Analytic Methods for Understanding the Temporal Patterning of Dietary and 24-H Movement Behaviors: A Scoping Review

Dietary and movement behaviors [physical activity (PA), sedentary behavior (SED), and sleep] occur throughout a 24-h day and involve multiple contexts. Understanding the temporal patterning of these 24-h behaviors and their contextual determinants is key to determining their combined effect on health. A scoping review was conducted to identify novel analytic methods for determining temporal behavior patterns and their contextual correlates. We searched Embase, ProQuest, and EBSCOhost databases in July 2022 to identify studies published between 1997 and 2022 on temporal patterns and their contextual correlates (e.g., locational, social, environmental, personal). We included 14 studies after title and abstract (n = 33,292) and full-text (n = 135) screening, of which 11 were published after 2018. Most studies (n = 4 in adults; n = 5 in children and adolescents), examined waking behavior patterns (i.e., both PA and SED) of which 3 also included sleep and 6 included contextual correlates. PA and diet were examined together in only 1 study of adults. Contextual correlates of dietary, PA, and sleep temporal behavior patterns were also examined. Machine learning with various clustering algorithms and model-based clustering techniques were most used to determine 24-h temporal behavior patterns. Although the included studies used a diverse range of methods, behavioral variables, and assessment periods, results showed that temporal patterns characterized by high SED and low PA were linked to poorer health outcomes, than those with low SED and high PA. This review identified temporal behavior patterns, and their contextual correlates, which were associated with adiposity and cardiometabolic disease risk, suggesting these methods hold promise for the discovery of holistic lifestyle exposures important to health. Standardized reporting of methods and patterns and multidisciplinary collaboration among nutrition, PA, and sleep researchers; statisticians; and computer scientists were identified as key pathways to advance future research on temporal behavior patterns in relation to health.


Introduction
Worldwide, the prevalence of suboptimal diet and insufficient physical activity (PA) remains unacceptably high [1,2].Sedentary behavior (SED) and insufficient sleep also characterize modern lifestyles and may further contribute to ill health [3,4].Adoption of healthy behaviors, including evidence-based recommendations for dietary and 24-h movement behaviors, which includes waking (i.e., PA and SED) and sleep behaviors, would confer substantial health and economic benefits such as a reduction in the global burden of noncommunicable diseases, lower health care costs, and higher workforce productivity [1,[3][4][5][6].
Dietary and 24-h movement behaviors occur at different times, frequencies, and sequences over time, herein referred to as temporal behavior patterns [7][8][9].Throughout a 24-h day, people consume a variety of foods at eating occasions (i.e., meals and snacks), accumulate their time in daily PA and SED by moving at varying intensities, and go to sleep and wake up.Across the life course, levels of these same behaviors may also fluctuate and change, especially during periods of major life transitions [10][11][12][13][14], and the health effects of long-term exposure to unhealthy lifestyles are well documented [15][16][17][18].However, within a 24-h day, the accumulation in the timing, frequency, and sequencing of dietary and movement behaviors may have implications on health via the body's circadian system.The body's central clock governs circadian rhythms (24-h cycles) and works with peripheral clocks located in body organs to modulate physiologic and metabolic processes [19,20], which are responsive to external cues including light/dark exposure, PA, and eating/fasting [19].Circadian misalignment of these cues impairs glycemic control, insulin responses, and blood pressure, and increasing evidence suggests that shift work, night eating, chronic social jet lag, sleep deprivation, and, conversely, oversleeping increase risk of obesity and cardiometabolic disorders [19,20].
Clustering describes how dietary, 24-h movement, and other behaviors occur in tandem and is a growing area of research [21][22][23][24][25][26][27].Clustering captures population differences in behavior pattern combinations and interactions that may be obscured if levels of behaviors for a population are examined with disregard of individual differences [28,29].For example, among youth, dietary and 24-h movement behaviors cluster in healthy, unhealthy, and mixed ways [22,23,25].Although mixed patterns (e.g., high PA and high SED) are most prevalent [23,25], unhealthy patterns (e.g., high SED, high intake of energy-dense foods) have been linked to higher adiposity [23,25].To date, these cluster patterns have mostly been based on daily/weekly averages of behaviors; temporal data have rarely been incorporated into patterns.An understanding of the when, frequency, duration, and type of behaviors may inform the design of precision health interventions.For example, some people may eat 6 times a day, consume most of their energy intake after 17:00 and engage in prolonged SED in the evening, whereas others may have evenly distributed energy intake across 3 mealtimes, be more physically active in the evening, and sleep 7 h a day.The temporal sequencing of behavior state transitions (i.e., going from active to sedentary, fasting to eating, awake to sleep) and behavior modalities [e.g., reading compared with watching television (TV), quality of sleep] also characterize daily temporal patterns and could be considered in the design of precision health interventions [30].
Dietary and 24-h movement behaviors are also influenced by a broad range of intrapersonal, social, cultural, economic, and political contextual factors that vary over time [31][32][33].Examples include location, social setting, political/organizational setting, mood, motivational level, and concurrent activities (e.g., running while listening to music, eating while watching TV).Further, contextual factors may interact with dietary and 24-h movement behaviors to shape health.For example, emerging research suggests that people who undertake PA in greenspaces or share more meals with family have a lower risk of poor mental health than those who do not [34,35].Understanding the accumulation of dietary and movement behaviors over time and their contextual determinants is key to identifying opportunities for intervention to shift these behaviors toward recommended levels [36,37].
Recent advancement of smart phone and wearable technology such as food diaries, activity applications, smart watches, accelerometers, and wearable cameras have allowed researchers to collect large amounts of time-stamped data on multiple dietary and 24-h movement behaviors and their accompanying contextual factors [33,38].These data facilitate the examination of temporal patterns but require advancements in analytic methods.Although a range of methods exist for analyzing the clustering of behaviors [25][26][27], they have mostly been applied to aggregated data whereby the temporal information about behaviors is lost.Broadly, these methods aim to uncover underlying patterns in the data by reducing high-dimensional data (i.e., number of observation features/covariates is larger than the number of subjects) into a smaller set of variables in a way that retains the variance in the original data [28,[39][40][41].Identification of methods suitable for analyzing temporal behavior patterns is a critical first step to advance understanding of complex lifestyles in relation to health.Therefore, the aim of this scoping review was to explore analytic methods for understanding the temporal patterning of multiple dietary and 24-h movement behaviors.Additionally, we explored methods that have incorporated contextual factors in the examination of temporal behavior patterns.

