Promoting active travel to school: a systematic review (2010–2016)

Background Interventions aiming to promote active school travel (AST) are being implemented globally to reverse AST decline. This systematic literature provides an update of AST interventions assessing study quality and theory use to examine progress in the field. Methods A systematic review was conducted to identify and analyse AST interventions published between 2010 and 2016. Seven databases were searched and exclusion criteria were applied to identify 18 AST interventions. Interventions were assessed using the Active Living by Design (ALBD) Community Action (5P) Model and the Evaluation of Public Health Practice Projects (EPHPP). Methods used to evaluate the effectiveness of each intervention and their outcomes and extent of theory use were examined. Results Seven out of 18 studies reported theory use. The analysis of the interventions using the ALBD Community Action Model showed that Preparation and Promotion were used much more frequently than Policy and Physical projects. The methodological quality 14 out of 18 included interventions were assessed as weak according to the EPHPP framework. Conclusion Noted improvements were an increase in use of objective measures. Lack of theory, weak methodological design and a lack of reliable and valid measurement were observed. Given that change is evident when theory is used and when policy changes are included extended use of the ALBD model and socio-ecological frameworks are recommended in future. Electronic supplementary material The online version of this article (doi:10.1186/s12889-017-4648-2) contains supplementary material, which is available to authorized users.


Background
Active school travel (AST) remains an important source of physical activity for children [1]. AST has been shown to provide benefits such as reduction in children's Body Mass Index that long-term leads to a reduction in obesity-related diseases [2], improvement in academic performance at school [3], and as part of a larger picture, reduction in car use benefitting the environment [4]. Compared with other forms of physical activity, AST has the additional advantage of being convenient and free of monetary costs [5]. However, there is evidence that AST has significantly declined over the past 30 years [6,7]. Studies investigating the reasons behind the decline in AST point towards increasing use of car transportation, change in social norms [8], and parental concerns about safety (e.g. abduction, traffic, crime, and strangers) [9] as key contributors to the decline, amongst other factors.
Behavioural change interventions have attempted to reverse the decline in AST. For example, a systematic review by Chillon and colleagues [10] identified 13 interventions reporting a trivial to strong positive impact on AST behaviour. However, opportunities for improvement in future studies were identified including measurement, methodology and use of theory in intervention design and/or evaluation. A review update to consider progress in the field is timely to extend understanding.
Systematic literature reviews offer two key benefits. Firstly, systematic literature reviews guarantee that a more reliable knowledge base can be developed without biases that can occur in narrative reviews [11,12]. Secondly, systematic literature reviews can inform policy makers and practitioners by reporting the effectiveness of interventions [12]. Therefore, the purpose of the current study is three-fold. First, we aim to conduct a systematic literature review and analysis of AST interventions published between 2010 and 2016. Second, we compare the results of our review with Chillon et al. [10] to assess whether significant differences in theory use, measurement and design are evident between time periods. Third, we assess the extent of theory use for AST interventions reporting theory.

Data source and search strategy
This study followed Chillon et al. [10] search terms and systematic literature review procedures (see procedures outlined in13, 14) to identify peer-reviewed journal articles reporting AST interventions published between 2010 and 2016. Seven databases (EBSCO All databases, Emerald, ProQuest All databases, Ovid All databases, ScienceDirect, Taylor & Francis, and Web of Science) were searched using the following terms: 1. active transport* OR active travel* 2. intervention* OR Randomi?ed Controlled Trial OR evaluation OR trial OR campaign* OR program* OR study OR studies 3. child* OR adolescent* OR parent* OR youth OR student* OR pupil* 4. school* The symbols '*' and '?' are used as wildcards to include possible plurals and American/British spelling versions of the relevant terms respectively. The search terms were determined by multiple experiments using different combinations of terms in database searches to maximize the likelihood of retrieving the most relevant results. The seven databases used in this review were selected as they include marketing-and health-oriented publications and these were consistent with databases reported in previous systematic literature reviews [13,14]. An additional file summarises the search strategy [see Additional file 1]. The numbers of articles retrieved from each database are shown in Table 1 below:

