Determinants of Change in Physical Activity in Children 0–6 years of Age: A Systematic Review of Quantitative Literature

Background Understanding the determinants of children’s health behaviours is important to develop successful behaviour-change interventions. Objective We aimed to synthesise the evidence around determinants (‘preceding predictors’) of change in physical activity (PA) in young children (0–6 years of age). Methods As part of a suite of reviews, prospective quantitative studies investigating change in physical activity in children aged 0–6 years were identified from eight databases (to October 2015): MEDLINE, Embase, CINAHL, PsycINFO, Web of Knowledge, British Nursing Index, Applied Social Sciences Index and Abstracts, and Sociological Abstracts. Determinants and direction of association were extracted, described and synthesised according to the socio-ecological model (individual, interpersonal, organisational, community, policy). Results Forty-four determinants, predominantly in the interpersonal and organisational domains, were reported across 44 papers (six prospective cohort, 38 interventional); 14 determinants were assessed in four or more papers. Parental monitoring showed a consistent positive association with change in PA; provider training was positively associated with change in children’s moderate-to-vigorous PA only. Five (sex, parental goal setting, social support, motor skill training and increased time for PA) showed no clear association. A further seven (child knowledge, parental knowledge, parental motivation, parenting skills, parental self-efficacy, curriculum materials and portable equipment) were consistently not associated with change in children’s PA. Maternal role-modelling was positively associated with change in PA in all three studies in which it was examined. Conclusions A range of studied determinants of change in young children’s PA were identified, but only parental monitoring was found to be consistently positively associated. More evidence dealing with community and policy domains from low-/middle-income countries and about lesser-explored modifiable family- and childcare-related determinants is required. International Prospective Register for Systematic Reviews (PROSPERO) Registration Number CRD42012002881. Electronic supplementary material The online version of this article (doi:10.1007/s40279-016-0656-0) contains supplementary material, which is available to authorized users.


Background
By the age of 5 years, over one in five children are overweight or obese the UK and USA [1,2]. Obesity in childhood is associated with a range of unfavourable outcomes including type 2 diabetes, hyperlipidaemia and psychosocial problems [3], with obesity known to track and be associated with unfavourable outcomes in adulthood [4,5]. Early childhood is a period of rapid growth and development, and the preschool years (defined here as up to the age of 6 years) are therefore ideal to both prevent and reverse unhealthy weight gain, by establishing healthy habits and behaviours.
As a result, interventions aiming to effect positive dietary, physical activity and sedentary behaviour change have been developed to prevent or halt obesity in the preschool years [6][7][8][9]. However, with a few notable exceptions [10][11][12], many of these interventional studies showed small effects which are not sustained over time, or have no effect at all [6][7][8][9]. One difficulty in establishing the reasons for a lack of intervention success is that multiple behaviours are often targeted simultaneously [8,9]. However, as each health behaviour has an independent significant impact on children's health [13,14], it is important to establish the most important determinants of each individual behaviour, and therefore how they may differ across behaviours. The socio-ecological model (SEM) [15] is a commonly used framework for categorising levels of influence on behaviours [16,17], classifying them into five broad categories: individual; interpersonal, organizational, community and public policy. By grouping potential influences on behaviour in this way, commonalities and differences can be identified and subsequently used to develop more targeted interventions to effectively change children's health behaviours [18].
In addition to consuming a balanced nutritious diet, children up to the age of 5 years are recommended to engage in 180 min of physical activity daily [19,20]. Higher levels of physical activity are associated not only with decreased adiposity in preschool-aged children but also positively associated with motor skill development, psychosocial health and decreased cardio-metabolic risk prospectively [13]. Cross-sectional studies in older preschool-aged children (2 years and over) also indicate that increased physical activity is linked to better gross motor control [21] and improved social skills [22]. Yet despite the importance of physical activity for young children's health and development [13], studies suggest that young children do not engage in sufficient levels of physical activity [23].
In order to specifically increase physical activity in targeted interventions, it is important to establish which factors influence activity behaviour [24]. A number of systematic reviews have been conducted to examine the associations between cross-sectional factors ('correlates') and young children's physical activity [16,25,26]. A broad range of correlates have been investigated, including demographic, biological, environmental, social and psychological influences. Although conclusions about the influences on physical activity differ between reviews [25,27], there is a suggestion that familial influences [16,25,26], time spent outside [25] and elements in the physical environment [25,27] may be associated with increased activity in preschoolers. An additional review [28] that included cross-sectional studies and a small number prospective cohorts also suggests that home influences may be key for young children's physical activity. However, it is difficult to draw firm conclusions about causality from cross-sectional studies. It is therefore necessary to use evidence from both prospective and interventional studies as these provide the best evidence to establish the longitudinal predictors (or 'determinants') of change in young children's physical activity, and to aid understanding of how to effect positive behaviour change.
This systematic review is part of a suite of reviews exploring the determinants of obesogenic behaviours in children aged 0-6 years (focussed on fruit and vegetable intake, sugar-sweetened beverages and unhealthy diet intake, physical activity and sedentary behaviour) [29,30]. It aims to synthesise the quantitative literature from prospective and interventional studies to ascertain the determinants (a 'preceding predictor') of change in physical activity in young children. It also aims to establish which (modifiable) determinants are associated with change; at which levels of influence these factors operate (i.e. individual, family, childcare setting, community or policy level); and where gaps in the literature exist for future research.