Methods
This scoping review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines [42].The protocol for this review can be accessed via the Open Science Framework at https://osf.io/7fg98/[43].

Search strategy
An initial literature search was conducted on 3 May, 2021, in the Embase, ProQuest, and EBSCOhost (PsycINFO, MEDLINE Complete, SportDiscus) electronic databases.The search strategy involved consultation with the institute librarian to ensure maximum coverage of key platforms such as PubMed and included peer-reviewed human studies published in English between 1 January, 1997, and 30 March, 2021.The year 1997 was chosen as the search start date because studies using real-time technology (e.g., accelerometers, electronic food diaries) to assess diet and 24-h movement behaviors became more prominent in the late 1990s and early 2000s [38,44].Keywords used for the literature search covered 4 main concepts, including dietary and movement behaviors, behavior context, temporal patterning, and novel analytic methods (Supplemental Table 1).Once the key terms were identified and refined, searches were completed for each of the databases and, where necessary, modified to accommodate the individual databases.An updated literature search using the same databases, search, and screening strategies was conducted in July 2022.

Eligibility criteria and article selection procedures
Supplemental Table 2 provides a summary of the study eligibility criteria.This review included peer-reviewed cross-sectional, prospective cohort, and intervention studies conducted in children and/or adults.To identify methods for determining temporal patterns based on combinations of behaviors, studies were required to have temporal information (i.e., time series data) on !2 of the 4 behaviors (diet, sleep, SED, PA).To identify methods for incorporating contextual factors into temporal behavior patterns, temporal information on !1 behavior and accompanying contextual information about that behavior (e.g., dietary intake and location of eating) was required.Articles were included if they used multivariate methods that analyzed the temporal behavior and context data as an integrated pattern.For example, studies were ineligible if they examined individual behaviors separately, across different models, or analyzed data in a way whereby the timing of the behavior was not incorporated.Articles were excluded if they did not involve humans, were not in English, were reviews, qualitative studies, or conference abstracts, focused on a single 24-h movement or dietary behavior only (i.e., did not include multiple behaviors or additional contextual factors).
On completion of the literature search, duplicates were removed and 2 reviewers (SEC and RML) screened all titles and abstracts of the remaining identified articles in Covidence [45].Following this, full-text screening was also performed independently by SEC and RML.For approximately a third of the full-texts, uncertainties about their inclusion were resolved through discussion of the inclusion and exclusion criteria between reviewers (Supplemental Table 3).The reference lists of the included full-text articles and recent reviews on behavior patterns [7,22,46] were also searched for relevant articles.

