Time-Use Patterns and Health-Related Quality of Life in Adolescents
OBJECTIVES: To describe 24-hour time-use patterns and their association with health-related quality of life (HRQoL) in early adolescence.
METHODS: The Child Health CheckPoint was a cross-sectional study nested between Waves 6 and 7 of the Longitudinal Study of Australian Children. The participants were 1455 11- to 12-year-olds (39% of Wave 6; 51% boys). The exposure was 24-hour time use measured across 259 activities using the Multimedia Activity Recall for Children and Adolescents. “Average” days were generated from 1 school and 1 nonschool day. Time-use clusters were derived from cluster analysis with compositional inputs. The outcomes were self-reported HRQoL (Physical and Psychosocial Health [PedsQL] summary scores; Child Health Utility 9D [CHU9D] health utility).
RESULTS: Four time-use clusters emerged: “studious actives” (22%; highest school-related time, low screen time), “techno-actives” (33%; highest physical activity, lowest school-related time), “stay home screenies” (23%; highest screen time, lowest passive transport), and “potterers” (21%; low physical activity). Linear regression models, adjusted for a priori confounders, showed that compared with the healthiest “studious actives” (mean [SD]: CHU9D 0.84 [0.14], PedsQL physical 86.8 [10.8], PedsQL psychosocial 79.9 [12.6]), HRQoL in “potterers” was 0.2 to 0.5 SDs lower (mean differences [95% confidence interval]: CHU9D −0.03 [−0.05 to −0.00], PedsQL physical −5.5 [−7.4 to −3.5], PedsQL psychosocial −5.8 [−8.0 to −3.5]).
CONCLUSIONS: Discrete time-use patterns exist in Australian young adolescents. The cluster characterized by low physical activity and moderate screen time was associated with the lowest HRQoL. Whether this pattern translates into precursors of noncommunicable diseases remains to be determined.
- CHU9D —
- Child Health Utility 9D
- CI —
- confidence interval
- HRQoL —
- health-related quality of life
- LSAC —
- Longitudinal Study of Australian Children
- MARCA —
- Multimedia Activity Recall for Children and Adolescents
- PedsQL —
- Pediatric Quality of Life Inventory
What’s Known on This Subject:
Lifestyle is a major determinant of adult health. Both lifestyle and health trajectories are well established by adolescence, with lifetime consequences. However, it remains unknown whether overall time-use patterns are already associated with health-related quality of life by this age.
What This Study Adds:
Four distinct time-use clusters emerged. Compared with the healthiest “studious actives,” health-related quality of life in “potterers” was 0.2 to 0.5 SDs lower. It is possible that promoting healthful time-use patterns, rather than individual activities, could improve adolescent health.
Both lifestyle and health trajectories are well established by early adolescence. If time use, a major determinant of lifestyle, already has a unidirectional or bidirectional relationship with health by this age,1 the consequences could be lifelong. One way to measure health is through health-related quality of life (HRQoL), which is increasingly used to guide health care resource allocation.2 Time use is modifiable, so if it is associated with HRQoL, it is possible that targeting nonoptimal patterns of behaviors could improve lifetime health.
However, measuring time use is methodologically challenging. Ideally, it is measured via a time diary that records all activities undertaken in a 24-hour day,3 which is important because whenever time spent in 1 activity domain increases, time spent in another must decrease.1,4 This contrasts with more commonly used stylized questions, where respondents estimate the usual duration of ≥1 specified activity domains. Furthermore, a person-centered analytical approach, which explores activity patterns by using methods such as cluster analysis, is preferred over a variable-centered approach, which investigates specific activity domains in isolation, because a variable-centered approach does not acknowledge the interdependence and complex relationships between activity domains.5
Little is known about how HRQoL varies according to adolescents’ time use on a person-centered level.5–7 Our recent scoping review (Wong M, Olds T, Gold L, et al, unpublished data) identified only 1 study (Hunt et al7) that examined this association, whereas 4 examined person-centered time use and well-being or functioning (an alternative to HRQoL),8–11 and 23 used variable-centered analyses to look at specific activities and HRQoL. Hunt’s study of 667 Irish 15- to 19-year-olds showed that a well-balanced lifestyle was not associated with better HRQoL in boys. However, 40% of girls in the “moderate study/higher leisure” group scored >0.5 SD units above the study mean for HRQoL, compared with only 18% in the “higher study/lower leisure” group.7 Of the 4 other person-centered time-use studies, Nelson and Gastic’s9 study of 6338 American 10th-grade 14 to 15 year-olds was the most robust. Their “unstructured recreation” group had the worst perceptions of school environment, academic performance, and levels of truancy across the 5 clusters identified, whereas their “study” cluster showed the best well-being and functioning. In the variable-centered studies, more frequent physical activity,12 less screen time,13 and adequate or more sleep14,15 in isolation were associated with marginally better HRQoL.
