Healthy urban environments for children and young people: A systematic review of intervention studies

This systematic review collates, and presents as a narrative synthesis, evidence from interventions which included changes to the urban environment and reported at least one health behaviour or outcome for children and young people. Following a comprehensive search of six databases, 33 primary studies relating to 27 urban environment interventions were included. The majority of interventions related to active travel. Others included park and playground renovations, road traffic safety, and multi-component community-based initiatives. Public health evidence for effectiveness of such interventions is often weak because study designs tend to be opportunistic, non-randomised, use subjective outcome measures, and do not incorporate follow-up of study participants. However, there is some evidence of potential health benefits to children and young people from urban environment interventions relating to road safety and active travel, with evidence of promise for a multi-component obesity prevention initiative. Future research requires more robust study designs incorporating objective outcome measures.


Introduction
Both globally and nationally, there has been increasing recognition of a need for action in building 'healthy communities' (Centers for Disease Control and Prevention, 2011;Edwards and Tsouros, 2006;National Institute for Health and Care Excellence, 2008;The Royal Town Planning Institute, 2009). Policy makers, urban planners and practitioners have a significant role in implementing and influencing policies to shape the urban environment in ways which enable people to live healthier lives. Historically, urban planning has focused on improving system efficiency or reducing environmental impacts (Handy et al., 2002). However, in more recent years there has been a focus on linking characteristics of the built environment with health behaviours and outcomes of the population.
There are increasing concerns that current 'lifestyles' in high income countries, particularly poor quality diet and sedentary behaviour, lead to chronic illnesses and health disparities which are socially and geographically patterned. Children can be particularly vulnerable as they have few opportunities to choose or change their environment. Systematic review evidence derived from cross-sectional or longitudinal studies have identified components of the built environment associated with physical inactivity (Sallis and Glanz, 2006), active travel (Pont et al., 2009), dietary intake (Sallis and Glanz, 2006), obesity (Dunton et al., 2009;Galvez et al., 2010), and mental health (Clark et al., 2007;Sinha and Rosenberg, 2013) in children and young people. Lack of sidewalks, distance to school or public open spaces, and density and availability of food sources are correlated with poorer physical health behaviours and outcomes (Dunton et al., 2009;Galvez et al., 2010;Pont et al., 2009;Sallis and Glanz, 2006). Worse mental health outcomes are associated with exposure to violence or crime in the neighbourhood (Clark et al., 2007;Sinha and Rosenberg, 2013). Although adaption of the built environment to overcome these factors may have the potential to improve health, robust intervention studies are required to provide evidence of a causal relationship and effectiveness.
A Cochrane review of built environment interventions for increasing physical activity in children and adults is yet to be published (Tully et al., 2013). However, in the protocol the authors argue that much of the previous evidence has been from crosssectional studies which demonstrate inconsistent associations between features of the built environment and physical activity, do not demonstrate a causal relationship and do not control for confounders such as more active people choosing to live in neighbourhoods that support physical activity.
In a recent systematic review of the impact of interventions to promote physical activity in green space (Hunter et al., 2015) the authors argue, given the significant investment by local authorities in maintaining and improving urban green spaces, there is a need to identify if investments are effective in increasing the use of such Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/healthplace spaces and whether there are public health benefits. Some evidence supported the use of built environment only interventions, but more promising evidence was found for physical activity programmes combined with changes to the built environment. However, the authors urged caution in interpreting the results because of the paucity of intervention-based research in this area. Furthermore, they highlighted the dearth of evidence in relation to children and adolescents.
The aim of the current systematic review is to examine evidence from intervention studies which involved changes to the urban environment and reported outcomes in relation to health related behaviours, and the physical or mental health outcomes, of children and young people.

Methods
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were followed throughout the design, conduct, and reporting of this systematic review (Moher et al., 2009).

Search strategy
A comprehensive search strategy to identify primary studies reporting interventions to the urban environment and the health outcomes of children and young people was developed by an experienced systematic reviewer (H.B.-F.) for the Medline database. This was refined following discussion with the second reviewer (S.A.). A combination of the following Medical Subject Heading (MeSH) terms indexed within the database, and relevant text words from previous systematic reviews, comprised the initial search strategy: 'Obesity'; 'Weight gain'; 'Weight loss'; 'Diet'; 'Dietary fats'; 'Exercise'; 'Physical activity'; 'Mental disorders'; 'Adjustment disorders'; 'Anxiety disorders'; 'Mood disorders'; 'Neurotic disorders'; 'Child'; Adolescent'; 'Child, preschool'; 'Infant'; 'Urban health'; 'City planning'; 'Urban renewal'; 'Environment design'; 'Public facilities'; 'Intervention studies'; 'Evaluation studies', and; 'Program evaluation' (Supplementary file 1). The indexing terms were modified to be applicable to other databases.