Data extraction and synthesis of results
Data extraction and entry of items into an Excel spreadsheet were conducted independently by 2 members of the review team (SEC and RML).First, the following information on study characteristics were extracted and included: author, year, study design and duration, study population (sample size, recruitment location, age, sex, health status), behaviors assessed, measurement of behaviors (assessment tools and duration, methods to classify levels of behaviors), contextual factors assessed, and measurement of contextual factors.Second, the following information on the methods used to assess temporal patterning were extracted: analysis aim, data structure (units, clustered/ nested, time segmentation), analytic sample size, analysis input variables (number, time segmentation, units, type of variables), details of the analytic method, software used, measures of model robustness, findings on the number and type of temporal patterns and any associations with health outcomes where reported, data visualization methods used to present findings, and observations on the strengths and limitations of the analysis methods.Following data extraction, the study findings were summarized and narratively described.

Search results
Supplemental Figure 1 shows a PRISMA flow diagram of the study selection process.The database searching and citation searching identified 32,607 (initial search), 6485 (updated search), and 3 (citation search) articles.After removing duplicates, 27,199 (initial search) and 6093 (updated search) article titles and abstracts were subsequently screened, and of these, 135 were assessed based on their fulltext for eligibility.Following screening, 14 studies were included in this review.

Study characteristics
Of the 14 included studies, 4 examined temporal patterns based on 2 or more behaviors, without contextual factors (Table 1 [47][48][49][50][51]).Study years ranged from 2017 to 2022 and were conducted in Europe [47,51], the United States of America [49], New Zealand, and 10 countries spanning Europe, North America, South America, and Australia [50].All but 1 longitudinal analyses of a prospective cohort study [47] were cross-sectional.One study included children and adolescents from the Gateshead Millenium Study [47], whereas the 3 studies included adults only, of which 1 included older adult patients with chronic obstructive pulmonary disease [50].All studies included males and females, with the percentage of male participants ranging from 42% to 65%.Three of the 4 studies examined multiple waking behaviors (e.g., PA, including different PA intensities, and SED), whereas only 1 study focused on both dietary and waking behaviors [49].
PA and SED assessment tools included accelerometers: Acti-Graph [47,49], Polar Active [51], and SenseWear armband/mini armbands [50].Device wear time mostly ranged from 3 d, including the weekend [47], to 14 d [51].Lin et al. [49] used a random valid weekday, defined as >10 h of wear time, from ActiGraph data collected over 7 consecutive days.The cut points used to classify PA intensities in adults varied but were mostly <2.0 metabolic equivalent of tasks (i.e., an estimate of the ratio of a person's energy expenditure rate during a specific PA, relative to energy expenditure while resting) used by the body during PA for classifying SED [50,51].Evenson cut points were used to classify both PA intensities and SED in the study of children [47].In the study by Lin et al. [49] that examined dietary behaviors with PA, energy intake at eating occasions defined as an eating window of 15 min was assessed using a 24-h weekday dietary recall.

