In this population-based cohort of 1179 children 11 to 12 years of age, equivalent benefits to adiposity and HRQoL were associated with different changes (trade-offs) in activities.
Video Abstract
Understanding equivalence of time-use trade-offs may inform tailored lifestyle choices. We explored which time reallocations were associated with equivalent changes in children’s health outcomes.
Participants were from the cross-sectional Child Health CheckPoint Study (N = 1181; 11–12 years; 50% boys) nested within the population-based Longitudinal Study of Australian Children. Outcomes were adiposity (bioelectrical impedance analysis, BMI and waist girth), self-reported health-related quality of life (HRQoL; Pediatric Quality of Life Inventory), and academic achievement (standardized national tests). Participants’ 24-hour time use (sleep, sedentary behavior, light physical activity, and moderate-to-vigorous physical activity [MVPA]) from 8-day 24-hour accelerometry was regressed against outcomes by using compositional log-ratio linear regression models.
Children with lower adiposity and higher HRQoL had more MVPA (both P < .001) and sleep (P = .002; P = .008), and less sedentary time (P = .02; P = .001) and light physical activity (P < .001; P = .04), each relative to remaining activities. Children with better academic achievement had more sedentary time (P = .03) and less light physical activity (P = .006), each relative to remaining activities. A 0.1 standardized decrease in adiposity was associated with either 55 minutes more sleep, 89 minutes less sedentary time, 34 minutes less light physical activity, or 19 minutes more MVPA. A 0.1 standardized increase in HRQoL was associated with either 64 minutes more sleep, 65 minutes less sedentary time, 72 minutes less light physical activity, or 29 minutes more MVPA.
Equivalent differences in outcomes were associated with several time reallocations. On a minute-for-minute basis, MVPA was 2 to 6 times as potent as sleep or sedentary time.
How children allocate their daily time to sleep, sedentary behavior, light physical activity, and moderate-to-vigorous physical activity is associated with a range of health and well-being outcomes. Optimal daily time use may be achieved through various trade-offs between activities.
Equivalent benefits to adiposity and health-related quality of life were associated with differing amounts of change (trade-offs) in accelerometer-measured time use. On a minute-by-minute basis, moderate-to-vigorous physical activity was 2 to 6 times as potent as sleep or sedentary behavior.
Even small improvements in attributes such as adiposity, health-related quality of life (HRQoL), and academic achievement could have individual and population benefits for children now and as adults. The importance of time-use allocation (daily hours spent sleeping, sedentary, and in light and moderate-to-vigorous physical activity [MVPA]) to all of these attributes1–3 is reflected in international optimal daily activity duration guidelines.4–9 For school-aged children, such guidelines typically prescribe a 24-hour day with ∼9 to 11 hours of sleep, at least 1 hour of MVPA, no more than 2 hours recreational screen time, and several hours of light physical activity.
Unfortunately, meeting these simple guidelines is evidently hard: estimates from 6128 children 9 to 11 years of age across 12 countries suggest that only 7% of children meet them.10 Time-use interventions (eg, to increase activity and/or sleep or decrease sedentary time) typically only have small and transient effects.11–13 Rather than unrealistic one-size-fits-all prescriptions, a way forward might be to support children, and those who guide them, to make the decisions that are individually most achievable and effective. However, families and clinicians must first know what trade-offs in the different components of time use might achieve what benefits to important child outcomes.
Because more time spent in any single activity necessarily means less time across all others in the 24-hour day, optimal daily time use may be achieved through various trade-offs between activities. Practically, these might differ according to personal choice, family factors, time constraints (such as school and work schedules), and environmental context, and these trade-offs could also work against each other. If a child needs to get up half an hour earlier to get to sports practice, will the benefits of the extra 30 minutes she spends at the gymnasium be attenuated by her abbreviated sleep?
Furthermore, any given time reallocation may have different impacts on different health outcomes: an extra hour’s sleep may be favorable for quality of life but not academic achievement.14 The same level of body fat may be associated with high MVPA and low sleep or high sleep and low MVPA.
Behavioral epidemiology15 has been reshaped by the analytical paradigm of compositional data analysis.16 Previously, behavioral epidemiology viewed daily activities (eg, physical activity, sleep, sitting) as independent activities, but the compositional paradigm views them as a mix of time-use activities called a composition. The question is not the relationship of any given health outcome with any individual component of time use but with the composition as a whole. An unexplored possibility is that the same health outcome may be achieved by quite different compositions. We can compare how much time change is needed in 2 different activities to have the same effect on a given outcome. Trade-offs can be compared by using a novel graphical interface called “equivalence curves,” created by plotting estimated effects of one activity against that of another activity, for a range of outcomes.
We aimed to describe equivalence curves for selected health and academic outcomes (adiposity, HRQoL, and academic achievement) in a large population-based sample of 11- to 12-year-old children.