Data sources
The following biomedical, geographical, and transportation databases were searched from inception to 29 October 2014: Embase; Geobase; Medline; PsycINFO; Transportation Research Information Services, and; ISI Web of Science & ISI Proceedings. Searches were not restricted by date of publication. All abstracts were saved using reference manager Endnote X3.

Study selection
Intervention studies (randomised controlled trial, controlled trial, controlled before and after, before and after, interrupted time series) were eligible if they: included a change to the built environment; reported outcomes in relation to children and young people's physical or mental health and well-being, or health behaviours such as dietary intake or physical activity, or counts of active transport or park use; were undertaken in urban areas; were published in English, and; were undertaken in high-income countries using the World Bank classification (available from http://data.worldbank.org/about/country-classifications). Studies were excluded if they: focused on changes to the school or home environment rather than the wider public realm; were undertaken in rural areas or low-or middle-income countries, or; were not published in English. Conference abstracts, dissertations, letters, and books were not eligible for inclusion but were checked for relevant publications. Separate publications presenting results from the same intervention were reported together.
After duplicates were removed, all records were reviewed by one reviewer (H.B.-F.) to consider their relevance for inclusion. A random 10% sample of the records was independently assessed by a second reviewer (S.A.), with 'very good' inter-rater agreement (kappa ¼0.90) (Altman, 1991). Full text articles were retrieved and independently assessed for inclusion by two reviewers (H.B.-F. and S.A.). Disagreements were resolved by discussion. Reference lists and bibliographies from relevant primary studies, reviews, and intervention protocols were hand searched for additional primary studies not retrieved by the electronic search (H.B.-F.).

Data extraction
One reviewer (H.B.-F.) extracted and entered the following information for each study onto an excel spreadsheet: study characteristics (authors' names, publication year, location, study period, objective(s), participants, and intervention setting), and; characteristics of study design (intervention, change to urban environment, sampling strategy, data collection methods, analysis, and main findings). These were double-checked by another reviewer (S.A.) to ensure accuracy.

Quality assessment
Quality assessment allows the methodical appraisal and evaluation of primary studies and is an established feature of systematic reviews of randomised controlled trials. Due to the anticipated nature of built environment interventions, which may preclude incorporation of randomisation and blinding within the study design, we did not exclude studies on the basis of quality. Quality assessment was carried out primarily to highlight the risk of bias and the resulting uncertainty of the results reported. Eligible primary studies were appraised by one reviewer (H.B.-F.) using a validated tool for non-randomised controlled trials (Sterne et al., 2014).

Results
A narrative synthesis approach (Popay et al., 2006) to reporting the results was taken because of the heterogeneity of outcomes, population groups and interventions. Further, there was a lack of suitable data to calculate standardised effect sizes (Higgins and Green, 2011). Primary studies were grouped according to the main focus of the intervention and reported narratively. Included primary studies are summarised in Table 1; the study designs and main findings are shown in Table 2, and; the assessment of bias is in Table 3. Table 4 offers a simplified overview of the key results for different types of intervention with their respective overall risk of bias.