Analytic methods used across the included studies
Tables 3 and 4 presents summarized findings from individual studies examining temporal behavior pattern (without and with contextual factors, respectively) and their analytic approaches.Supplemental Table 3 provides a more detailed description of the Abbreviations: BMI, body mass index; C/S, cross-sectional; diet, dietary behavior; EO, eating occasion; GIS, geographic information system; LPA, light physical activity; MET, metabolic equivalent; MVPA, moderate-to-vigorous intensity physical activity; NR, not reported; P, prospective; PA, physical activity; SED, sedentary behavior. 1For prospective studies, the baseline and follow-up years of data collection are shown. 2Duration of eating was not recorded, so all EOs were standardized to be 15-min long.analytic approaches used in each study.Of the 4 studies examining multiple behaviors only (i.e., PA and SED; diet and PA), 1 study used group-based multitrajectory models to examine longterm temporal patterns (i.e., yearly patterns) [47] whereas 3 studies used 1 or more machine learning approaches to examine daily temporal patterns [49][50][51].For example, a combination of distance-based clustering algorithms was applied to multidimensional accelerometer and dietary time series data by Lin et al. [49] and principal component analysis (PCA) and k-means cluster analysis were applied to 180 continuous waking behavior variables by Mesquita et al. [50].
Ten studies examined dietary and/or 24-h movement behaviors together with contextual factors (Table 4 [48,[52][53][54][55][56][57][58][59][60]).Vidal Bustamente et al. [55] and Yang et al. [57] used model-based clustering techniques (e.g., latent profile analysis) to determine sleep and PA patterns with contextual factors, respectively.In the study by Yang et al. [57], model-based clustering was used in conjunction with linear mixed-effect models to simultaneously determine cluster patterns and their predictor variables as part of a novel 1-step clustering procedure [57].For studies examining multiple 24-h movement behaviors with contextual factors (i.e., PA and SED; PA, SED, and sleep), 2 used a type of model-based clustering for longitudinal data (i.e., data with 3 time points or more) to determine joint temporal patterns for multiple behaviors over longer durations [48,56].
Most of the included studies on PA and SED used a combination of approaches to determine daily temporal patterns and their contextual correlates [52][53][54].For example, Farrahi et al. [54] used 2 clustering approaches to identify the temporal patterns of PA intensities and SED in relation to cardiovascular disease [51] before classifying the patterns according to 168 different contextual factors [54].Wang et al. [52] used latent growth-curve models to examine temporal growth trends in PA patterns first derived from PCA in relation to changing contextual factors over 637 d.
Two studies on children used compositional data analysis and data-driven visualization methods, specifically ring maps and time-activity diagrams [53], to understand compositional 24-h movement data and facilitate comparisons between groups and contexts.Of the 2 studies examining dietary behaviors only, 1 applied canonical correspondence analysis to food group and contextual variables to examine how eating is structured over time and space [59].The other study used a clustering algorithm to identify and characterize Parisian meal patterns from detailed interview data on mealtimes, meal frequency, and meal contextual factors [60].