Exclusion criteria
A total of 1553 records were identified in the search. All records were downloaded and imported into EndNote. After removal of all duplicated records (n = 696), 857 unique records were then checked against the following exclusion criteria to remove unqualified records: 1. not peer-reviewed journal articles, ensuring that all included sources had been peer-reviewed. Other types of records such as magazines, conference proceedings, newspapers, and dissertations were excluded; 2. not in English; 3. not related to AST; 4. policy related articles; 5. review/conceptual articles; 6. articles containing only formative research; 7. medical trials; 8. articles published before 2010.
After application of the exclusion criteria 27 qualified records remained. In the following stage backward and forward searches were conducted including examination of all reference lists of the 27 articles and searching authors' names and websites, and intervention names in Google Scholar. A further 13 articles providing additional information about already identified AST interventions and one additional new intervention were identified. The process produced a total of 40 peerreviewed articles published between 2010 and 2016 reporting a total of 18 AST interventions. PRISMA guidelines [15] were followed to systemically analyse the articles and report our review. Figure 1 demonstrates the search process, and a full list of 40 papers for each intervention can be found in the Appendix.

Data analysis
The following data was extracted and analysed from the papers: 1. Intervention strategy. In line with the method employed in Chillon et al. 's review [10], the Active Living by Design (ALBD) Community Action (5P) Model was adopted to analyse the strategies used in the interventions. This framework consists of five components: 1) Preparation, which includes "developing and maintaining a multidisciplinary community partnership, collecting relevant assessment data to inform program planning, providing relevant training, and pursuing financial Total and in-kind resources to build capacity" [16] (p. 315); 2) Promotion, which refers to engaging the target audience with dedicated messages and materials; 3) Program, which refers to ongoing organised activities that aim to engage individuals; 4) Policy, which refers to rules or standards that are set to regulate behaviours; and 5) Physical projects, which refer to environmental changes that are made to remove barriers to physical activity. Additionally, reported theory use was extracted and analysed as it has been linked to enhanced intervention outcomes [17], and theory use in AST interventions was previously found to be lacking [10]. The framework of assessing theory utilization was used in previous systematic reviews [18,19]. The framework consists of four levels, namely 1) Informed by theory, which means theory was identified but no or limited application of theoretical framework was used; 2) Applied theory, which means several components and measures were applied in the study; 3) Testing theory, which means more than half the theoretical constructs were explicitly measured and tested, or there exists theory comparison; 4) Building theory, which means revising or creating theory by measuring, testing, and analysing constructs. 2. Intervention design and delivery. The Evaluation of Public Health Practice Projects (EPHPP) [20] was adopted to assess the quality of the interventions and ensure consistency in reporting with earlier research [10]. EPHPP was developed to provide research evidence to support systematic intervention reviews by outlining step-by-step guidelines [21]. EPHPP has been used in a wide range of content areas, such as chronic disease prevention [22], family health [23], and substance abuse prevention [24]. EPHPP assesses six aspects of interventions: selection bias, study design, confounders, blinding, withdrawals and drop-outs, all of which is synthesised to Fig. 1 The systematic review process calculate a global study rating. In EPHPP, each of the aspects are rated on a three-point scale, and the final global rating is based on the rating of the six aspects and identified as strong, moderate, or weak, based on the EPHPP guidelines [20]. 3. Evaluation methods and outcomes. In contrast to the methods reported in Chillon et al. [10], who calculated the effect size for each intervention using Cohen's d, in this study we only extracted the information about the methods used to evaluate the effectiveness of each intervention and their outcomes as reported in the papers. Although Cohen's d could be an indicator of the effect size of the interventions, the identified heterogeneity of outcome measures among interventions makes the effect sizes incomparable with each other.
All data were extracted from the articles by two independent researchers and the final data were compared and verified to ensure accuracy. Discrepancies were minor and were resolved by discussion with a third researcher. In order to compare our results with Chillon et al.'s [10], we adopted the Fisher's exact test to calculate the p value. Fisher's exact test has been widely used to compare differences among small samples [13], and can offer more accurate results than the conventional Chi-Square method [25].

Intervention overview
Forty articles were identified in this review, covering 18 AST interventions. Full intervention details can be found in Table 2.

