Danish University Colleges Resemblance in physical activity intensities and physical activity behaviours within families with children Petersen, Therese Lockenwitz

(1)

Danish University Colleges

Resemblance in physical activity intensities and physical activity behaviours within families with children

Petersen, Therese Lockenwitz

Publication date:

2021

Link to publication

Citation for pulished version (APA):

Petersen, T. L. (2021). Resemblance in physical activity intensities and physical activity behaviours within families with children. [PhD, University of Southern Denmark]. University of Southern Denmark.

General rights

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.

• You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal

Download policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

(2)

activity intensities and physical activity behaviours within

families with children

Ph.D. thesis by

Therese Lockenwitz Petersen

Research Unit for Exercise Epidemiology Centre of Research in Childhood Health

Department of Sports Science and Clinical Biomechanics Faculty of Health Sciences

December 2020

(3)

Department of Sports Science and Clinical Biomechanics, Research Unit for Exercise Epidemiology, Centre of Research in Childhood Health, University of Southern Denmark, Denmark.

Submitted December 2020

Assessment committee:

Professor Per Kjær (chair of the committee), Department of Sports Science and Clinical Biomechanics, University of Southern Denmark, Denmark.

Director Peter Bentsen, Center for Clinical Research and Prevention (CCRP), Bispebjerg and Frederiksberg Hospital, Denmark.

Professor Paul Jarle Mork, Department of Public Health and Nursing, Norwegian University of Science and Technology, Norway.

Supervisors:

Professor Anders Grøntved, Department of Sports Science and Clinical Biomechanics, Research Unit for Exercise Epidemiology, Centre of Research in Childhood Health, University of Southern Denmark, Denmark.

Associate professor Peter Lund Kristensen, Department of Sports Science and Clinical

Biomechanics, Research Unit for Exercise Epidemiology, Centre of Research in Childhood Health, University of Southern Denmark, Denmark.

Professor Eivind Aadland, Department of Sport, Food and Natural Sciences, Western Norway University of Applied Sciences, Norway.

Randi Jepsen, PhD, Lolland-Falster Health Study, Centre for Epidemiological Research, Nykøbing F. Hospital, Denmark.

(4)

of Sports Science and Clinical Biomechanics, University of Southern Denmark (SDU), at the Lolland-Falster Health Study (LOFUS), Centre for Epidemiological Research, Nykøbing F.

Hospital, Region Zealand, and at University College Absalon. The PhD was financed by grants from University College Absalon and two scholarships, one granted from the Faculty of Health Sciences, SDU and one from Region Zealand. Steno Diabetes Center Zealand, SDU, Region Zealand, Nykøbing F. Hospital, and Edith & Henrik Henriksens Mindelegat contributed to running PhD costs and financing of the LOFUS data collection.

The foundation for my PhD journey was established in 2015 when I got involved in the selection and development of questions about physical activity (PA) for a household-based population study under planning. The Lolland-Falster Health Study (LOFUS) should include inhabitants of all age- groups living in Lolland or Guldborgsund municipalities from February 2016 through the following four years. I wrote the protocol for my PhD work and applied for funding, approvals, and university enrolment in 2016, partly while I was on maternity leave. In the beginning of 2017, the first funding for accelerometer measurement in LOFUS was achieved, and the device-based PA data collection could start including only families with children. Additional funding and loan of accelerometers from the National Institute of Public Health and the Centre of Research in Childhood Health

(RICH), SDU made it possible to expand the accelerometer measurement to all LOFUS participants from December 2018 and onwards. Following consultation with SDU and approval from the

LOFUS steering group, it was decided to use the dual accelerometer system Axivity AX3. I got involved in management of the data collection, including the development and continuous

adjustment of the logistics and procedures, teaching of the LOFUS staff, inclusion and instructions of participants, initialisation of accelerometers, and data download. The reduction of the raw accelerometer data was done by PhD, associate professor Jan Christian Brønd at SDU.

Along with the data collection and PhD courses, I started working on the first paper of this thesis, which is a systematic review. I learned that systematically searching, collecting, and assessing the literature is a hard and time-consuming but very educational work. Due to the time limit of my PhD, I could not await the termination of the LOFUS data collection in February 2020. Therefore, the

(5)

My PhD was established in a collaboration between SDU, University College Absalon, and LOFUS. The first year, I was involved in teaching and supervision of bachelor students at the Department of Sports Science and Clinical Biomechanics, SDU, and both the first and second year, I gave lectures to nurse students at University College Absalon. Besides, on most of my working days, I was located at LOFUS, which gave me the opportunity to be involved in and learn from every step of and detail in the data collection and to practice problem-solving on a day-by-day basis. It was interesting and educational to participate in discussions about ethical challenges and dilemmas occurring in the encounters between the research participants and the nurses, biomedical laboratory technicians, and secretaries who were the front personnel in booking of participants and the data collection in LOFUS. For me, it was a time-consuming, but very valuable experience to have hands-on the data collection throughout the whole process, which gave exceptional learning.

Also, my co-supervisor PhD Randi Jepsen worked at LOFUS, which gave me a unique opportunity to discuss ideas, challenges with the data collection, academic frustrations, methods, and the general research field on a daily basis.

SDU provided me with an academic environment including seniors and peers, with whom I could exchange scientific ideas and discuss methods and the general research field of and updates on child PA. My main supervisor Professor Anders Grøntved, one of my three co-supervisors Associate Professor Peter Lund Kristensen, and my close collaborator Jan C. Brønd contributed to the anchoring of my PhD project in their research group and introduced me to international networks.

Fellow PhD students gave me a sense of belonging and an arena for mutual support and exchange of experience.

During my PhD, I was lucky to have three periods of study abroad at the Department of Sport and Teaching at University of Applied Science in Sogndal, Norway. There, my co-supervisor Professor Eivind Aadland and his senior and junior colleagues formed a unique and including academic environment with a long experience in data collection and teaching in a rural area and research in the child PA field.

(6)

contribute to my learning in accordance with individual expertise. My PhD has been demanding and giving, and I would never have been without it.

(7)

years, I have learned how quickly life can change, and at the same time how strong we stand, when we are together. A lot of people have been involved in this journey, and they deserve a special thanks.

I will start thanking the participants. Your curiosity and willingness to try the “udvidet skridtæller”

made my research possible and challenged me to go further into the attachment method.

Furthermore, I would like to give a special thanks to my supervisor team, Anders Grøntved, Peter Lund Kristensen, Eivind Aadland and Randi Jepsen, what an amazing team, thanks to you all for contributing with your expertise and time. You all showed me collaboration at a high level and introduced me to three different research environments (RICH/ExE, LOFUS and HVL Sogndal) which I am grateful fore. Thank you for your support and for believing in me. Thanks to University College Absalon for giving me this opportunity.