Temporal behavior patterns identified across the studies
The number of temporal patterns identified across the 40 studies on multiple behaviors without contextual factors ranged between 3 [47] and 5 [50] (Table 3 [47,[49][50][51]).Most of these studies reported statistical measures used to determine the number of patterns [47,49,50] (Supplemental Table 3).Using PA and SED behaviors, Niemel€ a et al. [51] found 4 patterns labeled inactive, evening active, moderately active, and very active--mostly during day, with the inactive cluster having the highest risk of cardiovascular disease in the next 10 y, than the very active cluster [51].Mesquita et al. [50] identified 5 patterns in older patients with chronic obstructive pulmonary disease, 4 of which were characterized by varying degrees of SED [50].Of note, across all patterns, PA intensity was the highest before midday, however, "couch potatoes" had the poorest health risk profile than other patterns.Farooq et al. [47] found 3 patterns of joint PA and SED trajectories characterized by being inactive [i.e., <60 min moderate-to-vigorous intensity physical activity (MVPA) and 400-600 min SED per day] during childhood and adolescence, active during childhood only, and active during childhood and adolescence.Fat mass index was the highest in the inactive group at age 15 [47].
Four temporal patterns were identified in the 1 study examining dietary and waking behaviors [49].The first 2 patterns comprised more women and were characterized by 1) low PA and low variation in energy content across the main eating occasions (i.e., no distinct peaks of energy intake) and 2) an evening PA pattern also with low variation in energy content across main eating occasions.The remaining 2 patterns comprised more men, the first of which included an early morning PA pattern with the highest overall energy intake disperse across the noon and evening periods.The second was a high overall PA pattern (only 14% of participants) with 3 evenly spaced eating occasions that were also higher in energy intake in the afternoon and evening.Adults with the high overall PA pattern or the evening PA pattern had lower BMI (in kg/m 2 ) and waist circumference than those with the low PA pattern or the early morning PA pattern, with the most favorable estimates observed for the high overall PA pattern.Patterns were visually represented using heat maps of the distribution of energy-containing eating occasions and PA counts greater than zero [49].
The number of temporal patterns identified across the 10 studies on dietary and/or 24-h movement behaviors with contextual factors ranged between 3 [48,55,57] and 54 [54].Seven of these studies reported the statistical indicators or metrics used to determine the number of patterns [48,52,[55][56][57], however, the actual values for the statistical indicators by pattern number were rarely reported [52,56].
Among the studies on 24-h movement behaviors, Vidal Bustamente et al. [55] identified 3 sleep patterns linked to daily contextual factors and were characterized by the following: 1) high sleep quality, low stress, low negative affect, and higher subjective energy and PA levels; 2) highest sleep duration, regularity and quality, moderate stress, and lowest subjective PA levels; and 3) poor sleep quality, high stress, and high negative affect.The latter pattern was associated with higher Global Clinical Severity Scores for psychologic distress [55].Yang et al. [57] identified 3 groups of young adults who had consistently low MVPA between the ages of 14 and 23 y (maintainers), high MVPA at age 14 y but decreased over time (decreasers), or low MVPA at age 14 y but increased over time (increasers), and in each group, higher MVPA was predicted by unique contextual factors [57].For example, in the maintainers group, MVPA was predicted by friend support and more neighborhood parks, but self-efficacy and self-management strategies were important predictors of higher MVPA in the decreasers group.
In the study by Farooq et al. [48], having a safe playing environment and actively commuting to school during childhood and sports club participation in adolescence predicted the "active" trajectory pattern (>60 min MVPA and <400-600 min SED per day) during childhood and adolescence [48].In a similar study of joint SED and PA trajectories in children and Farrahi et al. [54] identified 54 subgroups of participants with active and inactive temporal patterns-originally identified in the study by Niemel€ a et al. [51]-which differed by their waking behavior accumulation patterns [54].Contextual factors that helped predict these subgroup patterns included weekday computer use time, sitting time at the workplace, number of public transport stops, and housing density [54].
Wang et al. [52] identified 4 key contextual factors that positively influenced college freshmen's daily PA patterns and included sleep, peer influence, social activity, being on campus, and temperature [52].Sleep also influenced daily 24-h movement behavior patterns in the study by Zhao et al. [53]; children from the most deprived neighborhood areas slept the least, went to sleep later, and were more sedentary in the morning school hours, than their counterparts in the least deprived areas.This pattern of less sleep and higher SED was also more prevalent in overweight or obese boys from deprived areas.G aba et al. [58] also reported sex-specific 24-h movement patterns.Although both boys and girls spent approximately two thirds of their school and out-of-school time in SED, boys spent more school time in vigorous-intensity PA and less time in out-of-school light-intensity PA, than girls [58].In girls only, replacing 30 min/d of out-of-school SED was associated with 14% lower fat mass index in compositional isotemporal substitution models [58].
Two [59] and 5 [60] temporal patterns were observed in the 2 studies of dietary behaviors with contextual factors.Both studies used objective statistical indicators to guide the selection of patters.Jaeger et al. [59] found 2 eating patterns using canonical correspondence analysis, which were structured by time of the day (e.g., bakery pastries and cereals more evident in the morning) and indicators of formality (e.g., sitting, social setting).For example, meals involving sitting with others usually included hot carbohydrates and proteins and took place in the houses of family/friends.In contrast, "casual eating" tended to involve sweets, chocolates, or sandwiches and take place at the office, while working on the computer or other locations.Five distinct meal patterns were identified by Riou et al. [60], of which 3 were characterized by 3 meals per day within the conventional time periods for French adults but differed by contextual factors such as where (home, outside) and with whom (family, alone) the meal was eaten.The 2 remaining patterns were characterized by the omission of breakfast (i.e., a 2 meal per day pattern) and meals were consumed mostly at home with the TV or outside often, respectively.