Intervention strategy and theory use
Chillon et al. [10] noted "the studies generally failed to describe their theoretical frameworks" (p. 8), however they did not report whether each of the studies reported or adopted theories. In our review, seven out of 18 interventions reported using theory. The most commonly used theory was Social Cognitive Theory reported (n = 5) [29,30,34,39,42], followed by the Social Ecological Framework reported (n = 2) [34,35] and the Theory of Planned Behaviour (n = 2) [37,39]. Two interventions [34,39] reported using more than one theory. In terms of theory utilization level, there are two studies [29,39] informed by theory and two studies [35,42] which applied theory. For example, in the "Beat the Street" [42] intervention, the researchers introduced competition to win points if children walk to school, underpinning social cognitive theory and learning theory. Three studies [30,34,37] tested theory. For example, in the "Traveling Green" [37] intervention, the factors of the Theory of Planned Behaviour were measured and tested to explain active commuting. None of the studies built theory. All interventions were analysed using the ALBD Community Action Model (see Table 3). Three interventions included all five strategies from the Community Action Model [36,40,41]. Five interventions included four strategies, three of which did not implement policy [7,27,29], and two did not implement physical projects [35,38]. Fisher's exact tests were used to compare the strategies used in the Chillon et al. review [10] and our review and none of the 5Ps were significantly different.

Quality assessment
The quality assessment of identified interventions was next conducted using the EPHPP tool 1 (see Table 4). Two researchers independently assessed all relevant articles and only minor discrepancies were identified and later resolved by discussion with the third researcher. Fourteen studies were assessed as weak in the global rating. None were assessed as strong. Comparing with the results reported in Chillon et al. [10], in which all 14 included studies were assessed as weak, a minor improvement was observed with four studies in the current review evaluated as moderate [29,31,38,39].
None of the studies reported representative sampling methods, which resulted in weak scores in category Aselection bias. In terms of study design, three studies reported using randomised control trial design [30,32,34] and were therefore assessed as strong; 13 studies were assessed as moderate, with nine cohort analytic (two groups pre + post) [26,28,29,31,[35][36][37][38][39], three cohort (one group pre + post) [7,41,42], and one interrupted time series design [33]. It is noteworthy that study [33] self-identified as quasi-experiment with pre-and posttests, measuring effects during the intervention, therefore the study was classified as interrupted time series. Two studies were rated as weak, including one longitudinal study [40]. Although one study [27] self-identified as quasi-experimental design, no evaluation was reported and therefore the study was assessed as weak. In terms of confounders, six studies reported confounders and no major differences were found between groups, which resulted in a strong rating. The rest of the studies did not report accounting for confounders or had only one group in the design and were therefore assessed as weak.
None of the studies reported to be double blinded. Four studies were assessed as weak as neither assessors nor participants were blinded [7,33,35,42]. The rest of the studies were rated as moderate with either one-directional blinding or no relevant information reported. In terms of data collection methods, ten studies provided evidence of both reliability and validity and thus were assessed as strong [28-32, 34, 35, 37, 40, 41]. Four were assessed as moderate with either reliability or validity being reported [27,33,38,39], and five were assessed as weak as they did not report reliability or validity [7,26,36,42]. Regarding the drop-out rate, seven studies were assessed as strong with more than 80% of participants completing the studies [7,29,30,32,33,35,36]. The remaining studies were assessed as moderate (n = 4) [28,31,38,39] or weak (n = 7) [26,27,34,37,[40][41][42] due to either low completion rates or not providing enough withdrawal information.
Fisher's exact tests were used to compare the EPHPP components between the Chillon et al. [10] review and our review. Table 5 shows that apart from Data Collection Methods (p = 0.011) none of the EPHPP components, including the global rating were significantly different.

Discussion
The purpose of this review was three-fold. First, we aimed to provide a contemporary review of AST interventions (2010-2016). Second, we aimed to compare the results of our review with Chillon et al. [10] to track progress in the field. Our review indicated that several issues identified by Chillon et al. [10] continue today and that theory use is limited in AST interventions. Third, we assessed theory utilization in AST   interventions. We will focus our discussion on three key aspects, namely theoretical, methodological, and empirical.

