Community wide interventions for increasing physical activity

See Characteristics of included studies; Characteristics of excluded studies

Summary of findings for the main comparison.

Community wide interventions for promoting physical activity
Patient or population: whole communities (adults, adolescents and children) Settings: community based Intervention: multi‐component of at least two physical activity interventions targeting the whole community Comparison: existing programmes and infrastructure
Outcomes [duration of follow up]	Summary of effects	Number of communities (studies)	Quality of the evidence (GRADE)
Physical activity % Physically active Intervention compared to control adjusted pre/post cross‐sectional sampling (end of intervention to 6 years)	Typically no evidence of benefit	25 (10)	⊕⊕OO¹ Low
Physical activity % physically active Intervention compared to control adjusted pre‐post cross‐sectional sampling (end of intervention to 3 years, 4 months)	Typically no evidence of benefit	160 (3)	⊕⊕⊕⊕ High
Energy expenditure METS/week score, adjusted mean difference (follow up; end of intervention to 4 years)	Typically no evidence of effect Range: ‐241 to +176	156 (5)	⊕⊕OO¹ Low
Physical activity Average daily minutes of moderate to vigorous (24 months)	No evidence of effect from the baseline of 36 minutes per day	2 (1)	⊕⊕⊕O² Moderate
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Substantial heterogeneity between trials regarding type of interventions and measured outcomes; wide and overlapping range of effects ²Findings based on a single study in only two communities

Background

Physical activity is recognised as being important for reducing the overall burden of disease (WHO 2009). Very strong scientific evidence based on a wide range of well‐conducted studies shows that physically active people have higher levels of health‐related fitness, a lower risk profile for developing a number of disabling medical conditions and lower rates of various chronic diseases than do people who are inactive (US Physical Activity Guidelines 2008).

Despite the positive health effects associated with regular physical activity, physical inactivity remains a common public health problem in high, middle and low income countries (Heath 2012). The prevalence of physical inactivity remains high, and in some cases has even increased in recent years (Bauman 2009; Guthold 2008). In addition, low income and ethnic minority adults have the highest rates of physical inactivity, people at the top of the socio‐economic scale appear to perform more leisure‐time activity than those at the bottom of the scale, and participation is patterned by age and gender (Belanger 2011; Crespo 2000; Crespo 2001; Gidlow 2006).

The lack of physical activity cannot be attributed solely to personal motivation and so countries that are tackling this complex issue are increasingly electing to employ multi‐component approaches (that is informational, behavioural, and environmental) in increasing a population's physical activity (Heath 2012; Kahn 2002; WHO 2004).

Description of the intervention

Community wide interventions are attractive in that they aim to improve the health risk factors (especially low physical activity) of a whole population. These strategies generally involve investment in visible infrastructure and planning initiatives with the aim of producing long‐lasting benefits for the community. They differ from singular community based strategies which may target only a particular subset of the population. Community wide interventions offer a number of advantages over offering only one approach to a population. They operate at a series of levels to impact on behaviour. These levels reflect social‐ecological models of health and include changes to policies and environments, and involve mass media and individually focused activities (for example primary healthcare screening).

One systematic review has categorised these interventions into four types (Cavill and Foster 2004). These are (1) comprehensive integrated community approaches, where physical activity is part of an overall risk factor reduction programme (for example the Minnesota Heart Health Project (Luepker et al 1994)); (2) community wide ‘campaigns’ using mass media (Renger 2002)); (3) community based approaches using person focused techniques; and (4) community approaches to environmental change. The third category includes programmes that use methods and strategies such as one‐to‐one counselling, classroom instruction, and cognitive‐behavioural strategies but in community facilities and settings such as church halls or community centres (Sharpe 2003). The final category includes programmes that use some form of community action, often including a coalition or advocacy group, to make positive changes to the physical environment (King 1994). These interventions are often delivered to communities in combinations.

How the intervention might work

We developed a logic model to capture the broad range of different approaches found in community interventions (Figure 1). This framework divides the actions into two phases, a community strategy development phase and an implementation phase, as there is some evidence to suggest community wide approaches appear more sustainable in the longer term (Foster 2000). The community strategy development phase describes the construction phase of a community intervention. Actions include identification of target groups, populations, the setting for delivery, stakeholders and intervention options. The implementation phase describes the delivery of actions to encourage physical activity behaviour change. Actions might include mass media campaigns, community participation or educational events, advocacy and environmental changes. The outputs of both phases might be measured in a range of variables as short to long‐term outcomes. For example, intermediate outcomes could include knowledge of the benefits of an active lifestyle or improved access to physical activity. Examples of long‐term outcomes could be a reduction in morbidity and mortality related to physical activity behaviour. Changes in the proximal and intermediate variables, such as knowledge or attitudes, are likely to be more amenable to change through communication campaigns (Cavill and Bauman 2004).