Study selection
Of 9686 records initially identified through the database searches, 7645 records were reviewed and 113 full text studies assessed for eligibility (Fig. 1). Of those full text studies excluded: 25 did not report a health behaviour or health outcome measure; 15 presented no data in relation to children or young people; 19 did not incorporate changes to the built environment; five were not intervention studies, and; two were not published in English. Fourteen systematic reviews were excluded following hand searches of reference and citation lists. A total of 33 relevant primary studies in relation to 27 separate interventions were included in  Davidson et al. 1991New York, USA 1983-1991 To evaluate the effectiveness of a community coalition to prevent severe injuries to children   Davidson et al. 1991New York, USA 1983-1991 To evaluate the effectiveness of a community coalition to prevent severe injuries to children  (Economos et al., 2007Folta et al., 2013) To test whether a community-based environmental change intervention could prevent undesirable weight gain in children at 1 and Quasi-experimental: Non-randomised controlled trial (one intervention site and two control sites) (Economos et al., 2007Folta et al., 2013) Objectively measured BMI (Economos et al., 2007 Baseline: 5940 potentially eligible students enroled in grades aged 6-8 years, 1696 (29%) consented to participate 30 schools-and the surrounding communityand homeenvironments Economos et al.
2013 Folta et al. 2013 Parent reported fruit and vegetable and 2 year follow-up (Economos et al., 2007 sugar-sweetened beverage consumption; number of organised sports and physical activities per year; walking to and from school (Folta et al., 2013) 1 year follow-up: Data available for 1178 (69.5%) children 2003(Chomitz et al., 2012 Quasi-experimental: Non-randomised controlled trial (one intervention site and one control site) (Chomitz et al., 2012) Chomitz et al. 2012 2 year follow-up: 1028 (60.6%) children  To describe the behavioural changes in children resulting from Shape Up Somerville (Folta et al., 2013) Student self-reported achievement of either moderate or vigorous physical activity guidelines (Chomitz V, et al. 2012) 2 year follow-up: 454 parents of children (Folta et al., 2013) To evaluate the Active Living by Design project implemented in Somerville (Chomitz et al., 2012) Baseline: 1098 (90%) middle-and 1382 (81%) high-school students 4 year follow-up: 926 (88%) middle-and 1125 (79%) highschool students Hendricks et al.   Davidson et al. 1991New York, USA 1983-1991 To evaluate the effectiveness of a community coalition to prevent severe injuries to children School Travel Plan programme  following implementation of the School Travel Plan programme   To evaluate the Canadian School Travel Planning intervention by examining changes in school travel mode and predictors of mode change (Mammen et al., 2014a) Quasi-experimental pre-post evaluation design without a comparison group (Mammen et al., 2014a) Proportion of students who changed to an Active Travel mode (Mammen et al. 2014a) Baseline and 1 year follow-up: Children aged 6 to 14 years old (number not provided) (Mammen et al., 2014a) 53 schools that participated in the School Travel Plan programme Mammen et al. (2014b) 1 year follow-up: 7827 of 24,893 (31.4%) families attending schools that implemented School Travel Plan programme (Mammen et al., 2014b) Proportion of parents who reported driving less (Mammen et al., 2014b) To evaluate the Canadian School Travel Planning intervention by examining child-, family-, and school-level characteristics (Mammen et al., 2014b) Quasi-experimental retrospective post-test assessment without a comparison group (Mammen et al., 2014b) McDonald et al.  Hospital surveillance data of injuries and death during the intervention period (1989)(1990)(1991) and the pre intervention period (1983)(1984)(1985)(1986)(1987)(1988) Poisson regression model During the intervention period, adjusted annual incidence rates of injuries in children aged 5-16 years decreased (relative risk:0.74, 95% CI: 0.62-0.89) compared with preintervention period. In the younger, nontargeted age group, no significant reduction in incidence occurred (  Objectively measured BMI at baseline, 1 and 2 year follow-up (Economos et al., 2007 Multiple linear regression, accounting for covariates and clustering by community 1-year follow-up time point: Average change in BMI z-score in the intervention community was À 0.10 compared to controls (95% CI: À 0.12 to À 0.09, po 0.001) (Economos et al., 2007) Economos et al. (2013) 2-year follow-up time point: Adjusted difference in BMI z-score change was À 0.06 compared to controls (95%CI: À 0.08 to À 0.04, p o 0.05)    (Boarnet et al., 2005b) Traffic signal improvement projects: 2 traffic signal improvement projects in school sites resulted in increased walking observations (Boarnet et al., 2005b) No observation of success for crosswalk and crosswalk signal or bicycle path improvement projects (Boarnet et al., 2005b) Crossing improvements: Adding crosswalks, installing in-pavement crosswalk lighting, and installing a pedestrian activated, "count-down" streetcrossing signal that warns pedestrians of the amount of time remaining to cross Traffic control: Installation of a traffic signal 1Boarnet et al.  Students: Children attending school on day of data collection (Mammen et al., 2014a) the morning (16.7% , n¼ 1118) and afternoon (17.1%, n ¼1211) periods at the 1 year follow-up time point (Mammen et al., 2014b) Parents: All parents with children attending schools implementing programme were invited to participate at 1 year follow-up (Mammen et al., 2014a(Mammen et al., , 2014b Infrastructure improvements and safety education were perceived by families as the most effective strategies implemented (Mammen et al., 2014b) Retrospective, post-intervention parent survey at 1-year followup (Mammen et al., 2014b) Schools that collected baseline data in the Fall and follow-up data in Winter saw a decrease of active travel by up to 5% (p o 0.05) (Mammen et al., 2014a) Parent-reported data showed less driving in Children's age, household distance, and middle class neighbourhoods schools were predictors of change (Mammen et  Active travel: Parent-reported data showed that 29% of students in the intervention group increased their walking, compared with 19% in the control group (net increase of 9.8%, p ¼ 0.05) at 1-year-follow up point There was no evidence for differences in active travel by studentreported data Students: All students aged 10-12 years present in participating schools on data collection days Student self-report by classroom survey for five consecutive days at baseline and 1-year follow-up Student-reported data showed distance from home to school and noncar use at baseline were predictors of non-car use at 1-year-follow-up time point (both p o 0.001) this review.
Eight interventions (29.6%) comprised modifications to parks and playgrounds, in which study participants included park users (6 studies), population-based samples (1 study), and school populations (1 study). The impact of changes to the built environment on road traffic safety measures was examined in three studies (11.1%) which all used population-based data sources. Of seven studies relating to four multi-component community-based health initiative interventions, six focussed on school populations and one considered the wider community. Finally, for the 12 interventions which aimed to promote active travel (44.4%, 15 studies), 13 studies selected participants from school populations and two focused on the wider community.
As part of a multi-component intervention, which included park and playground renovations in the USA, annual incidence rates of injuries in children aged 5-16 years decreased (relative risk: 0.74, 95% CI: 0.62-0.89) compared with the previous six years (Davidson et al., 1994).
Change to levels of park usage was reported in five studies (Bohn-Goldbaum et al., 2013;Cohen et al., 2009Cohen et al., , 2012Tester and Baker, 2009;Veitch et al., 2012). In an Australian-based study, the number of children and teen park users increased between baseline and six-months post-intervention, although p-values were not provided (Veitch et al., 2012). In another Australian-based study, no detectable difference between the mean number of children using the parks was observed at the two year follow-up time point (p ¼0.42) (Bohn-Goldbaum et al., 2013). Three studies undertaken in the USA showed a decline in the number of children and adolescent park users post-intervention (p-values not provided) (Cohen et al., 2009(Cohen et al., , 2012Tester and Baker, 2009).
Interventions did not appear to increase children's or young people's level of physical activity. Following park renovations and upgrades in the Australia (Bohn-Goldbaum et al., 2013) and New Zealand (Quigg et al., 2012), no evidence was demonstrated for a difference in the proportion of children engaging in Moderate to Vigorous Physical Activity (MVPA) at the two year follow-up time Proportion of students overweight/at-risk of overweight increased; proportion of students consuming 5 or more servings of fruit and vegetables a day decreased; no changes to proportion of students engaging in regular moderate PA x Serious Active travel Boarnet et al. (2005aBoarnet et al. ( , 2005b Parent-reported children walking/ bicycling less; increased walking observed in 3 of 5 sidewalk improvement projects, and 2 traffic signal improvement projects  -Goldbaum et al., 2013) or for objectively measured Mean Total Daily Physical Activity after one year followup (p-value not provided) (Quigg et al., 2012). In a USA-based intervention, removal of seating arrangements in the park did not change the likelihood of children standing or engaging in MVPA (p ¼0.35) (Roemmich et al., 2014).