Discussion
We conducted a comprehensive scoping review to explore analytic methods used by researchers to examine dietary and 24h movement behaviors as an integrated temporal pattern, and their contextual correlates, where presented.Fourteen studies were included in this review, and of these, 11 were published in the last 5 y, highlighting that this is a rapidly developing area of research.Only 1 study [49] combined temporal data on both waking and dietary behaviors, despite their roles in health and well-being and the evidence on the patterning of these behaviors in children and adults [23,25,61,62].Clustering methods were mostly used to determine temporal patterns.Although clustering methods aim to allocate observations into distinct groups, the data-driven nature of these methods and the types of specific patterns that are examined mean that patterns will vary by the study population and research aim (i.e., patterns are not generalizable).Studies also varied in the type of clustering used; some studies used probabilistic clustering methods (e.g., latent profile analysis, group-based trajectory analysis), while others used machine learning techniques with various clustering algorithms used across those studies.Further, some studies used a combination of data reduction methods.Together, these findings highlight the complexity involved with combining and analyzing time-series data for multiple behaviors and their contextual factors.Progressing analytic techniques for examining the temporal patterning of dietary [7] and 24-h movement behaviors [63], and the contexts in which they occur, may help us better understand health-promoting lifestyles and, subsequently, inform healthy eating and 24-h movement interventions.
The heterogeneity in the methods used to determine patterns and the behavioral and contextual input variables make the findings difficult to compare across studies.However, despite the diversity in types and range in number of patterns found, there were still several observations of note when comparing the included studies that examined waking behaviors.For example, temporal patterns characterized by high SED and low PA were associated with indicators of higher adiposity in children and adolescents [47,53,58], and higher adiposity [50,54] and risk of cardiovascular disease [51] in adults.Although the studies differed in the periods examined-for example, 2-3 yearly [47] compared with daily/weekly [51,53,54,58] temporal patterns, the findings are in line with previous literature that show patterns characterized by high SED are associated with higher adiposity in youth [23,25,26].Intervention studies in at-risk adults have also shown favorable changes in metabolic risk factors when breaking up prolonged SED time at a desk with regular brief bouts of PA [64,65].However, the examination of temporal and contextual aspects of patterns allows us to characterize useful information about when, where, and how behaviors occur.For instance, both G aba et al. [58] and Zhao et al. [53] identified after school and evening hours as critical windows for reducing SED and increasing PA in boys [53] and girls [58].In the study by Zhao et al. [53], weekday mornings were also    Abbreviations: ANOVA, analysis of variance; BMI, body mass index; CHAID, chi2 automatic interaction detection; FMI, fat mass index; LPA, light physical activity; MET, metabolic equivalent; MVPA, moderate-to-vigorous intensity physical activity; NA, not assessed; NR, not reported; OW/OB, overweight or obesity; PA, physical activity; SED, sedentary behavior; TV, television. 1 A valid day constituted !1200 min of a total wear time with !720 min of awake time, !320 min of sleep time, and 220-440 min of school time.
identified as a critical window for PA intervention in boys with overweight/obesity living in the most deprived areas.For long-term and favorable PA and/or SED trajectories in children and adolescents, friend support or co-participation [56,57] and access to safe play areas or local parks [48,57] appear to be important factors and are supported by previous literature on PA/SED determinants [66,67].These studies highlight that further investigation into temporal 24-h movement patterns on weekdays and weekend days may facilitate a more nuanced and meaningful understanding of how subpopulations use their daily time across different contexts and provide evidence for health-promoting interventions.Moreover, improved understanding of contextual determinants might provide insights into potential barriers and facilitators of behavior change, which are important considerations for the feasibility and effectiveness of behavior change interventions.
Very few studies included in this review examined the temporal patterning of dietary behaviors together with contextual factors (n ¼ 2) or 24-h movement behaviors (n ¼ 1).Although accelerometers collect time-stamped information on multiple 24h movement behaviors across the intensity spectrum, they do not collect contextual data.Dietary assessment methods commonly used in nutrition research include food frequency questionnaires, which do not include information about the timing of eating occasions or their contextual factors.However, more recently, 24-h dietary recalls and food diaries, including some used in some national nutrition surveys, have been modified to collect both time-based or eating occasion-based dietary and contextual data, which enable the examination of temporal patterns [38].The studies by Jaeger et al. [59] and Riou et al. [60] highlighted the role of context in food choices at eating occasions.For example, sandwiches, sweets, and chocolate were more commonly eaten when working on a computer, and fruit was more often eaten at the main meals later in the day in a sample of Spanish adults [59].Furthermore, eating only 1 or 2 meals per day (with breakfast being the most skipped meal) at home with a TV or outside of home was linked to more snacking and lower dietary quality in French adults [60].Neither of these studies examined associations with health outcomes; therefore, this is an area of investigation for future studies on temporal patterning of dietary behaviors.