Figure 1

Logic Model for Community Wide Interventions for Increasing Physical Activity.

Why it is important to do this review

Many studies of community wide interventions have been undertaken but, prior to our earlier review, few have published evaluations of their process or impact. Although the popularity of these interventions is increasing, there was a need to combine all the global evidence currently available in an up‐to‐date systematic review. We believed a review would enable a more in‐depth exploration of the effectiveness of the interventions as well as investigating equity and inclusiveness issues. Earlier reviews (for example Kahn 2002) do not contain the more recent studies and newer health promotion strategies built upon more recent research and health promotion theory. It is hoped that this update of the Cochrane review will be particularly useful to those decision makers with the responsibility of selecting and implementing community wide investments. The application of the logic model for this review illustrates the belief that community wide interventions should be understood more broadly than as being just the sum of several interventions that have been implemented in a community.

Objectives

Primary research objective

We sought to determine the effects of community wide, multi‐strategic interventions upon community levels of physical activity.

Secondary research objectives

We addressed the following predetermined research objectives.

To explore whether any effects of the intervention are different within and between populations, and whether these differences form an equity gradient.
To describe other health (e.g. cardiovascular disease morbidity) and behavioural effects (e.g. diet) where appropriate outcomes are available.
To explore the influence of context in the design, delivery and outcomes of the interventions.
To explore the relationship between the number of components, duration and effects of the interventions. As an addition to the published protocol, we sought to understand more explicitly whether the intensity of the community wide intervention could explain differences of effects between studies.
To highlight implications for further research and research methods to improve knowledge of the interventions in relation to the primary research objective.

Methods

Criteria for considering studies for this review

Types of studies

It is recognised that public health and health promotion interventions are evaluated using a wide variety of approaches and designs. We permitted the inclusion of cluster randomised controlled trials, randomised controlled trials (RCTs), quasi‐experimental designs which used a control population for comparison, interrupted time‐series (ITS) studies, and prospective controlled cohort studies (PCCS). Only studies with a minimum six‐month follow up from the start of the intervention to measurement of outcomes were included. The six‐month period was considered as the minimal time frame as physical activity behaviour changes, as understood by the Prochaska and DiClemente model (Prochaska 1992), are established in the action stage, which is when the individual actively engages in the new behaviour. For physical activity, the highest likelihood for relapse occurs within the first six months of starting a regular program (Dishman 1994).

Types of participants

The term community wide generally refers to either: 1) an intervention directed at a geographic area, such as a city or a town defined by geographical boundaries; or 2) an intervention directed toward groups of people who share at least one common social or cultural characteristic.

As the focus of the review was whole‐of‐community interventions, we defined participants in the included studies as comprising those persons of any age residing in a geographically defined community, such as urban, peri‐urban, village, town, or city. We excluded interventions which were whole of state or country. Although some of the strategies targeted individuals with chronic disease, collectively the participants included in the studies needed to be representative of the whole community and not restricted to a particular geographic subregion (for example a park) or subgroups (for example only elderly people). To be included, a strategy must have shown intent to be comprehensive in reaching the targeted community. Participants must have been free living and not part of any institutionalised community, such as those who were mentally ill, the frail or bedridden elderly population, or those incarcerated in prison.

Types of interventions

It is recognised that to achieve a whole of community approach requires more than a singular strategy, as changing behaviour is a difficult task (Mummery 2009). Although little is known about how to reach the most disadvantaged groups in the community (Mummery 2009), we defined a community wide approach as one which should include strategies that have, within their scope, outreach to many disadvantaged groups. For this review, we defined a community wide intervention as one which has at least two of the following six broad strategies aimed at physical activity. The list categories of suitable strategies, which would be components of an integrated community wide intervention, are consistent with the logic model.

1. Social marketing through local mass media (e.g. television (TV), radio, newspapers).

2. Other communication strategies (e.g. posters, flyers, information booklets, websites, maps) to raise awareness of the project and provide specific information to individuals in the community.

3. Individual counselling by health professionals (both publicly and privately funded), such as the use of physical activity prescriptions.

4. Working with voluntary, government and non‐government organisations, including sporting clubs, to encourage participation in walking, other activities and events.

5. Working within specific settings such as schools, workplaces, aged care centres, community centres, homeless shelters, and shopping malls. This may include settings that provide an opportunity to reach disadvantaged persons.