It therefore remains unclear how adolescent overall time use is associated with HRQoL. In a large population-based study of Australian 11- to 12-year-olds, we aimed to examine time-use clusters by using a well-validated 24-hour time-use diary with a person-centered approach and associations between time-use clusters and HRQoL. We hypothesized that distinct time-use clusters exist by early adolescence, and the most “balanced” time-use cluster (ie, moderate levels of participation in multiple different time use domains) would have the best HRQoL.
Design and Setting
The Child Health CheckPoint is a cross-sectional wave nested within the national population-based Longitudinal Study of Australian Children (LSAC).16 CheckPoint data collection took place February 2015 to March 2016 between LSAC’s sixth and seventh waves, when children were aged 11 to 12 years. The project was approved by The Royal Children’s Hospital (Melbourne) Human Research Ethics Committee (HREC33225) and the Australian Institute of Family Studies Ethics Committee (AIFS14-05). A parent or guardian provided written informed consent.
Using a 2-stage cluster randomized design, LSAC recruited a nationally representative sample of 5107 infants (age 0–1 years, birth or “B” cohort). LSAC has since followed children biennially, with 6 waves of data collection completed up to 2015. CheckPoint was a 1-time physical health and biomarker module offered to all families enrolled in LSAC’s B-cohort who participated in Wave 6 (retention rate 74%). Families opted in to hear more about CheckPoint by providing written consent to be contacted by CheckPoint.
From December 2014, the CheckPoint team sent an invitation and subsequently telephoned each family to ascertain interest and book an appointment, as the CheckPoint assessment center traveled sequentially to each state. Families received a prepaid electronic funds transfer card to offset travel expenses ($30–$120 depending on traveled distance). Most families attended a 3.5-hour appointment at the main assessment center, where they rotated through 17 stations. Mini-centers, operating in smaller regional cities, consisted of a slightly shorter 2.5-hour appointment with a subset of measures. Families who could not attend an assessment center were offered a 1.5-hour home visit. All center visits included collection of a time-use diary for the previous 24 hours, but home visits did not.
At the assessment, the CheckPoint team scheduled a time to call the child ∼7 days later. The date was selected such that, once the child had reported time-use data spanning the 2 days preceding the call, the CheckPoint should have obtained data on ≥1 school day and ≥1 nonschool day. At the assessment, children were also given an activity card to take home to note meal times and important events as a memory aid. Thus, ≤3 days of time diaries were available for children attending assessment centers and ≤2 days for those having a home visit.
Table 1 details all measures.
Time use was measured via the Multimedia Activity Recall for Children and Adolescents (MARCA), a computerized 24-hour recall time diary (recorded continuously at intervals of ≥5 minutes) with excellent test–retest reliability and good convergent validity.3,17 It includes 259 activities in 6 categories: screen time (any screen-based activities, eg, use of smartphones, television), physical activity (including active play and active transport; eg, cycling to school), school (structured classes, studying, homework including screen-based work, reading, and music), sleep, passive transport (eg, car ride), and domestic/work (social activities, grooming, eating, and chores), based on the system developed by Ridley et al.26 Adolescents were required to have completed the MARCA on ≥1 school day and ≥1 nonschool day, because distinct patterns arise on these days. These were used to generate an “average” day for analysis, with school and nonschool days weighted 1:1 (because children typically spend ∼50% of all days of the year in school). For example, if 2 school days and 1 nonschool day were collected, the school days were averaged, and this average was then averaged with the nonschool day.27
HRQoL was measured on the assessment day via 2 measures: a 9-item utility measure, the Child Health Utility 9D (CHU9D), and a 23-item profile measure, the Pediatric Quality of Life Inventory (PedsQL). The CHU9D is a recently developed utility measure for children. Item scores are valued based on population utility values generated from children for the CHU9D to produce a composite utility score along a continuum from −0.11 (negative values are worse than death) to 1 (perfect health), with 0 equivalent to death.21,28 The PedsQL is a well-validated and widely used profile measure, which provides Physical and Psychosocial Health Summary scores that each range from 0 to 100, with higher scores indicating better HRQoL.18
Potential confounders are also described in Table 1. They included sex, age, BMI z score based on Centers for Disease Control reference values,23 pubertal status, and neighborhood disadvantage (Socioeconomic Indexes for Areas Disadvantage Index).