This journey started because of Rolf Horne & Randi Jepsen. Thank you, Rolf & Randi, for seeing me and always believing in me. Librarian Peter Tværmose, thank you for your incredible help through the years. Also thanks to Liselotte Bang for your expertise and tremendous help with paper I.

My colleges at RICH and ExE, I am grateful for the way you all have welcomed me into the

department. Thanks to all of you for support and discussion through the years. A special thank goes to Jan C. Brønd. Your door has always been open for questions and discussion, as well as a talk about how life goes. I greatly appreciate your never-ending support and encouragement.

I would also like to thank the employees at LOFUS (my LOFUS family). Thank you for being dedicated and for providing me with invaluable help and support during the entire data collection.

A special thank goes to Sidsel and Anne Sofie, for discussion, support and not least cheering from the side-lines the last six month.

(8)

Mom, I don’t know how to express my gratitude for all the help you have been giving us the last three years. Your love and support are priceless. I am grateful for you always being there for Kamma and Frede.

My beloved children, Kamma & Frede, thank you for being you and insisting on that playing is a major priority in our life along with lots of laughter. Last but not least to my husband Anders, thank you for always being by my side, and for your support through the years.

I cannot express how grateful I am, but thank you to you all.

Therese Lockenwitz Petersen Nykøbing F., December 2020

(9)

Paper I

Petersen TL, Møller, LB, Brønd JC, Jepsen R, Grøntved A. Associations between parent and child physical activity: A systematic review.

International Journal of Behavioral Nutrition and Physical activity, May 2020.

DOI: https://doi.org/10.1186/s12966-020-00966-z

Paper II

Petersen TL, Brønd JC, Aadland E, Kristensen PL, Grøntved A, Jepsen R. Resemblance in accelerometer-assessed physical activity in families with children: the Lolland-Falster Health Study.

International Journal of Behavioral Nutrition and Physical activity, December 2020.

DOI: https://doi.org/10.1186/s12966-020-01067-7

Paper III

Petersen TL, Brønd JC, Aadland E, Kristensen PL, Grøntved A, Jepsen R. Resemblance in physical activity in families with children in time segments during the week: The Lolland- Falster Health Study.

(Submitted December 2020)

(10)

A complexity of physiological, psychosocial, familial, and environmental factors are associated with physical activity (PA) in children and adults. In the family context, family members may influence each other’s PA behaviours. Although numerous previous studies have examined the relationship between parents and children’s PA, a systematic review of the available evidence was missing, and specific research gaps should be addressed to advance the understanding in this area.

Therefore, the aims of this thesis were:

I) To systematically summarize the current research evidence on the relationship between PA levels of parents and children.

II) 1. To examine the degree of resemblance in PA within families with children and between parents and children and 2. to explorethe degree of resemblance across age of the child, the gender of parents and children, and the intensity and type of PA.

III) To examine the degree of resemblance in PA intensities and specific PA behaviours within the total family, among parents and children, among siblings, and among parents considering different time segments of weekdays and weekends.

Method

This thesis consists of three studies. Study I is a systematic review based on published studies, which were identified using electronic databases and manual searches of reference lists. Papers reporting on associations between objectively measured child PA and at least one measure of parental PA were included. The quality of the papers was assessed using a modified version of the ROBINS-I tool. For quantitative interpretation of the results across studies, we produced albatross plots for all studies combined and by age-groups (of children), gender of parents, gender of children, methodology of assessment of parental PA, and type of PA.

Study II and III were based on cross-sectional data from the Danish household-based population study “Lolland-Falster Health Study”. PA was measured using a dual-accelerometer system (Axivity AX3) for seven consecutive days. Accelerometer data were used to classify time spent in PA according to intensities and specific PA types. Households that provided at least four days of valid data (Study II: any days of the week; Study III: at least three weekdays and one weekend day)

(13)

from at least one child and at least one parent were included in the analyses. A linear mixed model regression analysis was used to determine the intraclass correlation coefficient (ICC) of family (random effect) for PA outcomes, adjusted for sex, age, parental education, and the interaction between sex and age. The analysis for Study III was restricted to three specific time periods of the week: 1) weekdays from 8.00 a.m. to 4.00 p.m., 2) weekdays from 4.00 p.m. to 10.00 p.m. and 3) weekend days from 8.00 a.m. to 10.00 p.m.). In Study II, we performed analysis on the total family and various parent-child combinations in relation to age of children and gender of parents and children. In Study III, we performed analysis on the total family, parent-child dyads, siblings, and parent-parent dyads.

Results

Study I: Thirty-nine studies were included in the review with sample size of parent-child dyads ranging from 15 to 1.267 (mean=353 dyads, median=299 dyads). The majority of studies were classified as having moderate, serious, or critical risk of bias. The albatross plot for all studies combined showed that the clear majority of studies observed a positive relationship between parent and child PA. The plot suggested an average magnitude of correlation across studies to be around 0.13, and the overall impression was that this was fairly similar across child age-groups and gender of parents and children. Studies using objective assessment of parent PA showed a stronger

relationship between parent and child PA compared to studies using self-report for parent PA (average magnitude of correlation around 0.16 vs 0.04). Further, dividing the studies into two groups based on risk of bias (low or moderate risk of bias vs serious or critical risk of bias), the plot suggested an average magnitude of 0.15 and 0.11, respectively. However, methodological

limitations were identified in many studies, which affected the size of estimated relationships. No clear evidence was found for the strength of relationship being dependent on type of PA outcome (total PA, moderate-to-vigorous physical activity (MVPA), steps); however, the relationship for light physical activity (LPA) appeared weaker.

Study II: The study sample consisted of 1,837 individuals nested in 605 families; 902 parents (58.5% females, 42.9 ± 7.1 years) and 935 children (55.0% girls, 11.0 ± 4.5 years, range: ten months to twenty-two years). In the analysis of within-family variation in PA, the ICCs across PA intensities and activity types ranged from 0.06 to 0.34. For the total family, we found stronger clustering in light/low intensity activities (LPA: ICC 0.22 (95% confidence interval (CI) 0.17; 0.28)

(14)

and sitting/lying: ICC 0.34 (95% CI 0.28; 0.40)) and for walking (ICC 0.24 (95% CI 0.19; 0.30)) than for higher intensities (e.g. MVPA: ICC 0.07 (95% CI 0.03; 0.14)). The ICC for biking was 0.23 (95% CI 0.18; 0.29). The analysis on parent-child dyads gave similar results. No interaction effects for gender and age were found (except for in biking).

Study III: The study sample consisted of 1,576 individuals nested in 523 families; 774 parents (57.9% females, 42.8 ± 7.0 years) and 802 children (54.2% girls, 11.1 ± 4.3 years, range: ten months to twenty-two years). Overall, the intra-family clustering for the total family was stronger during late afternoons/evenings of weekdays and during weekends (ICCs 0.11-0.38) than during mornings/early afternoons of weekdays (ICCs 0.02-0.19). The strongest clustering was found for siblings both in the mornings/early afternoons of weekdays (ICCs 0.08-0.33) and during late afternoons/evenings of weekdays and during weekends (ICCs 0.10-0.47) and for parent-parent dyads (ICCs 0.02-0.26 mornings/early afternoons of weekdays and ICCs 0.13-0.52 during late afternoons/evenings of weekdays).