Methods
Study Design and Participants
Data were from the population-based Longitudinal Study of Australian Children (LSAC) and its physical health and biomarker module, the Child Health CheckPoint, conducted when children were aged 11 to 12 years. Detailed methodology is reported elsewhere.17 Briefly, in 2004, LSAC recruited a nationally representative birth cohort of 5107 infants aged 3 to 19 months using a 2-step random sampling process from the Medicare database, Australia’s universal health care system. LSAC has since conducted biennial waves of data collection. At wave 6 (age 10–11 years; 2014), 3764 of the original birth cohort participants (74%) were retained in the LSAC study. They were invited to consent to their contact details being shared with the CheckPoint team. Consenting families were contacted, and a total of 1874 families (50% of the wave 6 LSAC sample) participated in CheckPoint.17
A parent or guardian provided informed consent for each child, and The Royal Children’s Hospital (Melbourne) Human Research Ethics Committee (HREC33225) and Australian Institute of Family Studies Ethics Committee (AIFS14-26) approved the study.
Procedure
CheckPoint assessments took place between February 2015 and March 2016. Most families attended 1 of 15 assessment centers across Australia, in which an extensive range of physical and biomarker measures were assessed, whereas 365 (19.5%) children accepted a shorter home visit.
Measures
Details of measurement procedures are provided in Table 1.
Measures Table
Construct . | Measure(s) . | Data Collection, Equipment, and Derivation Procedures . |
---|---|---|
Exposure measure | ||
Time use | Accelerometry | A research assistant fitted a GENEActiv accelerometer (Activinsights Ltd, Kimbolton, Cambridgeshire, United Kingdom) set to collect movement data at 50 Hz. Children were asked to wear the device 24 h/d for 8 d on their nondominant wrist, documenting any removals and sleep and wake times on activity logs. Data were downloaded by using GENEActiv PC software (Activinsights) and converted to 60-s epoch files. Markers indicating self-reported sleep (sleep onset) and wake (sleep offset) times from the activity logs were superimposed onto the daily accelerometer traces and visualized in a MATLAB-based program (Cobra), developed by members of the CheckPoint team to provide an interactive interface to the accelerometer trace. A research assistant manually checked each participant’s daily accelerometry traces and, if required, moved the sleep onset and sleep offset markers to match the trace. Wakeful periods between sleep onset and offset were detected by using the sleep algorithm from van Hees et al.40 Each waking 60-s epoch was classified as sleep, sedentary time, light physical activity, or MVPA by using validated cut points,41 linearly adjusted to accommodate the 50-Hz sampling frequency. For each valid participant (waking wear time ≥10 h, sleep ≥200 min/d, sedentary time ≤1000 min/d, ≥4 valid days), an average weekly composition was created by using a 5:2 weighting for weekdays versus weekend days. |
Outcome measures | ||
Adiposity | Percentage body fat; BMI; waist girth | Percentage body fat was obtained from bioelectrical impedance analysis by using a 2-limb Tanita BC-351 (Tanita, Kewdale, Australia) at home visits and a 4-limb InBody230 (Biospace, Seoul, South Korea) at center visits.42 BMI was calculated from measured weight and height (Invicta I0955 stadiometer; Invicta, Leicester, United Kingdom) in bare feet and light clothing. Waist girth was measured on the skin at the narrowest point between the 10th rib and iliac crest. If narrowing was not detected, the measurement was taken at the midpoint of these 2 landmarks43 by using a Lufkin W606PM steel tape (Apex Tool Group, Sparks, Maryland). Waist girth and BMI were expressed as z scores relative to international norms44,45 and body fat as a sample-specific z score. The 3 variables were considered as separate outcome variables in supplementary analyses, but for the main analyses, they were averaged to form a composite adiposity summary variable. |
HRQoL | PedsQL version 4.0; 8- to 12-y-old child self-report version | The 8- to 12-y-old self-report PedsQL version 4.046 is a 23-item instrument of high internal and external reliability (α coefficients ≥0.7).47 The PedsQL comprises 4 domains: physical, emotional, social, and school functioning. Responses for each item are reverse scored on a 5-point Likert scale (0 = never a problem, 4 = almost always a problem) and linearly transformed to a 0–100 scale (0 = 100, 1 = 75, 2 = 50, 3 = 25, 4 = 0), with higher scores indicating better HRQoL.48 Physical and Psychosocial (a combination of emotional, social and school domains) Health Summary Scores were considered as separate outcome variables in supplementary analyses, whereas the main analyses used the Total Scale Score. |
Academic achievement | NAPLAN | The NAPLAN is the numeracy and literacy test undertaken by all Australian children biennially from years 3 through 9. Year 7 (when children were ∼12–13 y old) scores were used because they are the most proximal measure of academic achievement subsequent to accelerometry, limiting reverse causality. We used the numeracy score and derived a literacy score as the arithmetic mean of reading, writing, spelling, and grammar scores. The overall academic achievement score was the mean of numeracy and literacy. |
Covariates | ||
Age | Age | Age of study child (in months) (obtained from wave 6 parent questionnaires) |
Sex | Sex | Sex of study child (female or male) (obtained from wave 6 parent questionnaires) |
SEP | SEP | LSAC releases a composite z score for family-level SEP with each wave, with higher scores indicating higher SEP. The score is derived from parent-reported household income levels, education levels, and occupation49 (obtained from wave 6 parent questionnaires). |