Time-use data were treated via the Compositions package for R (R Foundation for Statistical Computing, Vienna, Austria) in preparation for compositional cluster analysis.29 First, because the presence of zeros prevents the log-ratio transformation needed for the analysis of compositional data, zeros were replaced with small nonzero values.30 To accommodate the closed nature of the cluster inputs, the 6 time-use categories were expressed as isometric log-ratio coordinates, because they represent the relative proportions of the components while avoiding multicollinearity concerns.31,32
Second, cluster analysis is sensitive to the presence of outliers, so these were removed from the data set before cluster analysis.33 The criterion for removal was defined according to the robust squared Mahalanobis distance (limit set at 99.99th percentile of the χ2 distribution) after an isometric log-ratio transformation.33 Adolescents were stratified by sex to allow examination of sex-specific time-use patterns.
Identification of time-use clusters (Aim 1) was undertaken in the R program via compositional cluster analyses,29 as we have previously reported by using the MARCA data.36 This method does not generate goodness-of-fit statistics; rather, the best cluster solution was determined by a combination of approaches. First, to explore possible cluster structure within the data set, we generated a dendrogram (Supplemental Fig 3) through agglomerative hierarchical clustering (Ward’s method and Euclidean distances).33 Subsequently, the Calinski–Harabasz rule was used to define an optimum number of clusters for both boys and girls (see elbow plot in Supplemental Fig 4).37 Finally, k means partitioning cluster analysis was carried out with the cluster centers identified by the preceding hierarchical procedure.38 The characteristics of the resulting clusters were described through compositional means (the geometric mean of each component, adjusted to sum to 1440 minutes, ie, to the total minutes in a 24-hour day), and variation matrices were used to describe the multivariate spread of the components.39 After visual inspection, we then selected 1 cluster a priori as the reference category for subsequent regression analyses, based on the literature regarding “healthful” characteristics (see Results).
Relationships between time-use clusters and HRQoL (Aim 2) were examined in Stata 14.0.34 We treated time use clusters as “exposures” and HRQoL as “outcomes.” However, we acknowledge that the associations described do not imply directionality, given the study’s cross-sectional nature. We examined HRQoL scores between clusters visually by plotting marginal means and 95% confidence intervals (CIs), followed by analysis of variance and unadjusted and adjusted linear regression models. The adjusted model included sex, age, BMI z score, pubertal status, and neighborhood disadvantage.
Of 3764 eligible adolescents, 1874 (50% of Wave 6) participated in CheckPoint; 1455 (39%) of these had both MARCA and HRQoL data, comprising the analytic sample (see Supplemental Fig 5). The analytic sample came from slightly less disadvantaged neighborhoods (mean 1028 [SD 60]) compared with eligible nonparticipants (mean 1008 [SD 69]), but they were comparable in terms of age and sex.
Table 2 describes the participant characteristics. Age ranged from 11 to 12 years (mean 11.5, SD 0.5), boys and girls were approximately equally represented, 10% were prepubertal, and 76% were in early or midpuberty. CHU9D scores were similar to Australian norms.22 PedsQL Physical and Psychosocial Health Summary scores were marginally lower than the American normative values.19
Time-Use Clusters (Aim 1)
For MARCA data, 75 outliers were removed.33 To check the robustness of the cluster solutions, random samples of 500 boys and 500 girls were reclustered according to the same procedures. Excellent agreement was found between the 2 solutions (Cohen’s κ = .90 for boys and .93 for girls). Preliminary cluster analyses found characteristics of clusters to be similar for boys and girls, so the final cluster solution was rederived for both sexes together. Another 16 participants (1.1%) were excluded from the analyses because of incomplete questionnaires.
Figure 1 presents the time-use characteristics across the 4 clusters, across which mean sleep duration varied little. We labeled the clusters as follows:
Studious actives (22% of analytic sample): highest school and passive transport time, higher than average physical activity, low screen time; selected a priori as the reference category for Aim 2 analyses
Techno-actives (33%): highest physical activity, moderate screen time and passive transport, lowest school time
Stay home screenies (23%): highest screen time, lowest passive transport and domestic/work time
Potterers (21%): moderate screen, domestic/work, passive transport, school time, low physical activity
Associations of Time Use and HRQoL (Aim 2)
Visual inspection of plots of marginal means and CIs showed clear differences in HRQoL between the “studious actives” and “potterers” (Fig 2). Analyses of variance indicated that the mean PedsQL Physical (F:13.5) and Psychosocial (F:11.3) Health Summary scores were unlikely to be the same between the clusters (both P < .001), with weak evidence that CHU9D scores also differed (F:2.54, P = .06). The reference studious actives had the highest HRQoL across all measures. Adjusted and unadjusted regression results were similar for all analyses (Table 3). Compared with the studious actives, the potterers scored ∼0.4 to 0.5 SD lower (ie, poorer) on Physical and Psychosocial Health Summary scores and slightly lower on the CHU9D (0.13 SD). “Stay home screenies” also scored marginally lower on Physical and Psychosocial Health Summary scores compared with the studious actives. There was little statistical support for the “techno-actives” having poorer HRQoL than the studious actives, although all their HRQoL scores were slightly lower.