Methods
The protocol for this review project has been described previously [29]. The International Prospective Register for Systematic Reviews (PROSPERO) Registration number is CRD42012002881. Following established criteria for the rigorous conduct and reporting of systematic reviews [31,32], this review was carried out in three stages [33,34]. One search (led by HM) was conducted to identify studies across all reviews; at the data-extraction stage, smaller teams led each of the reviews focusing on specific behaviours of interest [i.e. physical activity (review lead: KH), fruit and vegetable consumption (COM), and sugarsweetened beverages (VP)]. KH also conducted the search update specific to physical activity in October 2015.

Identification of Studies for Review
A systematic search, common to all reviews, was undertaken in August 2012. Four sets of search terms were used related to: the population; study design (capturing observational, interventional and review articles); outcome; and exclusion of clinical populations. An extensive scoping phase was conducted prior to implementing the full search to maximize sensitivity and specificity of included papers. This involved contacting experts in the field and identifying key publications to be included for each behaviour, with searches run to ensure that these publications were captured. An electronic search was conducted in eight databases [MEDLINE, Embase (via OVID), CINAHL, PsycINFO (via EBSCO), Web of Knowledge (via Thomson Reuters), British Nursing Index (BNI), Applied Social Sciences Index and Abstracts (ASSIA) and Sociological Abstracts (via ProQuest)]. Citations were downloaded into Endnote citation management software (Thomson Reuters, Philadelphia, PA, USA). Included papers were searched for additional relevant publications, as were relevant reviews. No language restrictions were placed on the search, but articles were limited to published full texts. An updated search was conducted in October 2015 to capture studies with outcomes relating to physical activity only, published in the interim period (Electronic Supplementary Material  Table S1).

Study Selection
In 2012, two batches of 500 titles and abstracts were screened for inclusion by the review leads (KH, VP, COM) and checked for fidelity by a fourth reviewer (CS). With less than a 5% discrepancy, each reviewer subsequently screened approximately 12,000 papers individually. For quality control, two random 5% samples (total n = 3600) were double screened by two additional reviewers (RL and EvS). All full texts were obtained and distributed for the behaviour-specific reviews to progress in parallel. Additional texts retrieved in 2015 were screened by KH and a subsample (15%) reviewed by EvS.

Inclusion/Exclusion Criteria
Articles were included if (a) they reported results from a longitudinal observational study, randomized controlled trial (RCT) or controlled trial (CT), (b) quantified a withinchild change in physical activity behaviour (as primary/ second outcome in interventions) and (c) assessed at least one potential determinant of change. Children had to be aged between 0 and 6 years at baseline, and studies assessing physical activity using objective or subjective measures were included. Exclusion criteria included: (a) clinical populations (e.g. children who were malnourished, had asthma, cerebral palsy, cystic fibrosis, autism, etc.), (b) non-human studies, (c) quantitative cross-sectional studies, (d) qualitative studies, and (e) laboratorybased studies (e.g. validation studies).