Pubertal status | PDS | Children used iPads to report sex-specific pubertal signs using the PDS.50 The PDS consists of 5 sex-specific questions related to growth spurts, body hair, and acne; facial hair and voice deepening for boys; and breast growth and menstruation for girls. Responses were summed to obtain a total PDS score, used to categorize children into prepubertal, early pubertal, midpubertal, late pubertal, or postpubertal. |
Construct . | Measure(s) . | Data Collection, Equipment, and Derivation Procedures . |
---|---|---|
Exposure measure | ||
Time use | Accelerometry | A research assistant fitted a GENEActiv accelerometer (Activinsights Ltd, Kimbolton, Cambridgeshire, United Kingdom) set to collect movement data at 50 Hz. Children were asked to wear the device 24 h/d for 8 d on their nondominant wrist, documenting any removals and sleep and wake times on activity logs. Data were downloaded by using GENEActiv PC software (Activinsights) and converted to 60-s epoch files. Markers indicating self-reported sleep (sleep onset) and wake (sleep offset) times from the activity logs were superimposed onto the daily accelerometer traces and visualized in a MATLAB-based program (Cobra), developed by members of the CheckPoint team to provide an interactive interface to the accelerometer trace. A research assistant manually checked each participant’s daily accelerometry traces and, if required, moved the sleep onset and sleep offset markers to match the trace. Wakeful periods between sleep onset and offset were detected by using the sleep algorithm from van Hees et al.40 Each waking 60-s epoch was classified as sleep, sedentary time, light physical activity, or MVPA by using validated cut points,41 linearly adjusted to accommodate the 50-Hz sampling frequency. For each valid participant (waking wear time ≥10 h, sleep ≥200 min/d, sedentary time ≤1000 min/d, ≥4 valid days), an average weekly composition was created by using a 5:2 weighting for weekdays versus weekend days. |
Outcome measures | ||
Adiposity | Percentage body fat; BMI; waist girth | Percentage body fat was obtained from bioelectrical impedance analysis by using a 2-limb Tanita BC-351 (Tanita, Kewdale, Australia) at home visits and a 4-limb InBody230 (Biospace, Seoul, South Korea) at center visits.42 BMI was calculated from measured weight and height (Invicta I0955 stadiometer; Invicta, Leicester, United Kingdom) in bare feet and light clothing. Waist girth was measured on the skin at the narrowest point between the 10th rib and iliac crest. If narrowing was not detected, the measurement was taken at the midpoint of these 2 landmarks43 by using a Lufkin W606PM steel tape (Apex Tool Group, Sparks, Maryland). Waist girth and BMI were expressed as z scores relative to international norms44,45 and body fat as a sample-specific z score. The 3 variables were considered as separate outcome variables in supplementary analyses, but for the main analyses, they were averaged to form a composite adiposity summary variable. |
HRQoL | PedsQL version 4.0; 8- to 12-y-old child self-report version | The 8- to 12-y-old self-report PedsQL version 4.046 is a 23-item instrument of high internal and external reliability (α coefficients ≥0.7).47 The PedsQL comprises 4 domains: physical, emotional, social, and school functioning. Responses for each item are reverse scored on a 5-point Likert scale (0 = never a problem, 4 = almost always a problem) and linearly transformed to a 0–100 scale (0 = 100, 1 = 75, 2 = 50, 3 = 25, 4 = 0), with higher scores indicating better HRQoL.48 Physical and Psychosocial (a combination of emotional, social and school domains) Health Summary Scores were considered as separate outcome variables in supplementary analyses, whereas the main analyses used the Total Scale Score. |
Academic achievement | NAPLAN | The NAPLAN is the numeracy and literacy test undertaken by all Australian children biennially from years 3 through 9. Year 7 (when children were ∼12–13 y old) scores were used because they are the most proximal measure of academic achievement subsequent to accelerometry, limiting reverse causality. We used the numeracy score and derived a literacy score as the arithmetic mean of reading, writing, spelling, and grammar scores. The overall academic achievement score was the mean of numeracy and literacy. |
Covariates | ||
Age | Age | Age of study child (in months) (obtained from wave 6 parent questionnaires) |
Sex | Sex | Sex of study child (female or male) (obtained from wave 6 parent questionnaires) |
SEP | SEP | LSAC releases a composite z score for family-level SEP with each wave, with higher scores indicating higher SEP. The score is derived from parent-reported household income levels, education levels, and occupation49 (obtained from wave 6 parent questionnaires). |
Pubertal status | PDS | Children used iPads to report sex-specific pubertal signs using the PDS.50 The PDS consists of 5 sex-specific questions related to growth spurts, body hair, and acne; facial hair and voice deepening for boys; and breast growth and menstruation for girls. Responses were summed to obtain a total PDS score, used to categorize children into prepubertal, early pubertal, midpubertal, late pubertal, or postpubertal. |
MATLAB, matrix laboratory software; PC, personal computer; PDS, Pubertal Development Scale; SEP, socioeconomic position.
Time Use (Exposure)
Children’s 24-hour time use (ie, average daily time spent in sleep, sedentary time, light physical activity, and MVPA) was quantified by using accelerometry.
Adiposity, HRQoL, and Academic Outcomes
We selected 3 diverse outcome variables spanning public health (adiposity), perceived physical and psychosocial health (HRQoL), and education (academic achievement) domains that are important to children, families, and clinicians. Adiposity was operationalized as a composite of percentage body fat, BMI, and waist girth. HRQoL was measured via a 23-item profile measure, the Pediatric Quality of Life Inventory (PedsQL). Data on academic achievement were accessed from the LSAC data set, with 95% of LSAC parents consenting to linkage to standardized National Assessment Program – Numeracy and Literacy (NAPLAN). Before analysis, all outcomes were expressed as sample-specific z scores to standardize interpretation.