6. Environmental change strategies such as creation of walking trails and infrastructure with legislative, fiscal or policy requirements, and planning (having ecological validity) for the broader population.

Studies that were community based but did not include at least two of the six stated strategies were excluded. We recognised that single strategy interventions (for example mass media only) are likely to be topics of other reviews and they were beyond the scope of this review.

Types of outcome measures

Primary outcomes

Whilst it is desirable to focus on a small range of outcome measures, the context for research in this area of health is that measures of physical activity at a population level are complex (both the measures and the methods) and international consensus on gold standards has not been reached.

To be included in this review, studies needed to measure physical activity in the study population. Physical activity could be quantified using a variety of measurements, for example percentage of people active or inactive, frequency of physical activity, percentage meeting recommendations, percentage undertaking active travel; and other objective (for example accelerometers, pedometers) or subjective methods (for example self‐reported questionnaires, diaries) (Bassett et al 2008).

Secondary outcomes

Data on other related measures of health were extracted.

1. Measures of health outcomes and risk factor status (e.g. cardiovascular disease, body mass index (BMI), energy expenditure).

2. Measures of other health behaviours (e.g. sedentary behaviour, dietary patterns, or smoking).

3. Intermediate outcomes (e.g. knowledge of and attitudes toward the benefit of physical activity).

4. Any adverse outcomes that were reported (e.g. unintended changes in other risk factors, opportunity cost, and injuries).

Process measures

Measures relating to the process of implementing an intervention were also extracted.

Search methods for identification of studies

Electronic searches

We searched the following databases:

Cochrane Public Health Group Specialised Register in the Cochrane Register of Studies (CRS);
The Cochrane Library;
MEDLINE, MEDLINE In‐Process;
EMBASE;
CINAHL;
PsycINFO ;
LILACS;
ASSIA;
British Nursing Index (BNI);
Database: CAJ, CCND, CPCD, CJSS, CMFD, CDFD, Chinese CNKI databases (http://www.global.cnki.net/grid20/index.htm);
EPPI Centre;
DoPHER;
TRoPHI;
ERIC;
Health Management Information Consortium (HMIC) (grey literature);
Sociological Abstracts;
SPORTDiscus;
Transport Database TRIS;
Web of Science

- Science Citation Index, Social Sciences Citation Index and Conference Proceedings Citation Index,
- Science Citation Index, Social Sciences Citation Index and Conference Proceedings Citation Index.

We searched the following websites for relevant publications, including grey literature:

EU Platform on Diet, Physical Activity and Health;
Health Evidence (http://healthevidence.org);
IUHPE (International Union for Health Promotion and Education);
NCCHTA (National Coordinating Centre for Health Technology Assessment) (http://www.ncchta.org);
NICE guidelines (http://www.nice.org.uk);
SIGN guidelines (http://www.sign.ac.uk);
US Centres for Disease Control and Prevention (http://www.cdc.gov/);
World Health Organization (http://www.who.int/en/).

Searches were carried out for studies published from January 1995 to January 2014. The search strategies and details of the search dates can be found in Appendix 1 . The MEDLINE search was developed for precision and sensitivity with advice from the Public Health Group's Trials Search Co‐ordinator and tested against a set of 38 relevant studies from across the globe. The search was then adapted to the remaining databases using database‐specific subject headings, where available.

Searching other resources

In addition, reference lists of all relevant systematic reviews, guidelines and included primary studies were searched.

For the original review, the following experts in the field were contacted to ask if they were aware of any recently published, in press or unpublished studies: Dr Harry Rutter (National Obesity Observatory, Oxford), Dr Nick Cavill (Oxford University), Mr Glenn Austin (GP Links Wide‐Bay), Mr Jiandong Sun (Queensland University of Technology), Professor Kerry Mummery (University of Central Queensland), Professor Gregory W Heath (University of Tennessee College of Medicine) and Professor Ross C Brownson (Washington University in St Louis). Subsequent to the original review we had studies brought to our attention by experts and researchers.

The past 12 months of the six journals that contained two or more studies (completed or in progress) meeting the review inclusion criteria were handsearched in the original review, however for the update this was determined as unnecessary and was not repeated. The journals were:

American Journal of Public Health;
Australia Health Promotion;
BMC Public Health;
Norsk Epidemiologi;
Preventive Medicine;
Scandinavian Journal of Public Health.

Through various methods, including contact with authors, the review team obtained a full text PDF or an abstract containing sufficient details to determine eligibility of all potentially relevant studies. Non‐English study reports were all examined by readers with appropriate language skills to determine whether they were to be excluded or included.