Quality Assessment
For descriptive purposes, a quality appraisal of each of the included studies was conducted focusing on internal and external validity using assessment criteria adapted from those used previously [34,35] (Electronic Supplementary  Material Table S2). Criteria included: sample representativeness, size and retention, use of objective exposure and outcomes measures, appropriateness of analysis strategy, and randomisation method for RCTs. Scores out of 6 (or 7 for RCTs) were allocated and categorised accordingly (high quality: C5; medium: 3-4; low: 1-2).

Data Extraction
All full texts identified for inclusion were read by KH, and double screened for inclusion by EvS. For relevant papers, data were extracted using a standardized form. Data extracted included fırst author; publication year; country; study design, setting and population; and baseline descriptive characteristics. Data were also extracted about physical activity measurement and outcome; potential determinants; method of analysis; duration of follow-up; loss to follow-up; and results. All outcome measures used in prospective and interventional studies [e.g. percentage time or minutes spent at differing activity intensities (i.e. light (LPA), moderate (MPA), vigorous (VPA), moderate to vigorous (MVPA) or total activity (LMVPA)] were extracted. However, in some studies, activity was only assessed during specific periods (i.e. at weekends, during recess). In an attempt to standardise findings across studies, where more than one physical activity outcome was reported, we report total physical activity/counts per epoch (given current guidelines for young children's activity [19,20]), followed by MVPA, LPA and MPA/VPA. For interventional studies, each of the described elements targeted in the intervention (e.g. parental knowledge, parental modelling) were extracted as potential determinants of change in physical activity. For each determinant, the smallest included sub-sample was considered for extraction (e.g. if stratified by sex). Where results were stratified by specific times of the day, results for the largest time periods were reviewed and extracted. For longitudinal studies, the latest data available before the children were 6 years old were included; where two or more papers reported on the same study sample, both were included if they reported determinants associated with different outcome measures. For interventional studies, we assessed the difference in physical activity between control and intervention groups over time to classify determinants, as this provided evidence of factors targeted in interventions (i.e. determinants) which were associated with change. Where possible, results of multivariable rather than univariable models were included.

Data Synthesis
Narrative data synthesis was undertaken for all studies. Due to the heterogeneous nature of included quantitative studies and the physical activity outcomes used, meta-analysis was not appropriate. Each extracted determinant was scored based on direction and strength of evidence: '-' significant decrease in physical activity; '0' no significant association/effect or '?' significant increase in physical activity. Evidence from cohort and interventional studies was weighted equally, as both provide prospective determinants of change in physical activity behaviour. As per previous reviews [16,17,36], consistency across studies for any given determinant was then summarized according to the following metric: '0' (no association) if supported by 0-33% of individual studies; '?' (indeterminate/possible) if supported by 34-59%; and '?' or '-' if supported by 60-100%. Where four or more studies reported on a potential determinant, double indicators were used (e.g. '00', '??', '??' and '--') to indicate greater levels of evidence and therefore confidence in findings. Determinants, study score and consistency across studies were then presented according to the SEM (individual, interpersonal, organisational, community and policy) [17,36].

Results
A total of 37,686 (full review) and 3652 (physical activityspecific update) references were retrieved in 2012 and 2015 respectively, of which 220 were read in full and 44 papers included for review (representing 42 study samples: four prospective cohort and 38 interventional studies, see Fig. 1). A descriptive summary of the included study samples is presented in Table 1; study-specific information is provided in Tables 2 and 3.