Road traffic safety measures
Three studies examined road traffic injuries (Dimaggio and Li, 2013;Grundy et al., 2009;Ragland et al., 2014) of which one UKbased study investigated the impact of 20 miles per hour (mph) traffic speed zones on road traffic casualties from a non-randomised observational study (Grundy et al., 2009). All three were classified at an overall moderate risk of bias with none of the individual domains being classified as at serious risk of bias.
The effect of the USA Safe Routes to School programmes (which included installation of sidewalks and traffic calming measures) on road traffic causalities and rate of collisions was also reported: data in relation to 30 of 124 potentially eligible intervention sites (Dimaggio and Li, 2013) and 93 of 313 Safe Routes to School programmes (Ragland et al., 2014) were obtained for these studies. For all three studies, outcome measures were population-based and comprised routinely collected data in relation to hospital admissions and police investigations for crashes and road traffic accidents.
Introduction of 20 mph traffic speed zones was associated with an annual average decline of 3.4% (95% CI: 3.1-3.7) in the incidence of all types of casualties amongst children aged 0-15 years and an annual average decline of 3.9% (95% CI: 3.6-4.3) reduction in pedestrian casualties aged 0-15 years (Grundy et al., 2009). Evidence for the impact of the Safe Routes to School programme was equivocal. A post-intervention decrease in the annual rate of school-aged pedestrian injury (44%, 95% CI: 17-65) during schooltravel hours was reported in the New York programme (Dimaggio and Li, 2013). However, there was no strong evidence for a postintervention reduction in collisions within 250 feet of a built environment change (incident rate ratio: 0.47, 95% CI: 0.20-1.12) in the Californian programme (Ragland et al., 2014).
In the 'Shape Up Somerville' intervention, changes to the built environment included: traffic calming measures to and from the school environment; advocacy to paint crosswalks; installation of pedestrian crossing signs; opening and renovation of parks, and;  provision of bike racks. Selection of the study site for intervention was non-randomised and included either two (Economos et al., 2007Folta et al., 2013) or one (Chomitz et al., 2012) control site. Objectively measured BMI data was reported (Economos et al., 2007, in addition to parent-reported children's dietary and physical activity behaviours (Folta et al., 2013) and student selfreported levels of physical activity (Chomitz et al., 2012).
In comparison to the control communities, average change of the BMI z-score in the 'Shape Up Sommerville' intervention community was À 0.10 (95% CI: À 0.12 to À 0.09) at the one year follow-up time point (Economos et al., 2007) and À 0.06 (95%CI: À0.10 to À 0.05) at the two year follow-up time point . Parent-reported data indicated children consumed less sugar-sweetened beverages ( À 2.0 ounces per day; 95% CI: À3.8 to À 0.2) and were more likely to participate in organised sports and physical activities (0.20 sports or activities per year; 95% CI: 0.06-0.33) at the two year follow-up time point (Folta et al., 2013). Student self-report data at the four year follow-up time point suggested that high-school aged students were more likely to meet physical activity recommendations at follow-up after adjusting for demographic, health, and behavioural variables (OR: 2.36, 95% CI: 2.29-2.43) (Chomitz et al., 2012).
One study undertaken in California examined changes to physical activity following a move to a Smart Growth community, which was designed with features considered conducive to physical activity such as fewer barriers to connectivity, more parks and playgrounds, and traffic safety (Dunton et al., 2012). Using accelerometer data, there was no strong evidence that daily MVPA increased to a greater extent in the Smart Growth group (p ¼0.51) (Dunton et al., 2012).
Two of the community-based interventions used a before and after study design without a comparison group (Hendricks et al., 2009;Maddock et al., 2006). Changes to the built environment included improvements for pedestrian safety through the provision of walking paths and signs (Hendricks et al., 2009;Maddock et al., 2006). Using student self-report measures, increases in the proportion of students walking to school in Michigan were identified in four schools which had at least two years of data (Hendricks et al., 2009). However the Hawaii initiative showed, through self-report questionnaire data, that the proportion of students classified as overweight or at risk of being overweight increased and those consuming five portions of fruit and vegetables a day decreased at follow-up (Maddock et al., 2006).