Data collection and analysis

Selection of studies

The initial search strategy produced a listing of nearly 26,000 citations across the original review and this update. An initial screening of titles and abstracts was undertaken to remove those which were obviously outside the scope of the review. Authors were overly inclusive at this stage and, if in doubt, a paper was left in. The full text was obtained for the papers potentially meeting the inclusion criteria (based on the title and abstract only) and multiple publications and reports on the same study were linked together. All the full text papers obtained were then screened by two review authors (PB and shared between DF, JS, and CF) who compared the description of the intervention with the logic model (Figure 1) to assess whether the required components of a community wide intervention and permissible study designs were fully met. Where there was a persisting difference of opinion, a third review author was asked to review the paper in question and a consensus was reached between the three review authors.

Data extraction and management

Data were extracted for all the studies that met the inclusion criteria. For each study, two review authors (PB and shared between DF, JS, and CF) independently completed data extraction forms, which were tailored to the requirements of this review. Quality criteria questions for RCTs, controlled clinical trials (CCTs), controlled before and after (CBA) studies and ITS study designs were incorporated into the data extraction form. A checklist was used to ensure inclusion of data relevant for health equity (Ueffing 2009). In addition, multiple reports and publications of the same study were assembled and compared for completeness and possible contradictions. Data were extracted from companion studies that reported findings on the process evaluation of the intervention. The specific components present in the primary paper and companion publications were reviewed using the logic model (Figure 1) to assist in the categorisation of studies and interpretation of results where heterogeneity was present.

Numerical data for analysis were extracted from the included studies and managed in an Excel spreadsheet.

The data extraction form was first piloted by three review authors (PB, DF, and JS) to assess its ability to capture study data and inform assessment of study quality. Problems in the use of the form that were identified were resolved through discussion and the form was revised as required.

Where studies reported more than one endpoint per outcome, the primary endpoint identified by the authors was extracted. Where no primary endpoint was identified by the authors, the measures were ranked by effect size and we extracted the median measure (Curran 2007). Measures of physical activity or sedentary behaviour that were based upon meeting a national standard were noted and the potential for unequal comparisons identified. We collected information on how physical activity was reported, that is whether it was through self‐report in a telephone survey or devices such as pedometers. Data extracted independently by the review authors were compared and any differences were resolved through discussion.

Assessment of risk of bias in included studies

Only studies that met the inclusion criteria were assessed and reported in a risk of bias table as per the recommendation of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008).

Two review authors (PB and one other author) assessed the risk of bias for each study. Analysis of non‐randomised controlled trials followed the recommendations in Chapter 13 of the Cochrane Handbook for Systematic Reviews of Interventions. Where there was disagreement between review authors in risk of bias assessment, this was resolved by discussion and consensus.

Studies were assessed for the five general domains of bias: selection, performance, attrition, detection, and reporting, as well as for an additional category to capture any other concerns pertaining to the study quality that did not fit distinctly into either of the five domains. For example, this additional category included instances where the statistical analyses presented in the included study were problematic and failed to adjust for baseline differences between the control and intervention groups, or failed to address what appeared to be regression to the mean. This category was also applicable if there appeared to be a 'head‐start' or other advantage for the intervention community. Each was assessed with answers of 'Yes' indicating low risk of bias, 'No' indicating high risk of bias, and 'Unclear' indicating either lack of information or uncertainty over the potential for bias. Studies were judged overall as at 'low', 'unclear' or 'high' risk of bias after consideration of the study design and size, and the potential impact of the identified weaknesses noted in the table for each study.

Specifically, assessment of performance bias included identification of explicit statements of measures undertaken to avoid contamination (that can occur when the control group also receives the intervention) such as spatial separation, non‐delivery of the program to the control communities, and minimisation of wide‐reaching mass media. We also considered measurement of the community's awareness of the message obtained through community surveys, both of the intervention and control communities. Additionally, integrity of the intervention was considered and performance bias was assessed as being present when the study's process evaluation (perhaps an additional publication) described instances where the program was not delivered as planned.

Studies were assessed as at high risk of detection bias when incomplete data were inadequately defined or, particularly in cross‐sectional sampling, where the characteristics of the follow‐up groups varied significantly from the baseline groups.

Detection bias was assessed to be at low risk where measurement tools were used in their entirety, the outcome assessment was blind (if deemed appropriate), the outcome measure metrics were valid, the measure was of sufficient quality (for example assessed over the period > one day) and the sample was representative (for example random sampling of the community).

Reporting bias was assessed as being at low risk if the reports appeared to be free from selective reporting and the measures reported were complete and matched the aims of the studies. Studies where follow‐up measurement was absent, or appeared to be deliberately withheld, were assessed as at high risk of reporting bias.