Conclusion

In this thesis examining the resemblance in PA within families, we found positive associations between the PA of family members independent of age of children and gender of parents and children. The intra-family clustering of PA tended to be stronger during late afternoons/evenings of weekdays and during weekends than during mornings/early afternoons of weekdays. The strongest resemblance was found for siblings and parents, respectively.

However, a substantial proportion of the variability in family PA is unexplained by our findings and therefore, further research is needed.

(15)

Dansk resumé (Danish summary) Baggrund og formål

Et bredt spektrum af fysiologiske, psykosociale og miljømæssige faktorer er associerede med fysisk aktivitet (FA) for både børn og voksne. I en familiekontekst kan familiemedlemmer påvirke

hinandens FA adfærd. Tidligere studier har undersøgt forholdet mellem forældres og børns FA, men der mangler en systematisk gennemgang af den eksisterende litteratur på feltet. Desuden mangler der generelt mere viden om familiens betydning for FA. Derfor er formålene med denne afhandling, at:

I) Systematisk opsummere den eksisterende litteratur om forholdet mellem forældres og børns FA.

II) 1. Undersøge lighed i FA i familier med børn og mellem forældre og børn, og 2. undersøge graden af lighed ud fra børns alder, forældres og børns køn og intensiteten og typen af FA.

III) Undersøge graden af lighed i intensiteter og specifikke typer af FA i den samlede familie, mellem forældre og børn, mellem søskende og mellem forældre i specifikke tidsperioder på hverdage og i weekender.

Metode

Denne afhandling består af tre studier.

Studie I: Vi udførte en systematisk litteraturgennemgang. Litteratursøgningen blev foretaget via elektroniske databaser og manuel screening af referencelister. Studier, der undersøgte

sammenhænge mellem objektivt målt FA hos børn og mindst ét mål for FA hos forældre, blev inkluderet. Kvaliteten af studierne blev bedømt ved hjælp af en modificeret udgave af ROBINS-I værktøjet. For kvantitativt at sammenfatte resultater på tværs af studierne udarbejdede vi albatros plots ud fra præ-definerede børnealdersgrupper, forældres køn, børns køn, målemetode for forældres FA, samt type af FA.

Studie II og III er begge baserede på tværsnitsdata fra ”Lolland-Falster Undersøgelsen”, som er en dansk, husstandsbaseret befolkningsundersøgelse. FA blev målt ved hjælp af to accelerometre (Axivity AX3) i syv fortløbende dage. Accelerometerdata blev brugt til at klassificere tid brugt i forskellige intensiteter og typer af FA. Husstande med mindst fire dages valide accelerometerdata for mindst en forældre og et barn (Studie II: hvilke som helst ugedage; Studie III: mindst tre

(16)

ugedage og en weekenddag) blev inkluderet i analysen. En lineær mixed model regressionsanalyse blev brugt til at bestemme intraclass korrelationskoefficienten (ICC) for familie (random effekt) for FA-mål, kontrolleret for køn, alder, forældres uddannelsesniveau og interaktionen mellem køn og alder. I Studie III blev der foretaget én analyse for hver af følgende tidsintervaller: 1) hverdage mellem kl. 8:00 og 16:00, 2) hverdage mellem kl. 16:00 og 22:00 og 3) weekenddage mellem kl.

8:00 og 22:00. I Studie II udførte vi analyser for den samlede familie og forskellige kombinationer af forældre-børn ud fra alder på børn og køn på forældre og børn. I Studie III udførte vi analyser for den samlede familie, forældre-barn-par, søskende og forældre-forældre-par.

Resultater

Studie I: Niogtredive studier med 15-1267 forældre-barn par (gennemsnit=353 par, median=299 par) blev inkluderet i reviewet. Størstedelen af studierne blev klassificeret med moderat, betydelig eller kritisk risiko for bias. Albatrosplottet for alle inkluderede studier viste, at størstedelen af studierne fandt en positiv sammenhæng mellem forældre-FA og barn-FA. Plottet viste en

gennemsnitlig korrelationsstørrelse på cirka 0,13, og det overordnede indtryk var, at resultaterne af studierne lignede hinanden uafhængigt af børns alder og forældres og børns køn. Studier, der anvendte objektiv måling af forældre-PA viste en stærkere sammenhæng mellem forældre-barn FA end studier, som anvendte selvrapportering til måling af forældre FA (gennemsnitlig

korrelationsstørrelse på cirka 0.16 versus 0.04). Ved opdeling af studierne ud fra risiko for bias (lav eller moderat risiko for bias versus betydelig eller kritisk risiko for bias) viste plottene

gennemsnitlige korrelationsstørrelser på 0.15 versus 0.11. Størrelsen af de estimerede

sammenhænge kan dog være påvirket af metodiske begrænsninger i mange af studierne. Vi fandt ingen klar evidens for, at styrken af sammenhæng var relateret til type af FA-mål (total FA, moderat-til-høj FA, skridt), men associationen for let FA fremstod svagere.

Studie II: Udvalget bestod af 1837 individer fordelt på 605 familier; 902 forældre (58,5% kvinder, 42,9 ± 7,1 år) og 935 børn (55,0% piger, 11.0 ± 4,5 år, fra ti måneder til 22 år). I analysen af intra- familiær variation af FA varierede ICCerne fra 0,06 til 0,34 på tværs af FA intensiteter og

aktivitetstyper. Vi fandt den stærkeste lighed i familiemedlemmers FA i aktiviteter af let/lav

intensitet (let FA: ICC 0,22 (95% konfidensinterval (KI) 0,17; 0,28) og sidde/ligge: ICC 0,34 (95%

KI 0,28; 0,40)) og i gang: ICC 0,24 (95% KI 0,19; 0,30) sammenlignet med højere intensiteter (f.eks. moderat-til-høj FA: ICC 0,07 (95% KI 0,03; 0,14)). ICC for cykling var 0,23 (95% KI 0,18;

(17)

0,29). Analysen på forældre-barn-par gav lignende resultater. Vi fandt ingen interaktionseffekt for køn og alder (undtagen for cykling).