Safe Routes to School programme
There was some weak evidence to support the effectiveness of the Safe Routes to School programmes at increasing active travel. Across five USA states (Florida, Mississippi, Texas, Washington and Wisconsin), pre-and post-intervention data suggested that overall active travel increased from 12.9% to 17.6% (p o0.001). There was evidence that walking increased by 45% (9.8% pre-intervention, 14.2% post-intervention, p o0.001) and cycling increased by 24% (2.5% pre-intervention, 3.0% post-intervention, p ¼0.01) (Moudon and Stewart, 2012).
Similarly, through the Eugene programme (Oregon, USA), student reported data showed increased walking and bicycling to school after four years follow-up (number or p-values not reported) (McDonald et al., 2013). Sites which implemented a programme that combined education and two Safe Routes to School interventions were associated with a 20% increase in rates of walking (p r0.05), but no associated affect with cycling (p-value not reported). However, there was no evidence for a difference in sites which implemented education and crosswalk/sidewalk interventions (p-values not reported) (McDonald et al., 2013).
In the Californian programme, observations of students walking to school suggested an increase in three of five sites which implemented sidewalk improvement projects and two of 10 sites which implemented traffic signals, but no differences were observed for crosswalk and signal or bicycle path improvement projects (Boarnet et al., 2005b). However, retrospective parentreported data suggested there were more children who reduced their rates of walking or cycling (18.0%, 155/862) post-construction, than children who increased walking or cycling (10.6%, 91/ 862) (Boarnet et al., 2005a). In Seattle, the authors reported a 24% increase in number of students who walked to school (data collection method and p-values not reported) where speed limits were enforced (Deehr and Shumann, 2009).