Only 1 study was identified in this review that examined the temporal patterning of dietary and waking movement behaviors [49].Lin et al. [49] found that adults with a pattern of high PA counts across the day (mean PA was 5.7 Â 10 5 counts/d) and evenly distributed energy intake at main eating occasions during the noon and evening was associated with more favorable cardiometabolic risk parameters, than adults with higher PA intensity in the morning and the high overall energy intakes dispersed at main eating occasions across the noon and evening.This latter pattern may indicate undesirable energy compensatory behaviors in response to the morning PA (i.e., unhealthy food or overeating as a reward for exercise) and highlights the potential impact of combining data on both dietary and 24-h movement behaviors.However, studies using multiple days of data, including weekend days, are required to elucidate the health effects of daily dietary and PA patterns [68].Further, the lack of studies on the temporal patterning of dietary and 24-h movement behaviors may result from the practical challenges in collecting detailed longitudinal data on both dietary and 24-h movement behaviors due to participant burden, or cost, or difficulties in combining the 2 data sources for analyses.Although both dietary and 24-h movement behaviors are constrained to a 24-h day, how their respective data are collected and processed results in time-series data sets on different scales, making them challenging to combine for analysis.For example, accelerometer data have been typically collected in predefined epochs (i.e., movement counts in different intensities are averaged across each minute of the allocated device wear time).In contrast, dietary data from food diaries or 24-h dietary recalls varies within and between individuals by eating start time, eating duration, and eating frequency, with periods of fasting (i.e., times where no intake is recorded).Further, typically only information about eating start time is collected as part of dietary assessment, which does not allow estimation of time spent eating compared with fasting.These issues are illustrated in the study by Lin et al. [49] who used on existing accelerometer and 24-h dietary recall data collected in the NHANES 2003-2006.As there was no information on duration of eating (and is not typically collected in dietary assessment), Lin et al. [49] applied generic data smoothing, assuming a 15-min eating window, to the start of each reported eating occasion to calculate energy intake per minute.This allowed the input data for the clustering analysis to be on the same time scale as the PA data, which were recorded as vertical accelerations counts per minute.Hence, future research will need to consider measurement timescales of both dietary and 24-h movement behaviors to facilitate their joint temporal pattern analysis.Improvements in digital tools and wearable devices to assess dietary and 24-h movement behaviors has increased the feasibility of "real-time" data collection of multiple behaviors in larger scaled studies.These tools and technologies reduce participant burden and are becoming more affordable with their increasing availability.Further, consideration should be given to the infrastructure and interdisciplinary expertise required for collecting, storing, managing, and combining large data sets from different sources in the planning and design of studies on the temporal patterning of dietary and movement behaviors.
The choice of methods used for pattern analysis varied across the included studies but were often justified by the study authors for their suitability to answer the research question and appropriateness for the data type (e.g., continuous, categorical, or mixed) and structure (e.g., cross-sectional, longitudinal, multilevel, and multidimensional).For example, group-based trajectory modeling approaches were suited to examining long-term temporal patterns [47,48,56,57], and clustering data mining techniques (based on different algorithms) were used to examine a large number of input variables and correlates [49,50,51,54].In some cases, >1 technique was required for data reduction purposes [49,50,54].For example, Farrahi et al. [54] had to first simplify the temporal patterns identified by Niemel€ a et al. [51] to make the data suitable for the analysis of correlates using decision trees and identified 54 subgroups of active and inactive adults based on a large range of sociodemographic, behavioral, and contextual factors.The findings from studies using data mining methods may be applied to develop personalized behavioral support using mobile and wearable technology or generate empirical hypotheses for examination in future studies.Nonetheless, the selection of variables (predictors, outcomes, covariates) for inclusion in the analysis and interpretation of the resultant patterns should be guided by evidence-based theory.Further, model-based clustering approaches may provide some advantages over data mining clustering approaches because the selection of the optimal number of patterns is less subjective and is guided by probabilities and goodness-of-fit indicators estimated from statistical models.
The novel clustering method used by Yang et al. [57] simultaneously considered both predictors and MVPA in the model-based clustering procedure and allowed them to identify important contextual predictors of higher MVPA unique to groups with differing MVPA temporal patterns.Such information may be useful for developing interventions tailored to subpopulation's specific needs regarding appropriate social/family support mechanisms or changes to the home, school, or neighborhood environment.Despite the potential utility of these methods for examining the temporal patterning of dietary and 24-h movement behaviors, and their accompanying contextual factors, the types of patterns examined varied in their level of detail (e.g., day-to-day overall time compared with minute-by-minute bouts of time spent in different movement intensities) and the methods and the statistical analyses involved a complex set of steps, were described using field-specific and/or statistical jargon, or were conducted using specialized/custom software packages.Clear documentation of novel methods, in particular the statistical criteria used to determine the patterns, and open-source code analysis tools may facilitate the application and refinement of such methods in future studies [69].Further, development of a framework to standardize the characterization (i.e., terminology, definitions) and reporting of temporal behavior patterns and multidisciplinary collaborations across the fields of nutrition, PA, sleep, statistics, computer science, and computer engineering may further advance this field of research.In fact, a framework for defining and reporting waking behavior patterns has recently been published, with the aim to progress the synthesis of the study findings in this field [9].