The review authors determined a priori that the best evidence (both contextually relevant and representing the purpose of the intervention) was likely to come from cluster RCTs and CBA studies. Although this differs from the usual evidence hierarchy (NHMRC 1999) (which emphasises RCTs for assessment of interventions), it is considered a better approach than the problematic application of the usual criteria when appraising the evidence for social and public health interventions (Petticrew 2003).

Measures of intervention effect

The effect sizes for dichotomous outcomes were expressed as relative risk (RR) and risk difference (RD) in the first instance. For comparability across studies, given the important baseline differences between intervention (I) and control (C) groups, we calculated from the authors' data an adjusted estimate of effect based on the differences at baseline. Therefore, for dichotomous outcomes we calculated the following.

1. Net percentage change from baseline = ((I_post ‐ I_pre)/I_pre) ‐ ((C_post ‐ C_pre)/C_pre) x 100.

2. Adjusted risk difference = (I_post ‐ I_pre) ‐ (C_post ‐ C_pre).

3. Adjusted relative risk = (I_post / C_post)/(I_pre/C_pre).

Confidence intervals (95%) were calculated using the Wald test.

For continuous outcomes we calculated the following from the authors' data.

1. Post mean differences (PMD) = Imean_post ‐ Cmean_post

2. Adjusted mean difference = [(Imean_post ‐ Cmean_post) ‐ (Imean_pre ‐ Cmean_pre)]

3. Adjusted percentage change relative to the control group = [((Imean_post ‐ Cmean_post) ‐ (Imean_pre ‐ Cmean_pre))/Cmean_post] x 100.

The 95% confidence intervals could not be calculated using this approach.

Unit of analysis issues

Studies allocated by clusters that did not account for clustering during analysis were not re‐analysed. This was because these studies were not randomised and there was only a small number of clusters, and so clustering would have a minimal effect.

Dealing with missing data

Protocols and baseline publications for the studies were used to identify outcome data that were expected to be present in the follow‐up report which presented the outcomes. Incomplete data (that is less than 40% of data) were assessed during the risk of bias assessment. Data that appeared to be completely absent were noted as reporting bias. Missing data were also captured in the data extraction form and reported in the risk of bias table. The authors were contacted to try and acquire missing data for inclusion. In some instances this included the use of a Chinese speaking epidemiologist.

Assessment of heterogeneity

Due to heterogeneity in the study designs employed, the populations in which the interventions were conducted, and the interventions themselves no meta‐analysis was conducted.

Assessment of reporting biases

We considered plotting trial effect against standard error and presenting this in a funnel plot (Higgins 2008) to determine whether asymmetry could be caused by a relationship between effect size and sample size or by publication bias (Egger 1998). However, we decided against doing this given the high risks of bias in the data and the poor quality of measurement undertaken in the studies.

Intensity of intervention

We categorised the intensity of the community wide intervention to assess whether intensity could account for differences that existed in the outcomes between studies. The intensity of the intervention was categorised based on the following six characteristics and attributes that we hypothesised would be important in understanding differences in the effectiveness of the community wide intervention; two review authors (PB and DF) independently assessed each characteristic as 'more intensive', 'less intensive' or 'unclear':

development of community partnerships and coalition (first level of the logic model ‘Community/Strategy Development’), showing evidence of engaging stakeholders and building a community coalition;
levels of intervention (second level of the logic model ‘Implementation’), intervening at the individual (personal), social (interpersonal) and environmental (physical and legislative) levels;
reach of the strategies (second level of the logic model), the intervention reaches the whole of the community, multiple sectors of the community, targets subgroups, with awareness > 85%;
magnitude of the intervention, the extent of continuous provision of the intervention through the intervention period (volume of the intervention): frequency and duration of strategies, with high intensity typified as sustained integration of the intervention;
description of cost, where stated the cost per person for the intervention (excluding the evaluation) in the context of the year and the location, presumably indicating the magnitude of the intervention;
statement of intensity by the authors, descriptors found within the studies where the investigators themselves used descriptors such as 'high impact' or 'significant cost'.

We categorised the overall assessment of intensity for each study as 'high', 'medium', 'low', or 'unclear'. Given that the six categories we assessed on were not distinct, and the sufficiency of detail varied between the studies, each review author independently made the overall assessment using subjective informed determination rather than a predefined algorithm. Discrepancies were resolved by discussion.