School Travel Plan programme
Evidence for the effectiveness of the School Travel Plan programme at increasing active travel was inconsistent. In a Canadian programme, child self-report data showed a small increase in active transport (pre-intervention: 43.8%, post-intervention: 45.9%, p-value not reported) (Buliung et al., 2011). In another Canadian programme, there was no evidence for a difference in active travel using child self-report data (n and p-values not provided) (Mammen et al., 2014a) but retrospectively collected parental data suggested reductions in driving of 16.7% in the morning (n ¼1,118) and 17.1% in the afternoon (n ¼1,211) (Mammen et al., 2014b). Infrastructure improvements and safety education were perceived by parents to be the most effective strategies implemented (Mammen et al., 2014a). In the New Zealand programme, studentreported data suggested an increase in active travel at the second (5.9%, standard deviation (SD) 76.8%) (Hinckson and Badland, 2011) and third (OR: 2.65, 95% CI:1.75-4.02) year time points .

Ride 2 School programme
The authors reported a small increase in the proportion of active trips to and from school using parent-reported data (47.9% pre-intervention, 49.6% post-intervention, p-values not reported) but student-reported data suggested a decrease (51.1% pre-intervention, 48.7% post-intervention, p-values not reported) (Garrard and Crawford, 2010).

Central Sydney Walk to School Research programme
Parent-reported data from a randomised controlled trial showed a net increase of 9.8% (p ¼0.05) of students increasing active travel in the intervention group at the one year-follow up time point. However, there was no evidence for differences in active travel by student-reported data. Distance from home to school and non-car use at baseline were predictors of non-car use at the one year follow-up time point (both p o0.001) (Wen et al., 2008).

Single component interventions
Following the introduction of a cycle lane in an urban area in the USA, the mean number of youths observed cycling each day doubled (baseline 2.2, SD 3.1; 1-year follow-up 5.2, SD 7.4) (Parker et al., 2013). However, no differences were found in median counts of active travel in experimental schools between baseline (8.5) and the two year follow-up time point (9.0) (p-values not provided) after the retrofitting of urban greenway/trail in Knoxville, USA (Fitzhugh et al., 2010). In a UK traffic calming scheme (comprising speed cushions, zebra crossings and parking bays), there was strong evidence that the observed pedestrian count across the road in which the intervention was implemented in all three sites (Morrison et al., 2004).

Main findings
This study systematically reviewed the available literature on interventions which incorporated changes to the built environment and reported health behaviours or outcomes for children and young people. There was some evidence of promise for interventions to reduce road traffic injuries and interventions to increase young people's active travel to school, and in relation to a multicomponent obesity prevention health initiative. There was limited evidence that interventions to parks and playgrounds increased usage.
A diverse range of study designs, outcome measures and study settings were used in the primary studies. Evidence for effectiveness of such interventions is at present weak and compounded by the study designs which may be opportunistic, frequently incorporate non-randomised allocation of study sites, use subjective outcome measures, and do not incorporate follow-up of study participants. The uncertainty in the evidence currently limits our understanding of which changes to the built environment can improve health behaviours and outcomes of children and young people. This in turn creates challenges in developing specific recommendations for policy makers and the public sector to make decisions in relation to the implementation of urban development projects.
4.2. Comparison to the literature 4.2.1. Road traffic safety There was evidence from two studies that interventions may reduce road traffic injuries in children and young people and create safer communities. These studies had access to relatively robust routinely collected, population-based data for longer time periods than studies relating to other interventions. Similarly, in relation to the general population evidence from two systematic reviews and meta-analyses suggests that area-wide urban traffic calming schemes, such as speed limits and one way systems, on average reduce the number of injuries by about 15% (Elvik, 2001) and can reduce road crash related deaths (pooled rate ratio: 0.63, 95% CI: 0.14-2.59) (Bunn et al., 2003).