Covariates
Data Treatment
Each participant’s daily time use was conceptualized as a time-use composition. Following the principles of compositional data analysis, compositions were expressed as ratios of their parts.20 As described in detail elsewhere, a specific type of isometric log ratio (ilr) was used.16,21,22 Four sets of ilrs were created. The first ratio in each set represented one part (ie, sleep, sedentary time, light physical activity or MVPA), relative to the remaining parts. There were no zeros recorded in any activity that would prevent log-ratio transformation.23
Statistical Analyses
Descriptive and compositional data analyses were conducted in R version 3.6 (R Foundation for Statistical Computing, Vienna, Austria) by using the Compositions and zCompositions packages.24,25 Descriptive statistics were calculated for all variables and compositional means (ie, geometric means of activities, adjusted to a total of 1440 minutes) for time-use data.26
Relationships Between Time-Use Variables and Health Variables
Compositional multivariable linear regression models investigated how each health outcome was associated with the time-use composition (expressed as ilr coordinates). This solved the problem that time-use reallocation pathways cannot be explored by using standard statistical approaches because of perfect multicollinearity.27 Models included only participants with complete data and were adjusted for the a priori potential confounders.
As described previously,16,21,22 each set of ilr coordinates was iteratively used in the multivariable linear regression model. The regression coefficient of the log ratio represented the relative dominance of the first compositional part, relative to the geometric mean of all remaining parts. For example, the first log ratio of the set representing MVPA, relative to the remaining activities, had MVPA in the numerator and the geometric mean of the remaining activities (sleep, sedentary time, and light physical activity) in the denominator. Standardized log-ratio βs were reported to facilitate interpretable comparisons.
Equivalence Curves
We generated equivalence curves to explore equivalent trade-offs between children’s activities in relation to changes in health. The ilr multivariable linear regression models estimated how much time must be reallocated from and to different activities (relative to the remaining activities) to achieve a specific change from the mean in each standardized health outcome. For example, 10 minutes more MVPA might be equivalent to 20 minutes more sleep in relation to a 0.10 SD increase in HRQoL (both relative to the remaining time-use composition). The paired values (ie, the reallocations to and from each domain required to produce the same change in the health outcome) were then used to plot equivalence curves of changes in one activity versus changes in another. All estimates were made relative to the CheckPoint sample’s compositional center, which is analogous to the average daily activity durations observed in the sample.
Results
Sample Characteristics
The participant flow through LSAC to the final analytical CheckPoint samples is displayed in Fig 1. Characteristics of included participants are described compared with excluded participants in Table 2. Compared with the original LSAC birth cohort children, the analytical sample had a higher mean socioeconomic position (0.24 vs 0.00). Included CheckPoint participants had higher academic achievement scores and lower BMI z score (zBMI) than excluded participants.
Comparison between included and excluded participant characteristics
Characteristic . | Included (N=1181) . | Excluded (N=693) . | Difference P valuesa . |
---|---|---|---|
Age, mean (SD), years | 12.0 (0.4) | 12.0 (0.4) | 0.47 |
Sex, No. (%) female | 578 (49) | 341 (49) | 0.95 |
Socioeconomic position, mean (SD), z-score | 0.24 (0.99) | 0.06 (0.99) N=686 | <0.001 |
Pubertal status, No. (%) | 0.73 | ||
Pre-pubertal | 120 (10) | 45 (8) | |
Early pubertal | 297 (25) | 148 (27) | |
Mid-pubertal | 603 (51) | 282 (51) | |
Late pubertal | 155 (13) | 74 (13) | |
Post-pubertal | 6 (<1) | 3 (<1) | |
Activity behaviors: arithmetic mean (SD), min/d | |||
Sleep | 566 (46) | 566 (46) N=101 | 0.98 |
Sedentary time | 554 (81) | 542 (80) N=101 | 0.18 |
Light physical activity | 251 (57) | 258 (54) N=101 | 0.20 |
MVPA | 62 (34) | 65 (33) N=101 | 0.36 |
Activity behaviors: compositional meanb, min/d | |||
Sleep; Sedentary time; Light physical activity; MVPA | 577; 560; 250; 53 | 577; 549; 257; 57 | 0.56 |