Data synthesis

Continuous outcomes were reported on the original scale. where possible. We predetermined we would undertake a meta‐analysis only when data were clinically homogeneous. We followed Chapter 9: 'Analysing data and undertaking meta‐analyses' of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008). As data were not available that were sufficiently similar and of sufficient quality, a meta‐analysis was not performed. We predetermined that evidence from differing study designs and outcome types was not to be combined in a forest plot from standard meta‐analysis (Christinsen 2009). However, to identify trends and provide summary statements, simple forest plots were generated for three dichotomous outcomes (% physically active, % physically active during leisure time and % physically inactive).

Subgroup analysis

We predetermined that, where sufficient data were available, we would perform additional subgroup analyses to compare outcomes by: types of study designs; group effects for people who shared a common social, cultural, or health status characteristic (for example age, gender, ethnicity); reach of intervention and intensity of intervention (derived from use of the logic model and process evaluations). We had intended that a subgroup analysis would be used to explore whether there was likely to be a relationship of effect to disadvantage and whether an equity gradient was present. Given the limitations of the data, both in their quality and the absence of subgroup reporting, no further subgroup analysis could be undertaken.

Sensitivity analysis

The studies with low risk of bias have been grouped in the forest plots.

Summary of findings

We had intended to undertake a summary of findings table for the primary outcomes related to physical activity and sedentary behaviour using GRADE profiler (Cochrane IMS 2009). This was to be created using the measures for the primary outcomes identified as being most reliable and which predominated. Given that very few studies had reliable measures of physical activity and sedentary behaviour, and much of the data were incomplete, a modified approach was required in which we split the presentation of findings according to the risk of bias. We considered the primary challenge that all the community wide interventions were different and all of the communities unique and thus caution was required in potentially homogenising very different approaches. As conducting meta‐analyses was deemed inappropriate, a summary table has been prepared using narrative analysis of the included studies.

Results

Description of studies

Results of the search

Electronic searches from 1995 to November 2009, in the original review, yielded 17,538 hits following removal of duplicates (Figure 2), of which 207 were considered potentially eligible and were assessed in full text. The update search, to January 2014, identified an additional 9551 hits following removal of duplicates (Figure 2), of which 62 were considered potentially eligible and assessed in full text. The results of the searches of the electronic databases and websites are found in Table 1 and Table 2, respectively. The full search strategies, dates, and number of hits are given in Appendix 1. Twenty‐five studies were included in the original review (Brown 2006; Brownson 2004; Brownson 2005; De Cocker 2007; Eaton 1999; Goodman 1995; Gu 2006; Guo 2006; Jenum 2006; Jiang 2008; Kloek 2006; Kumpusalo 1996; Luepker 1994; Lupton 2003; Nafziger 2001; Nishtar 2007; NSW Health 2002; O'Loughlin 1999; Osler 1993; Reger‐Nash 2005; Sarrafzadegan 2009; Simon 2008; Wendel‐Vos 2009; Young 1996; Zhang 2003). Eight additional studies were identified in the update search (Gao 2013; Kamada 2013; Mead 2013; Nguyen 2012; Phillips 2014; Rissel 2010; Solomon 2014; Wilson 2014) resulting in a total of 33 included studies. We identified one study for which there is no published conclusion and have identified it as 'ongoing' (Davey 2011).

Figure 2

PRISMA diagram based upon Moher 2009.

Table 1. Search results for electronic databases

Database	Number of hits
ASSIA	1144
British Nursing Index (BNI)	105
CINAHL	2881
Chinese atabase:CAJ,CCND,CPCD,CJSS,CMFD,CDFD, http://www.global.cnki.net/grid20/index.htm	124
Cochrane Library	1841
Cochrane Public Health Group Specialized Register	31
EMBASE	4941
EPPI Centre DoPHER TRoPHI	38 200
ERIC	416
Health Management Information Consortium (HMIC)	308
LILACS	416
MEDLINE & MEDLINE In‐Process	5691
PsycINFO	1315
Sociological Abstracts	874
SPORTDiscus	365
Transport Database TRIS	49
Web of Science Science Citation Index, Social Sciences Citation Index and Conference Proceedings Citation Index	9108

Table 2. Search results for websites

Web sites	Hits
EU Platform on Diet, Physical Activity and Health	0
http://health‐evidence.ca	5
IUHPE (International Union for Health Promotion and Education)	0
NCCHTA http://www.ncchta.org	1
NICE guidelines http://www.nice.org.uk	4
SIGN guidelines http://www.sign.ac.uk	0
US Centres for Disease Control and Prevention http://www.cdc.gov/	0
World Health Organisation http://www.who.int/en/	1