Theoretical aspects
Previous research indicated that theory use in intervention design was associated with enhanced intervention outcomes [17], yet the extent of theory utilization had not been examined previously. In our review seven out of 18 studies reported theory use. Detailed examination identified that two were informed by theory, two applied theories, and three tested theories with examples highlighted in the results section. At the optimal level, theory should provide guidance on the constructs that become the strategic focus of a campaign. Moreover, the theory framework should be used to evaluate the intervention pre and post permitting comparisons of key theoretical constructs focussed upon to be made [43]. The importance of theory adoption and implementation in intervention design is advocated by many researchers [44][45][46]. Consistent with previous studies our results show that theory testing and building remains limited in AST. For example, Painter et al. [47], identified that 69.1% of health behaviour research used theory to inform a study, in 17.9% theories were applied, in 3.6% theories were tested, and only 9.4% of studies involved building/creating theory. We propose three recommendations for future AST intervention implementation and reporting. Firstly, future studies should use theory to inform intervention development, execution and evaluation, and detail theory use to facilitate its full comparative assessment across multiple interventions. For example, Schuster et al. [48] used the Theory of Planned Behaviour to gain insights to inform an AST intervention. Results of the study indicated that four variables were found to be highly important in distinguishing carers segments, namely distance to school, current walk to/from school behaviour, subjective norms and intentions to increase their child's walk to school behaviour. Given that theory can increase the effectiveness of interventions [19,47] extended application of theory in AST interventions is recommended. Research studies have been undertaken to systematically implement, assess, and report theory utilization in health promotion interventions, such as the UK MRC guidelines [49,50] and the four-step Theoretical Domains Framework [43], and these are recommended to guide future AST intervention design.
Secondly, the theories used in the interventions identified in our review, such as social ecological theory, social  14 Stockholm County Implementation [35] Interview* n/a AST(#); Policy(+) No intervention effect was found on AST.
17 Walking School Bus -Houston [30] Survey* Observation* and BMI PA(+); Attitude(+) n/a 18 Walking School Bus -Missouri [31] n/a Accelerometer* PA(#); Participation(#) n/a + = positive effect; − = negative effect; # = no effect; * = AST evaluation method cognitive theory and the Theory of Planned Behaviour, have been considered traditionally as behaviour explanation theories [51]. However, as the ultimate purpose of AST interventions is to change behaviours, predictive theories and model testing should be deployed in future to develop theories focussed on behavioural change. Among all three studies in our review that tested theory [30,34,37], theoretical examination depended on crosssectional regressions limiting understanding to explanation rather than causal understanding. AST interventions should embed predictive theory testing involving longitudinal design across multiple time points to simultaneously explore potential behavioural change determinants. This also requires researchers to focus on utilising more causal/predictive methods rather than variancebased explanation methods in future study design, which is consistent with calls to advance theory to examine behaviour change [52]. Social ecological theory, social cognitive theory and the Theory of Planned Behaviour were most frequently reported and this provides a rich avenue for future research. Lu et al. [53] notes that social ecological theory lacks sufficient specificity suggesting additional testing is needed [53] to establish reliable and valid measures. Individual focussed theories such as Theory of Planned Behaviour and social cognitive theory are limited overlooking structural factors (e.g. policy) which limits understanding of how behavioural change can be facilitated [54]. Therefore, we recommended that theories that were specifically developed in the AST context, such as the McMillan model [55] and the Ecological and Cognitive Active Commuting (ECAC) model [56], should be empirically explored in future AST interventions. For example, the ECAC model specifies three levels of determinants, namely policy, neighbourhood, and individual; that are correlated with AST, covering environmental, social, and psychological aspects providing a wider system view. The McMillan model has been shown to be effective among the general population [57] and young adolescents [58], whereas the ECAC model needs to be empirically tested.

Methodological aspects
The EPHPP framework was used in this review to assess methodological quality. Fourteen out of 18 studies were assessed as weak. Notably, selection biases, lack of double blinding, and not controlling for confounders were key issues identified in both the current and earlier review [10]. While selection biases arise from practical considerations [28,41] such as recruiting schools to participate in AST, making it difficult for researchers to control in all circumstances, the current study points to the need for large scale funding permitting optimal study design to be achieved. Issues such as controlling for confounders, on the other hand, can be implemented in most AST interventions. Use of statistical methods, such as case-matching sampling [59], MANCOVA [60], and multi-level modelling [61], are recommended for future AST interventions.
Due to the complexity and diversity of intervention aims, evaluation methods and outcome reporting, we were unable to make direct comparisons of effectiveness between reviews. Standardised outcome measures would permit comparisons and meta-analysis to deliver more detailed understanding in the future. In addition, we recommend that objective measurement methods should be carried out in future intervention designespecially given declining monetary cost of equipment (e.g. smart phones and wearable technology that can automatically capture data) [62][63][64]. Governments or other funding bodies need to call for more rigour in methodological design and measurement in future.