This novel population-based study examined the association between person-centered time use and multiple measures of HRQoL in young adolescents. We identified 4 robust time-use clusters that were similar in boys and girls. Although physical activity, study, and screen time were all important in differentiating the clusters, sleep unexpectedly did not discriminate. Two groups differed substantially in HRQoL scores, with studious actives (ie, more physical activity, low screen time) having scoring 0.4 to 0.5 SD higher on PedsQL Physical and Psychosocial Health Summary scores than potterers (ie, low physical activity, moderate screen time). The remaining 2 clusters showed intermediate values that were not substantively lower than those of the studious actives.
Interpretation in Light of Other Study Findings
We discuss our findings in light of the only comparable study.7 Our identified clusters were similar in boys and girls, unlike Hunt’s sex-distinct clusters in older adolescents,7 suggesting that sex differences in time use may become more distinct by age 15 to 19 years. Hunt found little difference in HRQoL between time-use groups, with the only notable difference observed in HRQoL for girls in the “moderate study/higher leisure” group, who scored better than the “higher study/lower leisure” group. It may be that Hunt et al’s7 31-item time diary or latent profile analytic methods were less sensitive to capturing subtle differences in time use than our validated 259-item MARCA and compositional cluster analysis, or simply that our larger sample size yielded greater power.
Similar to our findings, studies examining associations between overall time use and well-being or functioning suggest that adolescents who undertake a variety of activities (eg, study, physical activity, screen time) have better health than those who spend most of their time in leisure activities with less physical activity, similar to our potterers.8–11 Our results strengthen the case made from studies examining individual activities that high physical activity is associated with better health and high screen time with worse health, because our compositional cluster analysis accounted for variation in all time elements.12,13
Our study indicated that, compared with the healthiest studious actives, potterers showed the poorest HRQoL by 0.1 to 0.5 SD. Yet there were no meaningful differences in HRQoL between the other groups. This suggests complexity in the interactions between time use and HRQoL. For example, when the techno-actives and studious actives are compared, it is possible that the additional time the techno-actives spend in physical activity counteracts the health-compromising effect of more screen and less school-related time, yielding similar HRQoL.
One surprising finding was that duration of sleep was similar across our clusters. This finding does not necessarily indicate that sleep is not associated with HRQoL, merely that sleep does not differentiate between clusters in this data set. This could be because cluster analysis prioritizes factors with greatest dissimilarity between groups, when sleep did not vary as much as other factors. However, it did mean that we could not refute or support the variable-centered studies that have reported more or adequate sleep to be strongly associated with better HRQoL.14,15
Weaknesses and Strengths of the Study
Our findings may not generalize to children from more disadvantaged neighborhoods because this group was underrepresented, although the sample still covered a wide social and geographic range. This limitation may be important because these children may have different time-use patterns and are more likely to experience poorer HRQoL.40 Although the MARCA has proven excellent test–retest reliability and good convergent validity, potential errors associated with self-report instruments include recall bias and social desirability bias. The MARCA captured 2 to 3 days of time use, which gives a better indicator of usual patterns than a single day but is less robust than if based on, say, 7 or 10 recalled days and may not capture seasonal variations. Despite any noise that these limitations may have introduced, we were still able to detect meaningful reductions in HRQoL across the time use clusters.
Identifying patterns with compositional time data entails making judgments, for example on which outliers to use and whether to use latent class instead of cluster analysis, that could influence the groups that emerge.41 Our procedures were informed by the best evidence on compositional techniques, and the clusters formed meaningfully different groups that reflected some expected differences in time-use patterns (eg, physical activity and screen time). Furthermore, reclustering with random samples in this data set yielded similar cluster solutions, indicating robustness. Residual confounders may arise from endogenous factors (eg, how adolescents felt about doing each activity) that we did not account for, but there is little evidence to suggest that these are important missing factors.