As described earlier, the methods used in the included articles were highly technical, and so data visualization methods were often used as an adjunctive tool for conceptualizing the temporal patterns.Dietary and 24-h movement behaviors are compositional in nature, meaning that time in one behavior will affect time in another behavior in each 24-h day [70,71].Hence, visualization strategies may be useful for mapping behaviors by population subgroups across times of the day, geographical spaces, and environments.For example, Zhao et al. [53] presented estimated means for time spent in different 24-h movement behaviors across the day (i.e., behavioral compositions) using novel spatial and data visualization techniques to facilitate understanding of PA, SED, and sleep temporal patterns in youth and the contexts in which they occur.Further, the authors proposed such visualization techniques could add value to studies using compositional data analysis whereby time use differences between groups are presented as isometric log ratio coordinates, which may be challenging to interpret on their own.Jaeger et al. [59] visually mapped categories of foods consumed at eating occasions across time of day and contexts such as location of eating and activities while eating to provide insights into the meal formats (i.e., combination and sequencing of food intake) and the social and cultural practices involved in daily eating routines.However, given the large variation in food choices across eating occasions and contexts, collecting and coding the dietary data to capture this complexity involves substantial participant and researcher burden; integrating more nuanced contextual data with automatically coded 24-h recall data may be a more feasible way to progress research on daily eating routines and their contextual influences.
A strength of this review included a comprehensive search strategy, focusing on methods to analyze the temporal patterning of 4 key health behaviors (diet, PA, SED, and sleep) and their contextual correlates.However, the literature search for this review was last updated in 2022, and we did not include studies of non-English language, gray literature, or conference proceedings.Thus, we expect more articles to have since been published, given this is a rapidly evolving area of research.Of note, only 1 new analytic approach (e.g., compositional data analysis) was identified in the 6 studies included from the updated search.This review therefore provides important insights into the analytic methods available to researchers for examining the temporal patterning of dietary and 24-h movement behaviors.
In conclusion, this scoping review identified a range of methods that have been used to analyze the temporal patterning of dietary and movement behaviors, as well as their contextual correlates.Although the studies included in this review were diverse in terms of their analytic methods, behavior input variables, and periods examined, the evidence to date suggests temporal patterns characterized by high SED and low PA are associated with poorer health outcomes, than those with low SED and high PA.However, further research is needed to elucidate the importance of context in which high SED and PA occurs, considering existing evidence to suggest leisure time PA is more beneficial to health than occupational PA [72].Further, standardizing the reporting of methods used to derive patterns and permission to use and edit open-source code may facilitate the application and continued development of new methods.
The ability of identified methods to incorporate contextual information about the temporal patterns yielded insights into the role of social and environmental factors in promoting temporal behavior patterns.Future research examining existing microlongitudinal data or using study designs with daily assessments of behavioral and contextual factors may help elucidate the health implications of different temporal patterns and their determinants.However, this may require advancing data collection through use of wearables, smartphone, and geographic information system technology and the use of existing data on weather, green spaces and parks, food environments, and local infrastructure.Finally, this review identified only 1 study on the temporal patterning of dietary and 24-h movement behaviors, suggesting further research is needed to develop methods for analyzing dietary behaviors together with 24-h movement behaviors across the intensity spectrum.Overall, progression of the methods and procedures used to collect, integrate, and analyze dietary, movement and contextual data and multidisciplinary collaboration among nutrition, PA and sleep researchers, statisticians, and computer scientists/engineers will be important for advancing our understanding of temporal behavior patterns in relation to health and optimizing future behavioral and public health interventions.

TABLE 1
Characteristics of individual studies examining temporal behavior patterns based on 2 or more behaviors only (i.e., no contextual factors).

TABLE 2
Characteristics of individual studies examining temporal behavior patterns together with contextual factors.

TABLE 3
Findings of individual studies examining temporal behavior patterns based on 2 or more behaviors only (i.e., no contextual factors).

TABLE 4
Findings of individual studies examining temporal behaviors patterns together with contextual factors.