Studie III: Udvalget bestod af 1576 individer fordelt på 523 familier; 774 forældre (57,9% kvinder, 42,8 ± 7,0 år) og 802 børn (54,2% piger, 11.1 ± 4,3 år, fra ti måneder til 22 år). Generelt var

sammenhængen i den samlede familie stærkest påhverdage mellem kl. 16:00-22:00 og i weekender (ICCer 0,11-0,38) sammenlignet med hverdage mellem kl. 8:00-16:00 (ICCer 0,02-0,19). De stærkeste sammenhænge blev fundet blandt søskende både på hverdage kl. 8:00-16:00 (ICCer 0,08- 0,33) og på hverdage kl. 16:00-22:00 samt i weekenden (ICCer 0,10-0,47) og for forældre-forældre- par (ICCer 0,02-0,26 hverdage kl. 8:00-16:00 og 0,13-0,52 hverdage kl. 16:00-22:00 samt i

weekenden).

Konklusion

Denne afhandling undersøgte ligheder i FA indenfor familier, og vi fandt positive sammenhænge mellem familiemedlemmers FA uafhængigt af børns alder og forældres og børns køn. De intra- familiære ligheder tenderede til at være stærkere i tidsintervaller på hverdage kl. 16:00-22:00 samt i weekenden, sammenlignet med tidsintervallet fra kl. 8:00-16:00 på hverdage. De stærkeste ligheder i familien fandt vi mellem søskende og mellem forældre.

En væsentlig del af variationen i intra-familiær FA er stadig uforklaret, og derfor er der behov for yderligere forskning.

(18)

List of abbreviations

BMI Body mass index CI Confidence interval CPM Count per minutes DAGs Directed acyclic graphs

EPOC Elevated post oxygen consumption

Hz Hertz

ICC Intraclass correlation coefficient LOFUS Lolland-Falster Health Study LPA Light physical activity METs Metabolic equivalents MPA Moderate physical activity

MVPA Moderate-to-vigorous physical activity NMT Non-moving temperature threshold PA Physical activity

SD Standard deviation SES Socio-economic status VPA Vigorous physical activity WHO World Health Organization

(19)

1.0 Introduction 1.1 Physical activity

Physical activity (PA) is considered to be an important factor in the prevention of non-

communicable diseases in both children and adults (1-3), and PA contributes to overall physical and mental wellbeing (4, 5). Nevertheless, at a global level, the proportion of children and adults that do not meet the daily PA recommendations is concerning (1, 6, 7), and physical inactivity has been identified as the fourth leading risk factor for worldwide mortality by the World Health

Organization (WHO) (7).

According to the official Danish PA recommendations, children should execute at least 60 minutes of moderate-to-vigorous PA (MVPA) per day (8, 9), while adults should engage in at least 30 minutes of moderate PA (MPA) per day in order to obtain beneficial health effects (10, 11). PA levels above the recommendations lead to enhanced health benefits (12, 13). Self-reported data from The Danish National Health Survey 2017 indicated that 28.8% of the adult population did not meet the recommendation for PA (14). A recent study of global trends in PA among adolescents (aged 11-17 years) reported that the majority (84.5%) of the Danish participants were insufficiently physically active (15). Similarly, a Danish study on school children found that 74% of 11-15 year old children did not met the recommended PA levels (16).

PA is embedded in everyday life and covers every energy-demanding movement produced by the skeletal muscles ranging from light intensity to high intensity PA (17). The different PA intensities refer to the degree of effort a person must expend to perform an activity in terms of the required amount of energy expenditure (often given in metabolic equivalents (METs)) (18). PA includes different behaviours such as lying, standing, walking, running, and biking (19, 20). In a life course

(20)

perspective, studies indicate that especially PA during adolescence seem to track into adulthood, however stronger for boys than for girls (21). Other studies have revealed that life events, such as transition from preschool to school, getting a child, and getting married are potential factors for decline in individual PA (20, 22).

A multitude of multileveled, interrelated factors are potential determinants of individual PA behaviours (23). Ecological models can work as frameworks for complex health related problems, and King and Salmon have developed one such model specifically for PA (Figure 1) (23). The model illustrates influential factors on three levels. Perception of neighbourhood safety, access to recreational facilities, and urban planning policies are examples of potential determinants on the environmental/policy level. On the sociocultural level, King and Salmon included factors such as social norms, intra-family social support, and interaction with siblings and peers. Age, gender, socio-economic status, and barriers and beliefs are examples of potential determinants on the individual level included in the model (23).

(21)

Figure 1. King and Salmon`s ecological model of physical activity (21, page 189).

1.2 The role of family in PA

In the PA model by King and Salmon, intra-family factors are thus considered contributing factors to individual PA on the sociocultural level (Figure 1). Families with children are the objects under study in this thesis. Family is a social unit and an arena for interrelations (24). In Denmark, 37 different family constellations are registered in national statistics including one and two parent families, families with biologically and/or non-biologically related off-springs and siblings, and foster families (25). Interactions in the family environment are multidirectional and complex and affect the health-related behaviours of each individual family member, for instance their dietary habits, screen use, sports participation, and overall PA (26-31). Underlying mechanisms

contributing to regulation of individual behaviours could be intra-family routines, negotiations, resistance, and cooperation (24, 32). Everyday lives of families with children are carried out in many different settings, in which family members spend time together and time apart (32). PA within families includes both non-structured (e.g. playing, housework, and screen viewing) and structured activities (e.g. transportation, fitness, and sport), which request a variety of PA behaviours and PA intensities (33).

Within families, parents are assumed to play an important role in the development of children’s PA behaviours (34, 35). Parents are socialisation agents, who may transfer PA habits to their children.

Parents’ choice of parenting style (34, 36, 37), active role modelling (34, 38, 39), rules and

restrictions (40), facilitation of PA and co-participation (41), support for PA (34), and provision of PA equipment (42) are examples of parental influence impeding or facilitating specific PA

behaviours in children. Parent’s perceptions of factors such as safety in the local outdoor

(22)

(43, 44). Access to PA facilities close to the home such as basketball hoops and soccer goals and distance to school are other environmental factors potentially influencing child PA (42, 43).

Although the ecological PA model by King and Salmon (Figure 1) provides a framework for individual PA, it does not cover all potential determinants within the family context. Thus, genetics is not represented in the model, although it is assumed to explain 11-24% of similarity in PA among biologically related non-twin siblings (45) and 27-30% among twins (46, 47). A three-generation study including more than 1,500 monozygotic twin-pairs and their parents and grandparents demonstrated that resemblance in PA was stronger among same-generation compared to between- generation individuals (48).

1.3 Resemblance in PA in families with children

The majority of previous studies on intra-family resemblance in PA behaviour have focused on the parent-child association using different methods for PA assessment. Reported findings are

inconsistent (34, 35, 38, 39, 49) with some studies observing positive (50-53) and others observing no association (54, 55) in PA among parents and children. The reported PA outcomes vary between studies. Some have focused on the association between the total PA of parents and children (50), while others have examined VPA (55), MVPA (53), light PA (LPA) (52), or steps (51, 54). The first study using device-based measurement of the PA of both parents and children was to my knowledge published by Moore et al. in 1991 (50) and showed a positive association in total PA among parents and children. VPA has seldomly been used as PA outcome in parent-child studies, but Sallis and colleagues (55) investigated the association in self-reported parental VPA and

accelerometer assessed VPA of their daughters (age 12-18 years) and found no association. MVPA has been used more often, and for instance Lee et al. (53) found a positive association among

(23)

parents and children using self-reported data for parents and accelerometer data for children.