Anthropometry (N=1175) | |||
%Body fat, mean (SD) | 21.5 (8.4) | 22.2 (8.6) N=683 | 0.08 |
BMI, mean (SD), z-score | 0.44 (1.15) | 0.55 (1.15) N=696 | 0.05 |
Waist circumference, mean (SD), z-score | 0.80 (1.07) | 0.88 (1.11) N=692 | 0.10 |
HRQL (N=1181) | |||
PedsQL Physical Score, mean (SD) | 84.3 (12.1) | 84.4 (13.2) N=667 | 0.96 |
PedsQL Psychosocial Score, mean (SD) | 77.17 (13.8) | 76.6 (14.5) N=667 | 0.42 |
PedsQL Total Scale Score, mean (SD) | 79.7 (12.2) | 79.3 (12.8) N=666 | 0.51 |
Academic (N=931) | |||
NAPLAN numeracy, mean (SD) | 580 (70) | 563 (66) N=492 | <0.001 |
NAPLAN writing, comprehension, spelling, grammar, mean (SD) | 559 (61) | 546 (61) N=511 | <0.001 |
Overall academic achievement, mean (SD) | 569 (61) | 554 (59) N=489 | <0.001 |
Characteristic . | Included (N=1181) . | Excluded (N=693) . | Difference P valuesa . |
---|---|---|---|
Age, mean (SD), years | 12.0 (0.4) | 12.0 (0.4) | 0.47 |
Sex, No. (%) female | 578 (49) | 341 (49) | 0.95 |
Socioeconomic position, mean (SD), z-score | 0.24 (0.99) | 0.06 (0.99) N=686 | <0.001 |
Pubertal status, No. (%) | 0.73 | ||
Pre-pubertal | 120 (10) | 45 (8) | |
Early pubertal | 297 (25) | 148 (27) | |
Mid-pubertal | 603 (51) | 282 (51) | |
Late pubertal | 155 (13) | 74 (13) | |
Post-pubertal | 6 (<1) | 3 (<1) | |
Activity behaviors: arithmetic mean (SD), min/d | |||
Sleep | 566 (46) | 566 (46) N=101 | 0.98 |
Sedentary time | 554 (81) | 542 (80) N=101 | 0.18 |
Light physical activity | 251 (57) | 258 (54) N=101 | 0.20 |
MVPA | 62 (34) | 65 (33) N=101 | 0.36 |
Activity behaviors: compositional meanb, min/d | |||
Sleep; Sedentary time; Light physical activity; MVPA | 577; 560; 250; 53 | 577; 549; 257; 57 | 0.56 |
Anthropometry (N=1175) | |||
%Body fat, mean (SD) | 21.5 (8.4) | 22.2 (8.6) N=683 | 0.08 |
BMI, mean (SD), z-score | 0.44 (1.15) | 0.55 (1.15) N=696 | 0.05 |
Waist circumference, mean (SD), z-score | 0.80 (1.07) | 0.88 (1.11) N=692 | 0.10 |
HRQL (N=1181) | |||
PedsQL Physical Score, mean (SD) | 84.3 (12.1) | 84.4 (13.2) N=667 | 0.96 |
PedsQL Psychosocial Score, mean (SD) | 77.17 (13.8) | 76.6 (14.5) N=667 | 0.42 |
PedsQL Total Scale Score, mean (SD) | 79.7 (12.2) | 79.3 (12.8) N=666 | 0.51 |
Academic (N=931) | |||
NAPLAN numeracy, mean (SD) | 580 (70) | 563 (66) N=492 | <0.001 |
NAPLAN writing, comprehension, spelling, grammar, mean (SD) | 559 (61) | 546 (61) N=511 | <0.001 |
Overall academic achievement, mean (SD) | 569 (61) | 554 (59) N=489 | <0.001 |
aDifference tested using t-tests for continuous variables, chi-squared for categorical variables and compositional MANOVA for compositional variables; bCompositional mean was calculated by finding the geometric mean of each activity and linearly adjusting these values to collectively sum to 1440 (min/day). MVPA = moderate-to-vigorous physical activity; HRQL = Health-related quality of life; PedsQL = Pediatric Quality of Life Inventory; NAPLAN = National Assessment Program – Literacy and Numeracy
Compositional Linear Regression Models
Standardized log-ratio regression coefficients (βs) for each activity behavior relative to all remaining activity behaviors from multivariable linear regression models for summary outcomes (adiposity, HRQoL, and academic achievement) are in Table 3. Additional results for individual outcomes are in Supplemental Table 5. The βs correspond to the first log ratio from the respective regression model, representing the increase of one activity, relative to the decrease of all the remaining activities. For example, the β of −.21 for adiposity (model 1) indicates that the adiposity composite score is estimated to decrease by 0.21 SDs when the log ratio of MVPA relative to the remaining activities increases by 1 SD. Lower adiposity was associated with more time spent in MVPA (standardized β [βstd] = −.21; P < .001) and sleep (βstd = −0.17; P = .002) and less sedentary time (βstd = .14; P = .02) and light physical activity (βstd = .16; P < .001), each relative to the remaining activities. Higher HRQoL was associated with more MVPA (βstd = .15; P < .001) and sleep (βstd = .14; P = .008) and less sedentary time (βstd = −.20; P = .001), and light physical activity (βstd = −.07; P = .04), each relative to the remaining activities. Better academic achievement was associated with less light physical activity, (βstd = −.10; P = .006) and more sedentary time (βstd = .14; P = .03), each relative to the remaining activities.
Table 3. Associations between activity behaviors and outcomes estimated by compositional multivariable regression models.