Empirical aspects
In line with the results reported in Chillon et al. [10], the current review confirmed the heterogeneity of included studies in terms of their length, sample size, and objectives (see Table 2 for details). However, analysis of the interventions indicated that significant room for improvement remains in terms of broader application of intervention activities. The analysis of the interventions using the ALBD Community Action Model showed that Preparation and Promotion were used much more frequently than Policy and Physical projects. Policy implementation and infrastructure improvements remained limited despite documented positive effects [27,29] indicating policy use may be a necessary condition for effectiveness [53,65,66]. Consistent with the theoretical utilization in AST, Physical programs seem to be effective in promoting AST (see for examples [29,31]). Our findings are consistent with Lu et al. [53]. In their systematic review, Lu et al. [53] found that social ecological theory is widely adopted and can explain factors preventing children's walking to school. We recommend that intervention designers should incorporate more school and local policies and infrastructural improvement such as crime prevention and traffic control in order to reduce the perceived risk of AST among parents, observed in many previous reviews [67,68]. Moreover, habit was not identified as a behavioural determinant in any of included studies although transport habit is an important factor in AST [69]. Future intervention designs should consider facilitating longterm support to convert occasional AST behaviour to a habitual behaviour.
Many of the studies embedded compulsory educational workshops and informational sessions into curriculum (e.g., [32,39,40]). Evidence shows that curricular-based interventions results in low attendance and are less effective, which may explain drop-out rates observed for studies employing educational workshops and informational sessions [70]. Therefore, we recommend that instead of educating schools, parents, and children using traditional curricular-based strategies, approaches with more audience engagement be adopted. For example, gamification has been drawing increased attention from intervention designers in recent years, and programs such as GOKA [71] and ONESELF [72] have been shown to achieve substantial audience engagement while delivering outcome effects. Future research should test gamification within AST interventions to extend understanding.

Limitations
This review has several important limitations, many of which represent opportunities for future research. The search parameters used in the current review limit the studies identified. For example, grey literature and studies not in the English literature were not included in this review. Further, all of the 18 interventions were carried out in developed countries, yet physical inactivity among children is a significant challenge in many developing countries [73] suggesting there is an opportunity to extend AST intervention testing geographically.
A range of outcome measures and methods (self-report and non-self-report) were used to assess AST interventions including attitudes, policy, physical activity, active school travel, BMI, knowledge and infrastructure. The diversity of outcome measures prevents metaanalysis from being undertaken. Different evaluation methods limit potential comparisons between interventions. Further, given physical activity self-report data varies when compared to objective measures (non-selfreport) despite high correlations with objective forms such as pedometers and diaries [74], biases must be acknowledged [75]. Moving forward, a unified and consistent approach in reporting AST intervention outcomes is needed to enable meta-analysis to be undertaken in future. Standardisation of outcome reporting would permit effect sizes to be calculated enabling comparison between interventions. Future research is recommended to determine whether there is a relationship between EPHPP quality levels and effect sizean understanding that would inform AST practice.
Meanwhile, the analysis presented in this paper is also limited to the information reported in sources identified in the search process. Employment of a standard reporting framework for AST intervention reporting warrants future research attention ensuring that quality assessments take practicalities into account. For example, full blinding procedures such as those advocated in EPHPP may not be feasible in local government and State funded interventions thereby making this assessment component redundant. Such endeavours may assist to standardise reporting and in turn enhance quality assessment exercises informing future intervention development.

Conclusion
This review has provided a detailed analysis of AST interventions published in peer-reviewed journals between 2010 and 2016. Following systematic literature review procedures' 18 AST interventions were identified and subsequently analysed. The main findings of our study are:

1) Theory utilization in AST interventions published
between 2010 and 2016 is limited. Where theory is used, interventions informed by theory and interventions that apply theory are much more common than theory testing and building. 2) Considering the ALBD Community Action Model, Preparation and Promotion were reportedly used much more frequently than Policy and Physical projects. Given that change is evident where policy changes are made extended use of the ALBD model is recommended (Preparation, Promotion, Program, Policy and Physical projects). 3) Using the EPHPP framework, 14 out of 18 interventions were weak, largely due to selection biases, lack of double blinding, and not controlling for confounders. 4) Finally, an increase in more objective assessment measures in AST interventions published between 2011 and 2016 was observed, in comparison to the rates reported in Chillon et al. [10].
Issues such as weak methodological design and lack of reliable and valid measurements continue to persist in reported AST interventions, all of which indicate opportunities for further improvements in terms of intervention effectiveness and evaluation.

Endnotes
1 It is noteworthy that six interventions have multiple published papers. As the EPHPP was designed to assess studies rather than interventions [20], and in order to minimise the discrepancies between papers reporting different components of interventions, only one paper reporting information relevant to the EPHPP framework was selected for each intervention. Excluded papers did not provide any information that would affect the EPHPP rating of the study. The papers used in the EPHPP assessment are marked in the Appendix.