Hesketh et al. (52) observed a positive association between mothers’ self-reported LPA and children’s accelerometer assessed LPA. In pedometer studies, Jacobi et al. (51) found a positive association between parent and child steps, while Chiarlitti et al. (54) observed no association. To the best of my knowledge, no studies have studied intra-family resemblance of specific PA behaviours, such as sitting, walking, and running.

Previous studies have indicated that the PA levels of parents vary with the age of children, which may be explained by the child’s motor development stage and increasing independency (39). Using accelerometer data, Adamo et al. (56) found that mothers of young children (6 years) tended to engage in less PA compared to mothers of older children (6-11 years), whereas the contrary was observed for fathers. However, a review by Bellows-Riecken et al. (57) concluded that studies on age of children as predictor of parental PA were inconclusive. A review by Hesketh et al. (58) concluded that across socio-economic groups, having young children was time-consuming and worked as a barrier to PA in families.

Also gender may influence resemblance in PA within families with children. Most previous studies on parental influence on child PA have been conducted within mother-child dyads, which leaves a gap in the literature regarding the influence of fathers (59). However, based on ten studies,

Neshteruk et al. (59) concluded that the PA of fathers and children seemed to be modestly and positively associated.

Studies on the impact of having siblings on child PA have reported inconsistent findings; however studies using accelerometers or pedometers have given indications of increasing PA with the

(24)

number of siblings (60). Studies on similarities in PA among parents are few, but Chen et al. (61) found a positive association in overall PA among spouses using self-reports.

1.4 Measurement of PA

The use of many different methods for measurement of PA makes it a blurred field to navigate in, and the complex nature of PA makes it challenging to choose assessment tools for investigation of PA under free-living conditions (33). Studies may aim at examining PA in different domains such as school, occupation, transportation, leisure time, and household/domestic, and PA may be

expressed as duration, frequency, intensity, or type (17). Different research focuses call for different approaches (62, 63).

The tools most commonly used in studies on resemblance in parent-child PA have been self- or parent-reported questionnaires (34, 38, 49). Questionnaires differ in for instance complexity, time frame, and PA outcomes (33). However, the reliability and validity of questionnaires have been debated especially when used in children (64, 65), particularly because of lack of accuracy, overestimation, and insecurities about children’s ability to recall PA (66). Also when used in adults, questionnaires are known to be unprecise but still, they have shown better validity and reliability for determining general type, amount, intensity, and duration of PA in adults compared to children (67). Objective measurement methods, such as pedometers and accelerometers are able to capture non-structured activities and PA intensities with more accuracy than questionnaires, and as the technology and accessibility have evolved, device-based measurements have increased in the PA research field (33, 68, 69). Accelerometers are biomechanical movement sensors that record bodily movement by the changes in accelerations in up to three axis (70). Depending on the specific type of devise, accelerometers provide the option to evaluate PA both as intensity related to energy

(25)

expenditure and thus oxygen consumption and as mechanical behaviour expressed as time spent in different activity types (71). PA intensity and PA activity type provide two different but

complementary aspects of an individual’s PA behaviour. Despite the more precise quantification of PA using accelerometers compared to questionnaires, the methodological choices made in the processing of raw accelerometer data (e.g. epoch length, number of valid days, and wear time) are challenging and could affect given results (72).

1.5 Aims of the thesis

In summary, family may play a role for individual PA. Previous studies have more often focused on parental influence (e.g. support or parenting style) on child PA than on intra-family resemblance in actual and specific PA behaviours (35, 38, 49). Earlier studies on familial resemblance in PA have mainly focused on the parent-child relationship, but there is a lack of overview over this literature.

In addition, high-quality studies are needed to examine resemblance in PA between not only parents and children but also among siblings and among parents (44, 58, 73).

Therefore, the overall purpose of this thesis was to examine intra-family resemblance in PA within families with children. A systematic overview of the available evidence on association between parent and child PA was lacking, and specific research gaps in intra-family resemblance in PA should be further addressed to advance understanding in this area. Therefore, the aims of this thesis were:

1) To systematically summarize the current research evidence on the relationship between PA levels of parents and children. (Study I) (74)

(26)

2) 1) To examine the degree of resemblance in PA within families with children and between parents and children, and 2) to explore the degree of resemblance across age of children, the gender of parents and children, and the intensity and type of PA. (Study II) (75)

3) To examine the degree of resemblance in PA intensities and specific PA behaviors within the total family, among parents and children, among siblings, and among parents

considerering different time segments of weekdays and weekends. (Study III) (76)

(27)

2.0 Methods

This section provides an overview of the methodologies used in the three studies included in this thesis. Detailed descriptions are provided in appendices I-III.

2.1 Study I

The systematic review adheres to the statement for Preferred Reporting Item for Systematic-reviews and Meta-analyses (77, 78) and is registered in PROSPERO (CRD42019093462).

2.1.1 Data and methods

A search strategy combining control terms (e.g. MesH and EMTREE) and text words related to PA, parents, and children was developed, and a search for relevant literature was conducted in the following databases: PubMed, EMBASE (from 1947-April 2018 and from 1947 - week 15, 2018), PsycINFO, SPORTSDiscus, The Cochrane Library (Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials (CENTRAL), Cochrane Methodology Register) in March 2018. For identification of additional studies, reference lists of previous reviews in the field were manually screened (39, 59, 79). The full screening protocol was repeated for all

supplementary articles identified. Endnote^TMX8 was used to remove duplicates, and Covidence (www.covidence.org) was used as a management tool in the screening process.

Criteria for inclusion were: 1) Studies had to report findings for at least one parent/child dyad. The child/children should be between 0 and 17 years of age. ‘Parent’ was defined as a legal guardian of the child (e.g. biological parent or foster parent). 2) Studies must report associations between parents’ and children’s levels of PA. Child PA should be measured objectively using accelerometer or pedometer. Parental PA should be measured either by self-reporting or objectively using

(28)

pedometer or accelerometer. 3) Observational studies including cross-sectional data were included.

Experimental studies were included if they reported cross-sectional data regarding the control and intervention groups.

Studies reported in English, Norwegian, Swedish, or Danish were read in their original languages.

Two authors independently identified relevant studies through screening of titles and/or abstracts. If a study was eligible after screening of the title and/or abstract, the full text copy of the article was obtained. Further, the full texts of the articles were independently screened by two authors. Any disagreement over the eligibility of particular studies was resolved through discussion with a third author.