Overview of Prospective and Intervention Studies
A total of 44 potential determinants of change were reported (Table 4) across papers. The same cohort study [Children Living in Active Neighborhoods (CLAN) [59]] was described in three [60][61][62] of the six prospective papers. One paper describing this study contributed all 16 determinants identified across prospective studies in intrapersonal, interpersonal and temporal domains. This paper predominantly reported on determinants relating to parental influence on change in physical activity. The 38 interventional studies targeted 28 potential (modifiable) determinants at intrapersonal (n = 6), interpersonal (n = 10), organisational (n = 10) and community levels (n = 1). No determinants at the policy level were identified across included studies. Of the 38 interventional studies, 27 (68%) were classified as multi-level [11, 42, 44, 46-48, 50-52, 54-56, 58, 63-76]; these most commonly targeted individual/interpersonal (i.e. children, parents, teachers) and organisational (i.e. preschool/ home environment) factors. Of these, 11 multi-level interventions (42%) effected a positive change in children's physical activity [42,44,46,47,54,58,63,65,66,69,72], though no clear effective combinations of components emerged. Across all prospective studies, positive effect sizes were generally small, with increases of less than 10% in total activity or MVPA from relatively low baseline levels.
Five determinants, across the intra-and interpersonal domains, namely sex (positive association in 2/5 studies); motor skill training (5/10); parental goal setting (2/4); parental social support (2/5); and increased time for physical activity (usually within the care setting; 4/11) showed no consistent association with change in physical activity. In the case of sex, evidence from the CLAN study served to highlight how determinants may differ within the same sample depending on the outcome used and time of follow-up [i.e. no association with counts per epoch at first follow-up [60] but a positive association between (male) sex and MVPA at second follow-up [61]]. For motor skills training [46,47,54,65,66] and increased time for physical activity [38,53,65,66] the majority of interventional studies that found a positive association with change in physical activity used objective measures.

Determinants Identified in Fewer than Four Studies
Determinants assessed in three study samples in the intra-/ interpersonal domains included child monitoring         week, y years, mo months, N/A not applicable, WIC women, infants and children, ? statistically significant positive effect of intervention, 0 no effect of intervention a Out of 6 for prospective and 7 for interventional studies Determinants of Preschoolers' Physical Activity 1365   [42,70,82], parental role-modelling [70,76,82] and maternal role modelling [44,58,61], with only the latter shown to be positively associated with change in physical activity in all three studies (one using proxy-reported physical activity [58]). In the organisational domain, increasing the number of care providers within the childcare setting was found to be positively associated with change in two (out of three) interventional studies [49,65]. Community awareness showed no association with change in children's physical activity [71,72,81]. Positive associations with change in physical activity were also found for providing additional opportunities for play within the home (two studies) [44,58] and sibling co-participation (one study) [61], and with structured physical activity [53] and lowering playground density [43] in one study each within the organisational domain.

Main Findings
This review is the first to synthesise evidence from longitudinal studies relating to the determinants of change in physical activity in preschool-aged children. Forty-four determinants were identified; determinants at the interpersonal and organisational levels were most commonly evaluated. Fourteen determinants were identified in four or more quantitative studies: parental monitoring showed a consistently positive association with change in physical activity. Provider training was positively associated with change in MVPA, but showed no clear association with physical activity overall. Of the remaining 12 determinants, a further five showed no clear association, and seven were  consistently not associated with change in children's physical activity. Moreover, maternal role modelling was positively associated with physical activity in three studies [44,58,61]. A range of modifiable family-and childcarerelated elements also showed positive associations with change in young children's activity in fewer studies. Where positive effects on change in physical activity were seen, they were often small in magnitude, particularly in studies reporting accelerometer-measured outcomes. Despite identifying a range of determinants that have been assessed, there appears to be little evidence of what elements effect positive change in preschoolers' physical activity.
Where determinants have shown no positive effect (e.g. child/parental knowledge), researchers should divert emphasis instead to other potentially influential determinants. Both parental monitoring and maternal role modelling may provide feasible and effective determinants of change; given the lack of longitudinal evidence from the community and policy domains, and with no evidence to date from developing countries, further exploration of possible determinants of change in these areas is also required.