Outcome (z-score) . | Standardized beta (95% CI) . | P value . |
---|---|---|
Adiposity (N=1175) | ||
Model 1. MVPA: R | -0.21 (-0.28 to -0.15) | <0.001 |
Model 2. Light physical activity: R | 0.16 (0.10 to 0.23) | <0.001 |
Model 3. Sedentary time: R | 0.14 (0.03 to 0.26) | 0.02 |
Model 4. Sleep: R | -0.17 (-0.27 to -0.06) | 0.002 |
HRQL (N=1181) | ||
Model 1. MVPA: R | 0.15 (0.08 to 0.22) | <0.001 |
Model 2. Light physical activity: R | -0.07 (-0.14 to -0.00) | 0.04 |
Model 3. Sedentary time: R | -0.20 (-0.32 to -0.08) | 0.001 |
Model 4. Sleep: R | 0.14 (0.04 to 0.25) | 0.008 |
Academic achievement (N=931) | ||
Model 1. MVPA: R | -0.01 (-0.08 to 0.07) | 0.84 |
Model 2. Light physical activity: R | -0.10 (-0.17 to -0.03) | 0.006 |
Model 3. Sedentary time: R | 0.14 (0.01 to 0.27) | 0.03 |
Model 4. Sleep: R | 0.00 (-0.11 to 0.11) | 0.98 |
Outcome (z-score) . | Standardized beta (95% CI) . | P value . |
---|---|---|
Adiposity (N=1175) | ||
Model 1. MVPA: R | -0.21 (-0.28 to -0.15) | <0.001 |
Model 2. Light physical activity: R | 0.16 (0.10 to 0.23) | <0.001 |
Model 3. Sedentary time: R | 0.14 (0.03 to 0.26) | 0.02 |
Model 4. Sleep: R | -0.17 (-0.27 to -0.06) | 0.002 |
HRQL (N=1181) | ||
Model 1. MVPA: R | 0.15 (0.08 to 0.22) | <0.001 |
Model 2. Light physical activity: R | -0.07 (-0.14 to -0.00) | 0.04 |
Model 3. Sedentary time: R | -0.20 (-0.32 to -0.08) | 0.001 |
Model 4. Sleep: R | 0.14 (0.04 to 0.25) | 0.008 |
Academic achievement (N=931) | ||
Model 1. MVPA: R | -0.01 (-0.08 to 0.07) | 0.84 |
Model 2. Light physical activity: R | -0.10 (-0.17 to -0.03) | 0.006 |
Model 3. Sedentary time: R | 0.14 (0.01 to 0.27) | 0.03 |
Model 4. Sleep: R | 0.00 (-0.11 to 0.11) | 0.98 |
Analyses adjusted for age, sex, pubertal status and household socioeconomic position.
MVPA = moderate-to-vigorous physical activity; R = Geometric mean of remaining activities. HRQL = health-related quality of life. Adiposity is a composite of body mass index z-score, waist girth z-score and percentage body fat; HRQL is estimated as the PedsQL Total Scale Score; Academic achievement is the mean of Grade 7 numeracy and literary (mean of reading, writing, spelling, and grammar and punctuation NAPLAN scores).
Equivalence Values and Curves
Presented in Table 4 are the changes to or from each of the 4 time-use components, relative to all others, associated with a model-predicted −0.10, −0.05, +0.05, or +0.10 standardized change in each outcome. Broadly, it can be seen that smaller durations of MVPA needed to be reallocated for the same benefits to adiposity and HRQoL, compared with all other behaviors. The smallest reallocations associated with the same predicted benefit in academic achievement were those taking time away from light physical activity to the remaining behaviors.
Differences (min/d) in activity behaviors (relative to remaining behaviors) associated with differences in outcomes
. | Δ min/d associated with ± 0.05 or 0.10 SD difference in outcomes . | |||
---|---|---|---|---|
Outcome | -0.10 | -0.05 | +0.05 | +0.10 |
Adiposity | ||||
MVPA | 19 | 9 | -8 | -14 |
Light physical activity | -34 | -17 | 18 | 38 |
Sedentary time | -89 | -45 | 47 | 95 |
Sleep | 55 | 27 | -27 | -53 |
Health-related quality of life | ||||
MVPA | -19 | -11 | 13 | 29 |
Light physical activity | 93 | 44 | -38 | -72 |
Sedentary time | 67 | 33 | -33 | -65 |
Sleep | -61 | -31 | 32 | 64 |
Academic achievement | ||||
Light physical activity | 62 | 30 | -27 | -52 |
Sedentary time | -89 | -46 | 47 | 95 |
. | Δ min/d associated with ± 0.05 or 0.10 SD difference in outcomes . | |||
---|---|---|---|---|
Outcome | -0.10 | -0.05 | +0.05 | +0.10 |
Adiposity | ||||
MVPA | 19 | 9 | -8 | -14 |
Light physical activity | -34 | -17 | 18 | 38 |
Sedentary time | -89 | -45 | 47 | 95 |
Sleep | 55 | 27 | -27 | -53 |
Health-related quality of life | ||||
MVPA | -19 | -11 | 13 | 29 |
Light physical activity | 93 | 44 | -38 | -72 |
Sedentary time | 67 | 33 | -33 | -65 |
Sleep | -61 | -31 | 32 | 64 |
Academic achievement | ||||
Light physical activity | 62 | 30 | -27 | -52 |
Sedentary time | -89 | -46 | 47 | 95 |
MVPA = moderate-to-vigorous physical activity.
From these values, we generated pairwise equivalence curves for all outcomes (Fig 2: adiposity; Fig 3: HRQoL; Fig 4; academic achievement).
