intake make up for the disadvantages of low PA using the patterning methods described here. The heavy influence of diet in the patterns may also help to explain the stronger and more numerous associations between obesity-related indicators and TDPs compared with TPAPs. Interestingly, when diet was removed from the patterns, the TPAP cluster with the highest PA counts from 6:00-18:00 was significantly associated with lower BMI compared to the TPAP cluster with the lowest PA counts from 7:00-18:00. This suggests that PA counts are important to the relationship with health status indicators when only PA was considered, but when energy and PA counts were both taken into account, the temporal lifestyle behavior patterns exhibited a more complicated association with health and disease status indicators where energy intake was influential.

TDPs also had the higher R² and lower AIC compared to TPAPs, which indicated that TDPs had stronger associations with BMI, WC, and odds of being obese compared to TPAPs. Previous studies also supported that energy restriction plays a major role in weight loss compared with increased PA ^89,90. However, the results should be interpreted with caution because the scale of the dietary (0-4000 kcal) and PA (0-1.2×10⁵ counts per hour) metrics are different, making a direct and equitable comparison of the behavior roles with obesity challenging.

Few studies have evaluated the relationship of multiple behaviors compared with a singular behavior to health and disease status indicators, and none have considered the timing and frequency of those behaviors. The lack of developed methods to combine behaviors ^91-94 have been a limitation to investigate multiple aspects of diet and PA together in relationship to health. However, evidence of the combined impact of diet and PA has been considered in previous interventions ^91-94. A systematic review and meta-analysis summarized that a combined dietary and PA intervention resulted in greater average weight loss and maintaining a higher mean weight loss compared to an intervention of diet or PA alone after the same period of time ⁹⁵. Other studies also showed the importance of a combined effect of multiple lifestyle behaviors on health: Goldstein et al. ⁹⁶ summarized that multiple behavioral interventions including diet, PA, smoking, and alcohol use hold promise for secondary care prevention where health professionals promote preventive measures that lead to early diagnosis and prompt treatment of a disease, illness, or injury ^97-99. Policy and practice-oriented organizations have also promoted both diet and PA changes to improve health. For example, the American Heart Association/the American College of Cardiology/The Obesity Society guideline for the management of overweight and obesity in adults stated that “The principal components of an effective high-intensity, on-site comprehensive lifestyle intervention in facilitating weight loss and maintenance of lost weight include 1) prescription of a moderately reduced-calorie diet, 2) a program of increased physical activity, and 3) the use of behavioral strategies to facilitate adherence to diet and activity recommendations”.⁴ The latest version of the Dietary Guidelines for Americans 2020-2025 recommends that making healthful changes to dietary patterns and increasing PA will improve health and prevent additional weight gain and/or promote weight loss¹, and also recognizes the need for more studies to address multiple aspects of diet in relationship to health ¹⁰⁰. A prior systematic review ¹⁰evaluated whether and how time and dietary and PA behaviors interacted to influence health. Findings showed that those with PA after eating had beneficial impacts to postprandial glycemia compared with those who exercised before a meal, indicating that the timing of behaviors with respect to each other may impact health. The methods to create TDPAPs in a prior study were designed to account for the sequence of diet and PA over time in a day.²⁹ The results showed that combining these different behaviors temporally is possible and importantly, that pattern selection can be guided based on internal and external criteria, and visualization. The current study builds on the previous study²⁹, applying a similar methodology to create TDPAPs ²⁹and then using the same bandwidths to generate TDPs and TPAPs, also building from development of the single-behavior patterns ^27,28. Each of these studies^27-29 provided concurrent validity of the TDP, TPAP, and TDPAP by evaluating the relationship of the pattern clusters with a variety of health and disease status indicators, similar to the evaluation of the relationships discovered in this current study. Here, even though TDPAPs had more significant associations with health and disease status indicators, higher R², and lower AIC compared to TPAPs, the TDPAPs and TDPs showed similar relationship with obesity-related indicators: TDPAPs had one more significant association with BMI, WC, and odds of being obese, but slightly lower R² and higher AIC compared to TDPs.

Lifestyle behaviors like diet and PA not only co-exist but also influence each other ^101-105. Engaging in healthful dietary and PA behaviors may also be more commonly practiced among those who are already concerned about health. Results of a health communications intervention showed that physically active participants were more likely to change dietary behavior like eating more vegetables and fruits compared to other participants ¹⁰⁶. The combination of lifestyle behaviors in an intervention or in health promotion may also bring about extra benefits including improved self-efficacy ¹⁰⁷ and reduced health care costs ^107,108. Based on the temporal lifestyle behavior patterns generated in this study, certain temporal lifestyle behavior patterns have a more desirable relationship with health and disease status indicators than others. In addition, the importance of the timing and practice of multiple behaviors holds promise for obesity prevention. Therefore, interventions considering multiple lifestyle behaviors in relationship to health are worth considering because these behaviors exist together in life, influence each other, may synergistically be linked to health and disease, and have relevance to behavioral interventions, dietary and PA recommendations, and public health policies.

Strengths of this study include objective PA data, 24-hour dietary recall and reported times of eating as a basis for patterns that avoid pre-conceived ideas about how, what and when participants engage in behaviors. In addition, this study compared the associations between three temporal lifestyle behavior patterns and health and disease status indicators by using similar models, covariates, and bandwidths to generate three temporal lifestyle behavior patterns and facilitate a comparison holding constant these controllable aspects. Moreover, since the temporal patterns integrating dietary behavior were significantly associated with the odds of obesity, with validation of evidence from stronger future study designs, temporal patterns representing the integration of daily dietary habits could potentially predict obesity.

Limitations of this study include the small R² and large AIC in each respective model, suggesting there are many unaccounted factors in the models for predicting health such as environment, genetic characteristics, and other lifestyle behaviors ¹⁰⁹. Lifestyle behaviors such as sleep and additional multidimensionality of considered lifestyle behaviors such as type of PA and diet quality should be integrated in temporal lifestyle behavior patterns in future studies to explain more of the variation in the health and disease status indicators modeled. However, the small R² or large AIC did not discount the significant associations between temporal patterns and health and disease status indicators shown in the results. The cross-sectional nature of this study indicates association and not causation of weekday TDPAP with health and disease status indicators. Current temporal patterns were generated from weekday dietary recall and PA accelerometer data, in future studies, weekend dietary recall and PA should also be considered in the development of temporal patterns because those behaviors may not be the same as weekday temporal patterns ^43,110. Lastly, the PA monitors did not record water-related activities such as swimming and bathing and 10-hours wear time a day was considered as a valid wear day, which may not fully represent all of the physical activities of participants for the 24-hour day.

CONCLUSION

In conclusion, the TDPAPs and TDPs had more significant associations with indicators of obesity compared with TPAPs. Patterns representing the integration of daily behavioral habits hold promise for early detection of obesity.

Research Snapshot.

Research Question:

Which temporal lifestyle behavior patterns have stronger and more numerous relationships with health and disease status indicators?

Key Findings:

In this cross-sectional study that included 1,836 U.S. adults from the National Health and Nutrition Examination Survey (years 2003-2006), temporal dietary and physical activity patterns and temporal dietary patterns had stronger and more numerous significant associations with health and disease status indicators compared with temporal physical activity patterns.