Findings in the Context of Previous Research
As is also shown in cross-sectional studies [16,25], the association between the child's sex and change in physical activity [60,61,[77][78][79] was not consistent here. In general, boys' absolute levels of physical activity were reported to be higher than those of girls [61,79], suggesting that, regardless of change, boys may remain more active than girls over time. The aim of this review was not to assess whether a determinant was associated with increased physical activity over time, but rather if different levels of a determinant predict differences in change in physical activity over time. Sex is a good example of this: boys' physical activity may increase over time whilst girls' activity remains stable, or boys' activity may remain stable whilst girls' activity decreases. Although the data available do not allow us to explore the actual direction of change, this is an important consideration for future research. Based on current evidence and quality of measurement, boys appear to be more active than girls, but firm conclusions about the influence of sex on changes in young children's activity over time cannot be drawn. Determinants in the interpersonal domain were most frequently assessed. Only one determinant, parental monitoring, was consistently positively associated with change in physical activity in both prospective and interventional studies in this age group. This was operationalized in a range of ways by increasing parental awareness of the child's physical activity [66,69], including using log books [44] and pedometers [42]. Although evidence of parental monitoring effecting a positive change in physical activity prospectively in older children is sparse [85,86], cross-sectional evidence from a small sample of US children (n = 99) suggests that where parenting is permissive, parental monitoring may lead to increases in MVPA in children [87]. Evidence tends to suggest that parents tend to over-estimate their children's physical activity in general [88]. Yet conscious parental monitoring of the target behaviour may increase its salience, resulting in a greater number of prompts to be active and therefore greater subsequent physical activity.
Several reviews conducted previously suggest that elements in the preschool environment may be positively associated with children's activity [27,98]. Many interventional studies here specifically targeted the childcare environment, providing curriculum materials or modified elements within childcare settings, but no clear determinants were identified [11, 37, 39, 41, 43, 48-50, 53, 55, 56, 64, 71, 73, 75, 80]. Four of the interventional studies used variations of the same 'Hip-Hop-to-Health' intervention [11,55,56,73], targeting a range of elements in the childcare setting: only one [56] showed a positive sustained effect on accelerometer-measured activity in a predominantly African American population. This highlights that even with a consistent core intervention, factors including cultural variability, differing reported outcomes and intervention fidelity likely influence intervention success.
Although environmental childcare determinants showed inconclusive results, of 16 interventional studies incorporating provider training, eight noted positive increases in children's activity [38,44,46,47,49,53,54,66] and MVPA in particular. Interestingly, those interventions showing positive effects often incorporated few additional environmental elements, including providing additional curriculum materials [49,53]; they did however tend to include motor skill training [46,47,54,66], or parental elements [44,66], and/or allocate additional time for physical activity [38,53,66]. Introducing additional providers also led to increased physical activity in two out of three high-quality interventional studies, where external gym trainers [49] and professional coaches [65] led physical activity sessions.
Given the increasing amount of time children now spend in childcare, care providers feasibly play an important role in shaping children's health behaviours. It is not possible here to disentangle which elements of training resulted in positive physical activity change, but encouraging care providers to build on their skill-base and/or confidence in multi-component interventions may be important. Moreover, qualitative literature suggests that care providers perceive themselves to be both a positive [99][100][101] and a negative [99,102,103] influence on children's physical activity, yet no quantitative studies to date have specifically focused on care-providers own behaviour as a potential determinant. Doing so may be timely given providers believe they can influence children's activity and that young children should be active, but many are not aware of how much physical activity young children require [104].
Despite an obvious lack of observational research informing intervention development, the majority of interventional studies (68%) were classified as multi-level [11, 42, 44, 46-48, 50-52, 54-56, 58, 63-76], targeting determinants across a range of domains. Though these studies used notionally similar exposures, e.g. targeting children, their parents and changing the preschool environment, inconsistent results were seen. As with all multifaceted interventions, it is therefore difficult to tease out which components were effective and may explain in part why so few determinants were consistently associated with change in physical activity. Determinants across interpersonal and organisational levels may act synergistically or may counteract each other leading to null results. Although we attempted to determine how each interventional component influenced activity, no formal mediation analyses were identified and further exploration of how elements within an interventional result in positive change would be beneficial. For example, mixed-methods process evaluations may help to delineate determinants of children's physical activity and aid future intervention development.