2.1.2 Data extraction and quality assessment

The following data were extracted from the included studies independently by two review authors:

author and year of publication, country of study, study design, sample size, age of children, family structure, measurement method of PA of parents and children, and reported associations for children’s PA. Furthermore, we developed a tool for quality assessment of the selected studies using ROBINS-I (80) as the starting point and modifying some of the questions to suit our study aim. Some questions, for example “Is the reported effect estimate likely to be selected on basis of multiple outcome measurements within the outcome domain?” and “Is the reported effect estimate likely to be selected on basis of different subgroups?” were used in their original form, while others were changed to fit our review, e.g. “Exposure measurement method listed in the study” was modified to “Is the self-reported PA measured validly and reliably by this method?”. The items chosen from ROBINS-I were those which best captured the internal validity of the included articles.

The quality assessment tool covered four domains; selection bias, information bias, risk of bias in

(29)

selection of the reported results, and risk of type 2 error. We used the ROBINS-I categories for the overall risk of bias judgement, which categorised the studies as having low risk, moderate risk, serious risk, critical risk, or no information (80). The tool was used for assessment of study quality and evidence synthesis.

2.1.3 Synthesis of results

Because of heterogeneity among the included studies, conducting a meta-analysis was not possible.

The included studies provided non-homogeneous effect estimates (odds ratios, correlation coefficients, and regression coefficients) and insufficient amount of information to compute a homogenous effect size across all studies. Therefore, albatross plots were produced to assist the interpretation of the results across studies (81). Albatross plots were made for all studies combined and by age group (preschool-aged versus school-aged children), gender of parents (maternal, paternal, unspecified), gender of the child (boy, girl, unspecified), methodology for assessment of parent PA (objective, self-reporting), and type of PA measure examined (steps, LPA, MVPA, total PA). An albatross analysis provides a quantitative approximation of each study’s effect size based on its sample size, direction of effect, and p-value (two-sided). Thus, sample size, p-values, and estimate of effect were extracted for each study to make the albatross plots. In studies not providing a p-value, we estimated it based on the sample size and size of effect (e.g. regression coefficient). If multiple p-values were reported in a study because of analyses of various PA outcomes (e.g.

weekday and weekend estimates), we combined the p-values for calculation of an overall p-value for the study to be used in the albatross plot (82).

In albatross analysis, the estimation of approximate effect size is generally based on the relationship between the difference between the observed effect size and its hypothesized value under the null

(30)

hypothesis divided by the standard error (the Wald test). This equals the Z-statistic, which can be compared to a standard normal distribution. Rearranging the formula, it is possible to estimate the approximate effect size from the sample size and p-value (81). Thus, the albatross analysis can provide an approximate estimate of the overall magnitude of association across studies. The albatross analysis has been compared with formal meta-analysis with fairly similar results of obtained summary estimates. Yet, it does not provide an overall test of significance, which is a limitation (81).

2.2 Study II & III

2.2.1 Lolland-Falster Health Study

The Lolland-Falster Health Study (LOFUS) is a Danish household-based population study that enrolled 19,000 participants aged 0-96 years from a disadvantaged, mixed rural/provincial area of southern Denmark between 8 February 2016 and 13 February 2020 (83). Using the unique civil registration number (84), randomly selected individuals aged ≥18 years and their household members, if any, living on one of the two Danish islands Lolland and Falster were invited to participate and contribute with data for many different research purposes. Entire households were allocated to either an invited group or to an uninvited, non-contacted control group. Participation in LOFUS was voluntary for each invited individual. LOFUS’ data collection encompassed a

questionnaire and a visit to an examination centre, where a series of physical examinations and collection of biological samples were conducted (83). At the end of the examination, a subsample of participants was asked to wear accelerometers. In the period from 1 February 2017 to 30 November 2018, the criteria for inclusion for accelerometer measurement was that at least one adult and one child from a given household should participate. From 1 December 2018 to 13 February 2020, all

(31)

participants in LOFUS were eligible for inclusion in accelerometer measurement. LOFUS participants who could not walk were excluded from wearing accelerometers.

2.2.2 LOFUS-sample for the present thesis

The design of LOFUS with inclusion of entire house-holds allowed for examination of intra-family resemblance in PA within families with children. Therefore, data for Study II & III consisted of a subsample of LOFUS (83). We used data collected between 1 February 2017 and 2 October 2019 and included all families with children (≤ 22 years) who provided accelerometerdata. In both studies, we operationalised family as at least one parent and one child aged 0-22 years belonging to the same household. Parent referred to a primary caregiver, which could be a biological parent, a stepparent, a foster parent, or any other legal guardian (85).

2.2.3 Sociodemographic information

Sociodemographic information was obtained as standard information in LOFUS using a self- reported questionnaire preferably administered electronically. Paper versions of the questionnaire was only provided on demand (83, 86). The questionnaire items used in LOFUS regarding

sociodemographic factors have previously been applied in the Danish National Health Survey (14, 87). Seven response options on civil status were dichotomised as married/cohabiting or

divorced/separated/single/widow(er). Educational level was assessed with thirteen response options, which we merged into three categories: 1) medium (3-4 years) or longer higher education (≥5 years), 2) short higher education (2-3 years) or vocational education, and 3) one or multiple shorter courses or no formal education. Occupational status was assessed by combining sixteen response options into three categories: 1) in work (e.g. employees, employers, or self-employed), 2) being a

(32)

student (e.g. in high school, college, or vocational training), and 3) being out of work (e.g. on social benefits or unemployed).

2.2.4 Anthropometry

Height was measured without shoes to the nearest centimetre using a stadiometer (83). For participants ≥18 years, bioimpedance was obtained (model: Tanita Body Composition Analyser BC-420MA III or Tanita Body Composition Analyzer DC-430MA), providing the body mass index (BMI). For participants younger than 18 years, weight was measured using an electronic scale (Tanita WB 150 SMA). For these, BMI was calculated as body weight (kg) divided by the height squared (m²) (BMI kg/m^-2). All anthromometric measurements were made at the examination site before the recording of PA.

2.2.5 Recording of physical activity

The complex nature of PA as a human behaviour is difficult to capture, and it is sensitive to

response bias when using self-reported questionnaires (88). In the LOFUS data collection process, it was therefore decided to implement accelerometers in addition to questionnaire items about

engagement in specific PA related activities. The use of accelerometers would also provide

opportunities to compare results to other observational population studies using accelerometers (89, 90) and to merge and harmonise data within consortium collaboration (91).

Recordings of the PA of LOFUS participants of all age groups were obtained using two Axivity AX3 accelerometers (92). The Axivity AX3 combines a micro-electro mechanical system (MEMS) accelerometer with light and temperature sensors in a very small case weighing eleven grams.