Equivalence curves for adiposity. The values in brackets identify the difference (minutes per day) in activity behaviors, with the behavior in the x-axis before the comma and the behavior in the y-axis after the comma. For example, the “(+17, +52)” from (A) indicates an estimated 0.10 SD (beneficial) decrease in adiposity is associated with either reallocating 17 minutes to MVPA from the remaining activities or reallocating 52 minutes to sleep from the remaining activities. A, MVPA versus sleep. B, MVPA versus sedentary time (SED). C, MVPA versus light physical activity (LPA). D, Sleep versus SED. E, Sleep versus LPA. F, LPA versus SED.
Equivalence curves for adiposity. The values in brackets identify the difference (minutes per day) in activity behaviors, with the behavior in the x-axis before the comma and the behavior in the y-axis after the comma. For example, the “(+17, +52)” from (A) indicates an estimated 0.10 SD (beneficial) decrease in adiposity is associated with either reallocating 17 minutes to MVPA from the remaining activities or reallocating 52 minutes to sleep from the remaining activities. A, MVPA versus sleep. B, MVPA versus sedentary time (SED). C, MVPA versus light physical activity (LPA). D, Sleep versus SED. E, Sleep versus LPA. F, LPA versus SED.
Equivalence curves for HRQoL. A, MVPA versus sleep. B, MVPA versus sedentary time (SED). C, Sleep versus SED.
Equivalence curves for HRQoL. A, MVPA versus sleep. B, MVPA versus sedentary time (SED). C, Sleep versus SED.
By observing where a vertical line extending up from the x-axis at −0.1 SD intersects the curve in Figures 2−4, we can read the pair-wise activity trade-offs associated with a difference of −0.1SD in the outcome. For example, Figure 2A shows that an estimated adiposity difference of −0.1SD was associated with either +55 min sleep or +19 min MVPA (each relative to remaining activities). Similarly, by observing where a vertical line at +0.1 SD intersects the HRQoL curve (Fig 3A), we see that either +64 minutes of sleep, or +29 minutes of MVPA (each relative to remaining activities) are associated with an HRQoL difference of +0.1 SD. Compared with other time reallocations, MVPA increments were therefore much more potent for lower adiposity and higher HRQoL. On Figure 4, a vertical line at +0.1SD intersects the academic achievement curve at +95 min sedentary time and −52 min light physical activity, meaning more sedentary time was almost twice as potent as less light physical activity.
Discussion
Principal Findings
On a minute-for-minute basis, reallocations of time to MVPA were ∼2 to 6 times as potent as reallocations to sleep or away from sedentary time in terms of lower adiposity and higher HRQoL. Reallocation of time to light physical activity was unfavorably associated with all outcomes.
Comparison With Previous Literature
Although this report of equivalent time use “means” to achieve the same healthful “ends” is entirely novel, in other respects, our findings are consistent with the current literature. Thus, authors of both compositional and noncompositional studies have found higher MVPA/sleep and lower sedentary time to be favorably associated with HRQoL and adiposity,2 consistent with putative mechanistic explanations.28–31 Our adiposity findings are comparable to 2 compositional isotemporal substitution studies using zBMI measures,21,32 despite differences in accelerometer type, analytical algorithms, and sample characteristics. The unfavorable association between light physical activity and adiposity aligns with recent compositional evidence32 but challenges noncompositional evidence suggesting favorable associations of light physical with adiposity and other health indicators.33,34
The negative association between time reallocations to light physical activity (from all other activities) and academic achievement is consistent with a previous study.35 The finding of beneficial associations between sedentary time and academic achievement is also supported by previous studies.35,36 Our coefficients for the other activities (sleep and MVPA), each relative to the remaining activities, revealed no relationships with academic achievement, on the surface contradicting previous studies reporting favorable numeracy and literacy associations with more sleep, sedentary time, and MVPA.2,3,35–38 In fact, our study is congruent with these previous findings. We showed poorer academic achievement when light physical activity was increased while the remaining activities (sleep and MVPA) were decreased to compensate. Thus, the opposite is also true: taken together, more sleep, sedentary time, and MVPA (relative to light physical activity) were associated with better academic achievement.
Strengths and Limitations
Strengths of this study include the large sample drawn from a nationally representative cohort and valid, reliable, and widely used measures of both exposure and outcomes. In previous compositional studies, researchers have quantified how different reallocations of time between activity behaviors (isotemporal substitutions) are associated with health outcomes in different ways. Our study extends the compositional approach by exploring how different reallocations of time between behaviors may impact health in equivalent ways. This allows us to weigh up activities against each other, providing information for activity behavior choices in the context of a constrained 24-hour day. However, findings from compositional analyses can appear complex and their interpretation can be challenging. For example, while we talk about the increase of an activity (eg, MVPA), we are simultaneously considering the decrease of all remaining activities (sleep, sedentary time, and light physical activity).
Our “exposures” and “outcomes” terminology reflects public health priorities of seeking time-use solutions that improve health, but we acknowledge that differences in health outcomes may not represent effects of time reallocations because this was a cross-sectional study. Relationships could well be bidirectional (ie, time use may impact on health, but health could equally impact on time use), or both may be caused by an unrecognized third factor.
Although LSAC originally aimed to be nationally representative, our results may be less accurate for more disadvantaged children because of self-selection and attrition. In future studies, the use of population weights may improve generalizability of findings. Our findings are also specific to how the activity composition was measured. Different accelerometers and analytical algorithms will yield different estimates of time spent in each activity. Although accelerometers are considered valid and reliable for assessing activity behaviors, estimated durations can vary because of wear protocols and processing procedures. For example, estimates can be affected by device location (wrist, hip), the choice of cut points for classifying activities, and the epoch length. Accelerometer counts were converted to 60-second epoch files, which is long in comparison with other pediatric studies, and may have contributed to relatively low MVPA estimates in this study.