Included studies

Communities in the included studies

Twenty‐five of the included studies were set in high income countries (using the World Bank economic classification). Of these, 11 studies were conducted in North America (Brownson 2004; Brownson 2005; Eaton 1999; Goodman 1995; Luepker 1994; Mead 2013; Nafziger 2001; O'Loughlin 1999; Reger‐Nash 2005; Wilson 2014; Young 1996), three in Australia (Brown 2006; NSW Health 2002; Rissel 2010), one in Japan (Kamada 2013) and 10 in Europe (De Cocker 2007; Jenum 2006; Kloek 2006; Kumpusalo 1996; Lupton 2003; Osler 1993; Phillips 2014; Simon 2008; Solomon 2014; Wendel‐Vos 2009). Of the remaining eight studies, two were set in lower middle income countries: one in Pakistan (Nishtar 2007) and one in Vietnam (Nguyen 2012); and six were set in upper middle income countries: five in China (Gao 2013; Gu 2006; Guo 2006; Jiang 2008; Zhang 2003) and one in Iran (Sarrafzadegan 2009).

A total of 267 communities were included in the review. The size of the community in which the intervention took place varied greatly, from two small villages with a total population of less than 1000 inhabitants (Kumpusalo 1996) and clusters of villages greater than 500 (Solomon 2014) to a large region with a population of 1,895,856 (Sarrafzadegan 2009). Similarly, the location of the communities varied with 12 studies taking place in what could be considered rural or remote settings and the remaining 21 studies located in urban centres or cities.

Interventions in included studies

When assessed against the six categories, we found substantial differences in the combinations of interventions used in the included studies. Almost all of the interventions included a component of building partnerships with local governments or non‐government organisations (NGOs) (29 studies). Other strategies used in the interventions included some form of individual counselling by health professionals (20 studies), mass media campaigns (23 studies) or other communication strategies (26 studies). Some studies were delivered in specific settings (18 studies) and used environmental change strategies (14 studies).

Only four interventions that were investigated by the included studies contained elements of all six of the components described in the inclusion criteria (Brown 2006; Gao 2013; Goodman 1995; Luepker 1994) (see Methods section). Three interventions were comprised of five components, 10 of four components, seven of three components and two of two components (Table 3).

Table 3. Categories of strategies included in interventions

Study	Mass Media	Other communication	Individual	Partnerships	Settings	Environmental	Total
Brown 2006	X	X	X	X	X	X	6
Brownson 2004		X	X	X		X	4
Brownson 2005	X	X	X	X			4
De Cocker 2007	X	X		X	X	X	5
Eaton 1999		X		X	X	X	4
Gao 2013	X	X	X	X	X	X	6
Goodman 1995	X	X	X	X	X	X	6
Gu 2006		X	X				2
Guo 2006	X	X	X				3
Jenum 2006	X	X	X	X		X	5
Jiang 2008		X	X	X			3
Kamada 2013	X	X		X	X		4
Kloek 2006	X		X	X	X		4
Kumpusalo 1996		X	X	X	X		4
Luepker 1994	X	X	X	X	X	X	6
Lupton 2003	X		X	X	X		4
Mead 2013	X			X	X		3
Nafziger 2001	X	X		X	X		4
Nguyen 2012	X	X	X		X		4
Nishtar 2007	X		X	X			3
NSW Health 2002	X	X		X		X	4
O'Loughlin 1999		X	X	X		X	4
Osler 1993	X		X	X			3
Phillips 2014		X		X		X	3
Reger‐Nash 2005	X	X	X	X	X		5
Rissel 2010		X		X	X		3
Sarrafzadegan 2009	X		X	X			3
Simon 2008				X	X	X	3
Solomon 2014	X	X		X	X	X	5
Wendel‐Vos 2009	X	X		X	X		4
Wilson 2014	X	X		X		X	4
Young 1996	X	X		X			3
Zhang 2003		X	X				2
Total	23	26	20	29	18	14

2 components ‐2 studies; 3 components ‐ 10 studies; 4 components ‐ 13 studies; 5 components ‐ 4 studies; 6 components ‐ 4 studies.

See Types of interventions for examples of suitable strategies which would be components of an integrated community wide strategy