Active travel
Accessible pavements and street connectivity are considered important to facilitate active travel. A recent systematic review examining interventions which promoted active travel to school (including educational programmes without any changes to the built environment), found only three of the 14 interventions had a large, or very large, effect size on rates of active travel (Chillón et al., 2011). In the current systematic review, 13 of the 15 primary studies identified suggested an increase in active travel (Boarnet et al., 2005a(Boarnet et al., , 2005bBuliung et al., 2011;Deehr and Shumann, 2009;Garrard and Crawford, 2010;Hinckson and Badland, 2011;Mammen et al., 2014a;McDonald et al., 2013;Morrison et al., 2004;Moudon and Stewart, 2012;Parker et al., 2013;Wen et al., 2008) but the findings need to be treated with caution: some studies did not provide evidence for size of effect (Deehr and Shumann, 2009;Fitzhugh et al., 2010;Garrard and Crawford, 2010;McDonald et al., 2013;Moudon and Stewart, 2012), reported very small effects (Buliung et al.,2011;Garrard and Crawford, 2010;, or used retrospective parental report methods (Boarnet et al., 2005a;. Many of the changes to the built environment in the primary studies of the current review were implemented as part of a multicomponent programme. Therefore, attributing behavioural outcomes to specific changes to the built environment is difficult to untangle. In addition, few authors attempted to address this in the analysis. None of the studies used objective measures of active travel, such as accelerometer data. This is a significant weakness given that there were two studies in which parents and children, participating in the same study, self-reported opposing results (Garrard and Crawford, 2010;Wen et al., 2008).

Park and playground interventions and physical activity
It is argued that the provision of clean, safe, and accessible public open spaces can offer children and young people opportunities for physical activity and social interaction (National Institute for Health and Care Excellence, 2008). However, the limited findings from this review, focussing on the public realm, suggest minimal effects of playground and park interventions in creating positive changes to usage or physical activity levels in children and young people.
The effect of physical changes to the playground environment has been more frequently researched in relation to the school setting. Systematic review evidence has shown that manipulation of pre-school playground environment with markings or equipment (Temple and Robinson, 2014) and school playground markings plus physical structures (Escalante et al., 2014) increased physical activity levels in children and young people. It may be that children are exposed to the intervention more often in the school environment compared to public parks which they may visit infrequently. Additionally in the school environment, researchers may find it easier to implement more robust study designs which incorporate follow-up of participants, as children are likely to attend the school for several years.
In the pre-school environment, a systematic review and metaanalysis showed that physical activity interventions which included environmental changes had a larger effect on increasing pre-school children's physical activity than interventions solely focused on physical activity (Gordon et al., 2013). Similarly, a systematic review examining the impact of interventions in children and adults to promote physical activity in urban green space, suggested combining changes to the built environment with physical activity programmes could be more effective than changes to the built environment alone (Hunter et al., 2015). This suggests future robust studies are required to evaluate the impact of combining structural improvements to public parks and playgrounds with behavioural change interventions.

Strengths and limitations
We followed a systematic and comprehensive process including: a search strategy applied to multiple databases in different research fields to uncover all relevant studies; a diverse range of eligible health outcome measures, and; no restrictions on the basis of publication date.
Nevertheless, there are a number of limitations. Publication bias may be present if interventions to the built environment that did not show positive results are less likely to have been submitted or accepted for publication. To allow comparability between studies, interventions undertaken in low-or middle-income countries were not eligible and the applicability of the present study findings in these settings is unknown. English language publication bias may also be present as studies not published in English were excluded.
The majority of primary studies were considered to be at serious risk of bias. Many of the studies used quasi-experimental designs, and the incorporation of statistical analyses that controlled for confounding variables was infrequent. Where possible, future interventions should incorporate experimental studies with a randomised controlled design at the level of the study site (e.g. park, school or city). Only two studies in this review used objective measures of physical activity. Future intervention studies should incorporate objectively measured outcomes: for example, a study using Geographical Information System (GIS) and accelerometers (Coombes et al., 2014) was able to show that children attending schools with a more supportive local environment were more likely to maintain active travel behaviours than those with less supportive environments. In addition, given the complexity of public health interventions, researchers should undertake process evaluation alongside implementation (Moore et al., 2015) to gain greater understanding about which parts of multi-component interventions are most effective in improving the health outcomes and behaviours of children and young people.
The built environment is thought to be an important contributory factor to the persistence of health inequalities in the UK (Glasgow Centre for Population Health, 2013). This suggests addressing the 'upstream' social determinants of health by making changes to the built environment has the potential to reduce population-wide health inequalities (Marmot et al., 2010(Marmot et al., , 2008. Addressing barriers and facilitators operating at policy and neighbourhood levels, is likely to be a more effective and far reaching strategy for public health improvement than short-term, behavioural interventions targeted at the inter-or intra-personal level alone (McLeroy et al., 1988). Despite this, few studies examined or controlled for differences by socioeconomic status, ethnicity or gender. Therefore, the potential impact of built environment interventions on reducing inequalities remains unknown. Future studies need to be adequately powered to allow exploration of different effects amongst different population groups.
No primary studies reporting mental health or well-being outcomes were captured in this review. This is important because associations between poor mental health outcomes and exposure to community violence or crime have been shown (Clark et al., 2007;Sinha and Rosenberg, 2013). In addition, the UK governmental strategy on the built environment and health recognises that population improvements to mental health and well-being need to be evaluated, in addition to physical health outcomes such as obesity and unintentional injury (The Scottish Government, 2008). Further research is required to examine whether changes to the built environment can have positive effects on children's mental health and well-being outcomes.