Finally, the cross-sectional nature of the study meant that causal directions could not be determined. Omorou et al12 suggest that better HRQoL predicts more time spent in physical activity, and achieving recommended physical activity level also drives better HRQoL. However, low HRQoL appeared to be a consequence (rather than driver) of a sedentary lifestyle, suggesting the high complexity in these associations.12 Nonetheless, our results did suggest that physical activity was important to HRQoL, because potterers who spent little time in physical activity had the lowest HRQoL scores.
Major strengths include the use of a comprehensive validated 24-hour time diary (yielding similar summary statistics to the Australian 2007 National Children’s Nutrition and Physical Activity Survey42) and the first cluster analysis with compositional input to capture time use as a whole, which is more accurate and valid than assuming time spent in different activities is independent. Our validated HRQoL profile and utility measures allow our findings to be considered within a health economics framework. It also increases confidence in our findings that the 3 HRQoL indicators (PedsQL Physical and Psychosocial Summaries and CHU-9D), despite their different recall time windows and instrumental structures, showed consistent relationships with the time use clusters. We adjusted for multiple potential confounders, including standardized BMI and pubertal status (which could affect both time use and HRQoL), although this adjustment had little impact on estimates or conclusions. Other strengths include our well-characterized participant selection process and the large population-derived sample.
Our study found distinct time-use clusters in healthy Australian young adolescents that show associations with HRQoL. When all other activities are taken into account, time-use patterns characterized by low screen time seem to be associated with the best HRQoL and patterns characterized by low physical activity with the worst HRQoL. Next steps would be to examine these same clusters against a range of physical health phenotypes and longitudinal outcomes as they become available in this cohort. If time-use patterns in early adolescence predict differential future HRQoL, this finding would strengthen causal inferences and the case for developing interventions based on time use. It will also inform clinicians and parents about the best time-use patterns for their children as they enter adolescence, when healthful lifestyles that can influence their future HRQoL are consolidated.
Prof Melissa Wake, Dr Kate Lycett, Dr Tim Olds, Ms Dorothea Dumuid, Mr Josh Muller, and Dr Monica Wong 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. This article uses unit record data from Growing Up in Australia, the LSAC. 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 those of the author and should not be attributed to the Department of Social Services, the Australian Institute of Family Studies, or the Australian Bureau of Statistics. Some study data were collected and managed via Research Electronic Data Capture electronic data capture tools, a secure, Web-based application designed to support data capture for research studies.
- Accepted March 28, 2017.
- Address correspondence to Melissa Wake, MD, Department of Paediatrics, Child and Youth Health, Auckland City Hospital, Grafton, Auckland 1023, New Zealand. E-mail:
FINANCIAL DISCLOSURE: The National Health and Medical Research Council supported Prof Wake (Senior Research Fellowship 1046518), Prof Gold (Early Career Fellowship 1035100), and Dr Mensah (Career Development Fellowship 1111160) in this work. Ms Dumuid is supported by an Australian Postgraduate Award, and Prof Wake is supported by Cure Kids, New Zealand.
FUNDING: The Child Health CheckPoint study was supported by the Australian National Health and Medical Research Council (1041352, 1109355), the Royal Children’s Hospital Foundation (2014-241), Murdoch Childrens Research Institute, The University of Melbourne, National Heart Foundation of Australia (100660), Financial Markets Foundation for Children (2014-055), and Victorian Deaf Education Institute. Research at the Murdoch Childrens Research Institute is supported by the Victorian government’s Operational Infrastructure Program. The funding bodies did not play any role in the study.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
- Centers for Disease Control and Prevention
- Mekary RA,
- Willett WC,
- Hu FB,
- Ding EL
- Bartko WT,
- Eccles JS
- Metzger A,
- Crean HF,
- Forbes-Jones EL
- Wake M,
- Clifford S,
- York E, et al
- Stevens K,
- Ratcliffe J
- Pink B
- Zuzanek J
- van de Boogaart K,
- Tolosana-Degado R
- Aitchison J,
- Barceló-Vidal C,
- Martín-Fernández J,
- Pawlowsky-Glahn V
- Palarea-Albaladejo J,
- Martín-Fernández JA,
- Soto JA
- Hair J,
- Black W,
- Babin B,
- Anderson R,
- Tatham R
- Varni J
- Dumuid D,
- Olds T,
- Lewis LK; the ISCOLE research group, et al
- Desgraupes B.
- Hartigan J,
- Wong M
- von Rueden U,
- Gosch A,
- Rajmil L,
- Bisegger C,
- Ravens-Sieberer U; The European KIDSCREEN Group
- CSIRO Preventative Health National Research Flagship
- Copyright © 2017 by the American Academy of Pediatrics