We only considered average daily durations of activities; however, the characteristics of accumulating these activities may also be important, for example, the timing of activity (morning, evening), fragmentation (short or long bouts), and day-to-day consistency.
Interpretation and Implications
Our findings have important clinical and policy implications: children with the same health status may have different time use. This suggests that the same improvement (or decrement) in health might be achievable by selecting from a menu of several different activity-change strategies. Such trade-off choices may allow flexibility to fit in with preferences or with life constraints, such as school and home schedules. There are many lifestyle recommendations for children (eg, sleep more, exercise more, sit less), but families are busy. There are only 24 hours in a day and any behavioral adjustment will have ripple effects across other activities. Although children with different durations of activities may have the same health status, clinicians may recommend increasing MVPA because this seems to require less time to bring about the same beneficial associations as increasing sleep or decreasing sitting. Because children with longer sleep had better adiposity and HRQoL, clinicians may advise that any increases in MVPA should not replace sleep. Another healthful option may be to reduce sedentary time and light physical activities to increase sleep while maintaining MVPA.
The best way of implementing these findings is not yet clear. Our trade-offs were based on the model of increasing one behavior by taking equal proportions of time from all remaining behaviors. It may be that swapping one time-use behavior for another (one-for-one trade-offs) may facilitate public health messages that mean more to individuals.39 Alternatively, it may be more sensible to model the type of trade-offs people actually make in real life under imposed demands (empirical trade-offs). There has been little research in this area, but intervention data sets are available.
On a minute-for-minute basis, MVPA is exceptionally potent (2–6 times as potent as sleep and sedentary time in this study). However, it may be more meaningful to express time reallocations in terms of how much time committed to each activity has actually been shown to be achievable by interventions. For example, it has proved difficult to increase time allocated to sleep.12
Focusing on activity compositions rather than individual activities opens up new intervention possibilities. Rather than addressing just one domain (“increase MVPA”), one could look at how to change activity patterns (“trade-off screen time against sleep and MVPA”). This is consonant with stealth interventions in which the real target activity domain is concealed behind an ostensible target (eg, trying to increase sleep duration by reducing screen time at night).
Future Directions
Our novel equivalence curves illustrate how alternative alterations to activity compositions may be associated with the same benefits or decrements to specific health outcomes. Longitudinal and intervention studies could demonstrate causality and help understand the nature of empirical reallocations. In studies, researchers should encompass broader outcomes, such as bone and cardiovascular health,3,36 and different populations and subgroups. Alternative reallocation modes (one-for-one, empirical) should be explored for greatest clinical and public health gain.
Acknowledgments
In this article, we use unit record data from Growing Up in Australia: The Longitudinal Study of Australian Children. The study is conducted in partnership between the Department of Social Services, the Australian Institute of Family Studies, and the Australian Bureau of Statistics. The findings and views reported in this article are solely those of the authors and should not be attributed to the Department of Social Services, Australian Institute of Family Studies, or Australian Bureau of Statistics. Emily Ng (The University of Melbourne) and Dorothea Dumuid (University of South Australia) had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Research Electronic Data Capture tools were used in this study. More information about this software can be found at www.project-redcap.org. We thank the LSAC and CheckPoint study participants, staff, and students for their contributions.
Ms Ng and Dr Dumuid analyzed and interpreted the data, drafted the manuscript, had full access to all the data in the study, and take responsibility for the integrity of the data and the accuracy of the data analysis; Prof Olds conceived the compositional equivalence concept, is a study investigator involved in the conception and oversight of the Child Health CheckPoint, contributed to the collection, analysis, and interpretation of the data, and drafted the manuscript; Prof Wake is the principal investigator involved in the conception and oversight of the Child Health CheckPoint, contributed to the collection and analysis and interpretation of the data, and drafted the manuscript; Dr Lycett and Assoc Prof Edwards contributed to the acquisition, analysis, and interpretation of the data and critically revised the manuscript; Dr Le contributed to the analysis and interpretation of the data and critically revised the manuscript; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
FUNDING: Supported by the National Health and Medical Research Council (Early Career Fellowships 1162166 to Dr Dumuid and 1091124 to Dr Lycett; Principal Research Fellowship 1160906 to Dr Wake). This work was also supported by the National Heart Foundation of Australia (Postdoctoral Fellowship 102084 to Dr Dumuid and Honorary Postdoctoral Fellowship 101239 to Dr Lycett). The CheckPoint study is supported by the National Health and Medical Research Council (project grants 1041352 and 1109355), The Royal Children’s Hospital Foundation (2014-241), the Murdoch Children’s Research Institute, The University of Melbourne, the National Heart Foundation of Australia (100660), and the Financial Markets Foundation for Children (2014-055 and 2016-310). Research at the Murdoch Children’s Research Institute is supported by the Victorian Government’s Operational Infrastructure Support Program.
COMPANION PAPER: A companion to this article can be found online at www.pediatrics.org/cgi/doi/10.1542/peds.2020-042168.
Abbreviations
References
Competing Interests
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
Comments