Further technical specifications of the accelerometers can be found elsewhere (92). The two

(33)

accelerometers were attached to the participant’s skin using adhesive plaster to enable full 24-hour recordings and to facilitate wear compliance (93). One was placed on the front of the right thigh, centrally between the hip and the knee, and the other was placed on the right musculus latissimus dorsi at the lower part of the lumbar curve at a minimum distance of one centimetre horizontally from the spine (Figure 2). Participants were instructed to wear the accelerometers at all times for seven consecutive days, including during sleep and water activities and to reapply the

accelerometers with plaster if they fell off. If possible, the entire household should wear the accelerometers simultaneously, but for some households it was not possible to fulfill this criterion due to practical matters, which gave non-concurrent recordings in some households.

Being asked to wear dual devices continuously for seven days may be anticipated as a burden and may thus affect the willingness to wear the accelerometers and the total wear time. Using two accelerometers instead of only one may increase the risk of device failure and thus the quality of the accelerometry data. However, the use of two devices gives several advantages, which may

counterbalance the risk of enrolling less subjects or obtaining less wear time. The accelerometer placed on the thigh provides data for a robust classification of specific PA behaviours such as sitting, standing, walking, running, and biking (94). The identification of lying is most optimally identified using the combination of the thigh and the back worn accelerometer (95).

(34)

2.2.6 Reduction of accelerometer data

The open source OMGUI software (version 1.0.0.30) was used for initialisation of the

accelerometers and download of the raw data (96). The chosen procedures for initialisation of the accelerometers (e.g. hertz (Hz), epoch length, and number of measurement days) and the

instructions of the research participants (e.g. weartime of 24 hours for seven consecutive days) are important for data quality due to the intermittent and sporadic nature of PA (72). However, there is no overall consensus about methods to process raw accelerometer data and thus, no single optimal method to apply (72). In particular, inconsistent use of cut-points for PA intensities and lack of consensus about using either the same or age-specific cut-points for samples including for instance both children and adults make it difficult to compare results across studies (97-100).

In the data reduction of the raw accelerometer data from LOFUS, PA intensities were determined by generating ActiGraph counts using 10 seconds-epochs from the raw acceleration measured at the back (101). An ActiGraph activity count is a metric used to describe PA intensity measured by an accelerometer. A count is calculated by summarising the acceleration over the given epoch length.

Summarising acceleration in the sense of physics provides velocity, however, acceleration recorded by accelerometers worn for instance on the hip, the wrist, or the thigh is determined by the internal

(35)

and external forces reacting on the body. Thus, count is not the exact representation of velocity, but it captures velocity to some extent (102). When accelerometers are worn on the hip, the PA

intensity processed from intermittent activities such as soccer and basketball become

underestimated (103). It has been speculated that this is caused by the band pass filtering introduced in the method generating Actigraph counts (101). However, a recent study indicated that this is not the case, as the band pass filter may actually improve the accurate identification of intermittent activities as vigorous (104). Alternative correction methods have been proposed for improving the assessment of intermittent activities of Actigraph counts such as the two regression method

proposed by Crouter et. al. (105-107). However, this method uses a 15 second epoch length and is thus not optimal for capturing the sporadic PA behaviors of children. Capturing accurate intensity assessment of sporadic and intermittent PA seems to require the consideration of the elevated post oxygen consumption (EPOC) between activity bouts performed consecutively with less than 10 seconds distance. Considering EPOC in the generation of Actigraph counts has been applied in a new method developed at University of Southern Denmark (104). This method accounts for the aspects of EPOC using an epoch length of 1 second and movement in all three axis, and the final output is provided in 10 second epochs. The results from this new method improves the intensity assessment of intermittent PA activities like basketball and also demonstrates an improved explained variance with respect to a cardiometabolic health outcome in children (108). Previous attempts to compare the use of triaxial versus uniaxial data with respect to agreement with PA energy expenditure in adults (109, 110) and PA levels in children (111, 112) have shown either small or no difference in size of agreement.

In Study II and III, we wanted to use different PA outcomes to be able compare our results to previous studies. The majority of earlier studies on resemblance in intra-family PA have used MPA

(36)

as their primary PA outcome (74). Engaging in PA of moderate or vigorous intensity has been suggested to improve fitness (113-116) and health in general (117) and is therefore of interest when examining resemblance in PA among family members. Moderate intensity is defined as 40% of an individual’s VO2max and in absolute terms of 3 METs (113). The 3 METs absolute moderate intensity level is commonly used in calibration studies for determining the corresponding

accelerometry output (97, 100, 118, 119). However, in a recent study by Arvidsson et al. (120), it was suggested that using 3 METs as an absolute threshold was not appropriate if a study covers subjects in the age range from 6-60 years of age, which applies to the LOFUS study. The ideal solution would be to derive individual cut-points, but this solution is not feasible in studies including a large number of participants. The metabolic and mechanical cost of walking at self- selected speed is similar across a large age range (121-123) and is performed at an intensity corresponding to 30-35% of an individual’s VO2max. Moreover, the metabolic cost of running is performed at an intensity >60% of VO2max, suggesting that running at any speed requires a vigorous effort. This suggests that a moderate intensity cut-point can be defined as the average counts for walking at self-selected speed irrespective of age. The vigorous cut-point is the counts threshold at which most subjects are considered running (the lower limit of the 95% confidence interval (CI) of counts per minute during running). Thus, in Study II and III, the time spent in different PA

intensities (LPA, MVPA, and VPA) was determined using age-specific cut-points established using an internally conducted validation study (104). This experiment included 133 individuals in the age range of 5 to 50 years. The individuals were divided into a pre-school group (5-6 years, N=29), a child group (9-11 years, N=35), an adolescent group (14-16 years, N=31), and an adult group (>18 years, N=38). In Study II and III, the following cut-points were used for MPA; 1680 for children aged 0-6 years, 3075 for children aged 7-12 years, 3522 for adolescents aged 13-17 years, and 3522 counts per minute for adults ≥18 years. For VPA the cut-points 3368, 5543, 5755, and 6016 counts

(37)

per minute were used for the same age-groups (104). The LPA cut-point was set to 100 counts per minute for all age-groups, which corresponds to earlier studies (124-126). Because the validation study did not include a cross-validation setting, the calculation of accuracy of classifying activity as LPA, MPA, or VPA using these derived cut-points on the same validation dataset were likely invalid (sensitivity and specificity would most likely be grossly overestimated). Thus, we refrained from calculating these. However, a previous validation study by Pate et al. (127) in 3-5 year old preschool children using a fairly similar approach identifying Actigraph counts cut-points for MPA and VPA through visual inspection of the count value corresponding to the VO2 level that best discriminated slow walking and brisk walking (the MPA cut-point) and brisk walking and jogging (the VPA cut-point), found identical Actigraph cut-points as our internally conducted study among 5-6 year olds. The Pate et al. (127) study also conducted a cross-validation of the derived cut-points and reported a sensitivity and specificity for the MPA cut-point of 96.6% and 86.2%, respectively, and 65.5% and 95.4% for the VPA cut-point, respectively.