Future Research Directions
This review highlights where research evidence and gaps exist. A large number of (interventional) studies have targeted determinants such as child motor/skills training; child and parental knowledge; provision of extra time for physical activity or curriculum materials; and provider training, with the studies overall showing no or indeterminate effects. Comparatively few studies have assessed a wide range of other determinants such as child/parent goal setting, and provider monitoring or social support. There is also a lack of studies assessing paternal determinants, and where this information is provided, studies tend to use maternal report. Only one determinant has been assessed in the community domain and none in the policy domain; no studies have been conducted to assess determinants in developing countries. Focusing research where such gaps exist will yield novel evidence, potentially prevent wastage of resources and promote physical activity change.
Moreover, little work has been conducted to explore how children's activity levels change from infancy to the preschool period, with only six studies including children aged 2 years or younger [57,58,69,70,83,84]. Questions remain about the optimal method for assessing physical activity in infants and toddlers [105]. Moreover, assessing physical activity across developmental periods may necessitate different measurement and processing protocols, complicating the assessment of change in physical activity. Nevertheless, given that the early years represent a period of rapid development and a crucial window for positive habit formation, it is important to determine for whom, how and why physical activity may change throughout early childhood, and whether behaviour and potential inequalities in health manifest and remain in later years.
Finally, determinants may be time or situation specific. Very few prospective observational studies have assessed determinants of physical activity change in young children. Including both prospective and interventional studies (and treating interventional components as determinants in the latter) allowed us to identify a wider range of factors that have been posited to effect change in physical activity. This review also indicates that determinants may differ within the same cohort depending on measurement method and follow-up period [i.e. in the CLAN study, there was no association between sex and counts per epoch at first follow-up [60] but a positive association between (male) sex and MVPA at second follow-up [61]]. Prospective studies allow assessment of change in behaviour over relatively long periods of time; interventions, with generally much shorter follow-up periods than prospective studies, may be able to capture more short-term fluctuations in behaviour. Both types of study also tend to assess differing types of determinants. Prospective studies have focused on child's sex and parental psychosocial and temporal factors, whereas interventional studies target child skill and knowledge, parental knowledge and behaviour, and elements in the preschool environment including care-provider training and provision of curriculum materials. Both types of study are therefore beneficial to establish whether a determinant is associated with behaviour change, and whether change is sustained over time. In combination, a more comprehensive picture of the determinants' landscape in children 0-6 years of age can emerge; this will ensure future research focuses on where gaps in the current evidence exist, whilst focusing work on areas where potential positive gains in changing young children's physical activity are most likely to be made.

Strengths and Limitations
This is the first systematic review, to our knowledge, to specifically explore determinants of change of physical activity in children aged 6 years and under across prospective cohort and interventional studies. Given that cohort and interventional studies offered the most appropriate design to extract determinants of change, our research strategy was restricted to prospective studies. We applied rigorous review methods and did not exclude papers based on language, but it is possible that all relevant publications may not have been included, as illustrated by the identification of an additional study at the data-extraction phase. As this review was restricted to published studies, publication bias cannot be discounted. One determinant (sex) was assessed in the same study twice and contributed by more than one paper [60,61]; however, in general, our methods reduced potential bias by lending more weight to determinants assessed in four or more studies. The inclusion of a range of study types and measures of activity is both a strength and a limitation of this review; studies using pedometers and questionnaires tended to report positive interventional effects. Studies also used differing accelerometer cut points and adjusted for differing covariates in regression models. This heterogeneity highlights how differing study methods may influence findings and intervention success. All studies were conducted in high-income countries and approximately half of the studies had small final sample sizes (n \ 50; studies = 15), which may have limited their statistical power to detect significant associations. Although we attempted to standardise outcomes across studies, five and 23 different outcome measures were used in prospective and interventional studies, respectively, preventing the use of meta-analysis here.

Conclusions
This review identified a range of predominantly interpersonal and organisational determinants of change in young children's physical activity; however, only parental monitoring of their child's physical activity emerged as a consistently positive determinant of change, with provider training positively associated with change in children's MVPA. Maternal role modelling was also positively associated with change in all three studies in which it was examined. Many determinants were explored in fewer than four studies, and multiple determinants were targeted within each interventional study. This heterogeneity in the determinants considered, and also in outcome measures used, limited the ability to identify consistent evidence for specific determinants. Future work should investigate potentially important lesser-explored or overlooked modifiable family-and childcare-related determinants; explore how determinants influence physical activity throughout the day and week; and deconstruct how the multiple elements within an intervention result in positive behaviour change. Assessment of determinants in the community and policy domains, in addition to studies conducted in developing countries, is also required. Such information will provide more robust evidence about the determinants of change in activity in preschool-aged young children, which is needed to inform the development of successful targeted interventions to increase activity levels in this population. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http:// creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.