Challenges and recommendations
Ogilvie et al. identified a number of challenges in evaluating the provision of new walking and cycling infrastructure, including the selection of variables to be measured for both exposure and outcomes, and suggest this relates in part to the different perspectives of public health and transport research (Ogilvie et al., 2012). This challenge was also evident in the studies included in the current review. For example, undertaking a head count in a park or playground at two time points may provide evidence of increased use, but is not sufficient to conclude an increase in children's physical activity or a reduction in BMI. However, the difficulty in attributing health outcomes should not be a reason to suggest local authorities should refrain from improving parks. Increased park use may be beneficial for reasons other than physical activity and a wider perspective may be necessary when considering effectiveness and cost effectiveness. Wildlife, education, safety or crime reduction may all be relevant or an urban park may, for example, assist storm water management (Hartig et al., 2014). Nevertheless, if the purpose is to improve health, then a robustly measured health outcome is required. Hunter et al. (2015) list a number of methodological considerations to inform the design, implementation and evaluation of physical activity interventions in urban green spaces which are relevant to the wide range of studies identified in this review: reporting and justifying sample size and accounting for clustering where appropriate; using multiple data sources (observations, surveys, interviews, and objective measures such as accelerometer and GPS data); collecting baseline and long-term follow-up data, and; identifying an adequate (albeit not perfect) control condition. We would also highlight the need to consider and minimise the risk of bias. The tool used in this review (Sterne et al., 2014) is stringent and requires that a judgement of serious risk of bias within any domain should be applied to the study as a whole, irrespective of which domain is being assessed ( Table 3). The tool has been developed for non-randomised studies of interventions and, while built environment interventions pose particular challenges, it is important to be aware of and minimise bias due to confounding, participant selection, measurement of interventions, departure from intended interventions, missing data, measurement of outcomes, and selection of reported results.
Elsewhere, it has been argued that postponing action until a strong evidence base has been developed has the potential to cause more harm than good (Frank et al., 2012). Policy makers will continue to make decisions about the neighbourhood environments in which children and young people live, while public health researchers and transport or urban planners differ in their opinions about what constitutes evidence of effectiveness. It is important that inter-disciplinary teams share expertise across transport, planning, public health and other relevant disciplines. In the UK, the decision to locate public health within local authorities offers the potential for greater collaboration in building an evidence base that is sufficiently 'robust' for a public health audience while not inhibiting policy makers and planners.
We suggest that public health researchers, policy makers and the public sector should embrace opportunities to work collaboratively to implement and evaluate built environment interventions which address some of the methodological weaknesses identified in this review.

Conclusion
The interventions captured in this review related to active travel, park renovations, road traffic safety, and multi-component community health initiatives. Although the majority of studies had a serious risk of bias, there was some evidence of effectiveness of in relation road safety measures and active travel. Future research studies should involve collaborations between researchers, policy makers and planners, and consider using randomised controlled study designs which incorporate objective outcome measures. A joint agenda to generate and act upon the best available evidence will aid decision-making for those who must choose between, or prioritise, different options to improve the health and well-being of children and young people who live in urban environments.

Conflicts of interest
The authors have no conflicts of interest to report.