The activity types for each LOFUS participant were determined using the method described by Skotte et al. using both the thigh and the back accelerometer (94). A two second data window was used to identify the activities with this method and due to the use of 50% overlapping windows, it provided a resolution of the activity type information at 1 second. Skotte et al. (94) have

demonstrated a very high degree of sensitivity and specificity with the identification of several activity types in adulst. Their method has also been evaluated with children aged three to 16 years, and the results demonstrated similar sensitivity and specificity as with adults (128). A median filter was applied to eliminate sporadic misclassification of activity types. Findings of multiple

assignments of different activity types due to the median filter were not allowed and were reduced to the most likely PA behaviour type. Thus, if running and walking were identified in combination

(38)

with standing, then running and walking were selected. As such, all PA data were interpreted into activity types. Furthermore, the category lying represents all lying activity including sleeping due to the limitation of accurate classification of sleeping by an accelerometer.

Valid wear time was identified by considering both acceleration and temperature. The raw

acceleration was band-pass filtered (0.1-4 Hz) and temperature low pass filtered (0.05 Hz) using a fourth order Butterworth filter (zero delay). A non-moving temperature (NMT) threshold was individually determined from the temperature recorded during movement. Periods of no movement (consecutive acceleration below 20 mg) for longer than 120 minutes were identified as non-wear, and shorter periods from 45-120 minutes were identified as non-wear if the average temperature was below the estimated NMT threshold. Periods of 10 to 45 minutes with no movement were only identified as non-wear if the average temperature was below the NMT threshold, and if the end of the period was within the expected period of wakefulness (6 a.m. to 12 p.m. (in Study II) or 6 a.m.

to 10 p.m. (in Study III)). Periods of active movement were identified as device transport (device moving but not worn by the participant) if the average temperature was below the NMT threshold (129).

The specific data reduction criteria for including a participant in the final PA analysis for Study II and Study III are presented in Table 1.

(39)

Table 1. Physical activity criteria for inclusion in the final analysis in Study II & III.

Study Wear time

Time restriction,

weekdays

Time restriction,

weekend days

Family Week- days

Weekend days II ≥8 6:00 a.m. to

11:59 p.m.

6:00 a.m. to 11:59 p.m.

At least one child and one parent from a household

≥4

III ≥8 8:00 a.m. to 3:59 p.m. &

4:00 p.m. to 10:00 p.m.

8:00 a.m. to 10:00 p.m.

At least one child and one parent from a household

≥3 ≥1

2.2.7 Statistics

Statistical analyses were performed using Stata version 16.0 (StataCorp, College Station, Texas, USA). Characteristics of the sample are presented as mean values ± standard deviation (SD) for continuous variables and as percentages for categorical variables. Scatter plots were created (of parent and child PA) to graphically explore linearity. With no implication of a non-linear

relationship between parent and child PA, mixed linear regression analysis was used to estimate the size of clustering of PA within families, within parent-child dyads, among siblings, and within parent-parent dyads (random effect). Because the size of the included families ranged from two to six members, and because the subgroup analyses included more than two participants, a mixed linear regression analysis was chosen to account for the dependency of each family member and the possible intra-family clustering of PA (130). Also, using a mixed linear model, we were able to account for factors such as age and sex by including these as fixed effects. Thus, the use of a linear mixed model provided us with a flexible model to estimate the degree of resemblance in PA among family members. The analyses were adjusted for gender, age, highest achieved parental education, and the interaction between gender and age (fixed effects). Our decision to adjust for these specific effects was theoretically supported by directed acyclic graphs (DAGs) based on a priori knowledge

(40)

about potential confounders or effect modificators in the relationship between the PA of family members. DAGs provide a graphical and theoretical aid to frame confounding and relationships between exposure, outcome, and covariates and are used to identify paths that should be measured and controlled to obtain unconfounded statistical estimates (131).

Based on the variance components of the estimated random effect (total family and combinations of parents-children, siblings, and parents), we calculated the ICC with 95% CI to determine within- family (or within parent-child dyads, among siblings, or within parent-parent dyads) versus

between-family (or between parent-child dyads, between siblings, or between parent-parent dyads) variations in PA. The ICC is a measure of similarity within a cluster (132). An ICC of 0.00

indicates no resemblance, whereas an ICC of 1 indicates perfect resemblance. The literature does not provide clear cut-points for the interpretation of ICCs as weak, moderate, or strong in studies like the present, but taking the complexity and multi-factorial determinants of PA behaviour into consideration, we decided to interpret an ICC >0.20 as indicating weak, 0.20-0.50 as indicating moderate, and <0.50 as indicating strong intra-family resemblance. We abstained from calculating CI for low ICC values due to a close-to-zero denominator problem in the calculation of 95% CI with the delta method.

The degree of intra-family clustering in PA was estimated for different combinations of family members across several PA outcomes in both Study II & III (Table 2).

(41)

Table 2. Overview of the analysis units and physical activity outcomes used in Study II & III.

Study Analysis units PA outcomes

PA intensities PA types II 1) The total family

2) Parent-child dyads

3) Parent-child aged 0-6 years 4) Parent-child aged 7-11 years 5) Parent-child aged 12-22 years 6) Father-daughter

7) Father-son 8) Mother-daughter 9) Mother-son

CPM LPA MVPA

VPA

Sitting Lying Sitting+lying

Standing Walking Running Biking

III 1) The total family 2) Parent-child dyads

3) Parent-child aged 0-6 years 4) Parent-child aged 7-11 years 5) Parent-child aged 12-22 years 6) Siblings

7) Parent-parent dyads

CPM LPA MVPA

VPA

Sitting Lying Sitting+lying

Standing Walking Running Biking

PA: physical activity, CPM: counts per minute, LPA: light physical activity, MVPA: moderate-to-vigorous physical activity, VPA: vigorous physical activity.

2.2.8 Ethics

Region Zealand’s Ethical Committee on Health Research approved LOFUS (SJ-432). The Danish Data Protection Agency approved LOFUS (REG-24-2015) and the LOFUS sub-project: Patterns and correlates of physical activity and sedentary behaviour in families on Lolland-Falster (REG- 147-2016). LOFUS is also registered in Clinical Trials (NCT02482896). Written informed consent was obtained at the LOFUS examination centre. Two content forms were used: one for participants aged 0-14 years to be signed by their custodians and one for participants aged 15 years. Except from the possible inconvenience of wearing two accelerometers for seven consecutive days, having to return the accelerometers to LOFUS, and a risk of local skin reactions to the devices or the plaster, no disadvantages in accelerometer measurement were expected. LOFUS did not return results of accelerometer measurement to the participants.

Danish University Colleges Resemblance in physical activity intensities and physical activity behaviours within families with children Petersen, Therese Lockenwitz

activity intensities and physical activity behaviours within

families with children

Contents