Theoretical perspectives

Interventions were developed from a variety of theoretical perspectives, although many studies did not identify any such perspective in their papers. Nine of the studies sought to increase physical activity in a community by developing an intervention based on an ecological approach (Brown 2006; Brownson 2004; Brownson 2005; De Cocker 2007; Gao 2013; Jenum 2006; Mead 2013; Simon 2008; Wilson 2014). Six studies developed interventions with the stages of change model as their guiding framework (Kamada 2013; Kloek 2006; Luepker 1994; Phillips 2014; Reger‐Nash 2005; Rissel 2010) while four studies used the social learning model (Eaton 1999; Luepker 1994; O'Loughlin 1999; Osler 1993). Two studies used the community empowerment model for developing their interventions (Jenum 2006; Lupton 2003). Other theoretical approaches used included behaviour change of self‐efficacy (O'Loughlin 1999), persuasive communications theory (Luepker 1994), social cognitive theory (Mead 2013), active friendly environments (Solomon 2014), social marketing (Rissel 2010; Wilson 2014) and community organisation principles (Kloek 2006; Osler 1993). Of note, a number of studies described basing their interventions or components of interventions on multiple models. However, 11 did not explicitly state a theoretical model (Goodman 1995; Gu 2006; Guo 2006; Jiang 2008; Nafziger 2001; Nishtar 2007; NSW Health 2002; Sarrafzadegan 2009; Solomon 2014; Young 1996; Zhang 2003).

Intensity of Interventions

A subjective assessment of the intensity of each intervention was conducted based on the consideration of six criteria, as described in the methods section. Ten studies were judged to be high intensity, 14 of medium intensity and nine of low intensity (Table 4). The categorisation of high intensity was typically assigned to an intervention which acted on multiple levels within a community via multiple strategies as understood by the logic model (Figure 1). For example, the Brown 2006 study used mass media as well as other forms of communication to increase awareness of physical activity. The study also promoted self monitoring and goal setting using a website and provided access for individuals to pedometers and logbooks. Counselling by health professionals was another mode of intervention and a number of setting‐specific initiatives were conducted. The investigators also collaborated with the local government in improving the environment for physical activity by repairing walking tracks and creating signage and maps. Importantly, this intervention had the express intent of increasing the physical activity of the whole population, whereas some interventions included in this review targeted a range of behaviours other than physical activity. O'Loughlin 1999 was one such study which, with quite a modest budget (when compared to some of the larger interventions), employed multiple strategies in targeting smoking and diet along with physical activity. Given these factors it was considered to be of moderate intensity.

See: Summary of findings for the main comparison

Table 4. Assessment of intensity of the interventions

Study	High	Medium	Low	Unclear
Brown 2006	X
Brownson 2004		X
Brownson 2005		X
De Cocker 2007		X
Eaton 1999	X
Gao 2013		X
Goodman 1995			X
Gu 2006	X
Guo 2006		X
Jenum 2006		X
Jiang 2008	X
Kamada 2013			X
Kloek 2006			X
Kumpusalo 1996		X
Luepker 1994	X
Lupton 2003	X
Mead 2013		X
Nafziger 2001	X
Nguyen 2012		X
Nishtar 2007			X
NSW Health 2002			X
O'Loughlin 1999		X
Osler 1993			X
Phillips 2014		X
Rissel 2010			X
Reger‐Nash 2005		X
Sarrafzadegan 2009		X
Simon 2008			X
Solomon 2014			X
Wendel‐Vos 2009	X
Wilson 2014	X
Young 1996		X
Zhang 2003	X
Total	10	14	9	0

Intensity was assessed subjectively and independently based upon six characteristics as described in Data collection and analysis

The interventions studied by Gu 2006, Jiang 2008 and Zhang 2003 reached every individual in their target communities through quite substantial contacts such as repeated door‐to‐door visitation and health screening. The extensive reach of the intervention, combined with what was a potentially significant dose, led to their classification as high intensity interventions despite them being very different to Brown 2006. Conversely, most of the interventions judged as being of low level intensity had a much poorer reach into the communities. Indeed, several of the studies judged as being of low intensity were described by their authors as being of low intensity or low cost (Osler 1993; Simon 2008). In the case of Osler 1993, the low cost of the intervention was demonstrated in the limited amount of activity that took place compared to the more intense interventions. Similarly, Simon 2008 was judged as a low intensity intervention as, while it aimed to reach the whole community, the vast majority of its activities were targeted at one section of the community (in this case adolescents attending school). Overall, some studies appeared to have good reach (Gao 2013) whilst others (Solomon 2014) identified that very few residents were even aware of, and participated in, the intervention. Several of the studies provided descriptions of people participating in the components.

Outcome measures

To be included in the review, the study had to include a measurement of physical activity. A variety of dichotomous and continuous outcomes were used in these studies. Thirteen studies reported the proportion of participants attaining a certain level of physical activity (Brown 2006; Gao 2013; Jiang 2008; Kamada 2013; Kloek 2006; Lupton 2003; NSW Health 2002; Phillips 2014; Reger‐Nash 2005; Rissel 2010; Sarrafzadegan 2009; Solomon 2014; Wendel‐Vos 2009). The inverse of these outcomes was the reporting of the proportion of participants who were physically inactive, that is failing to attain a defined level of physical activity (Eaton 1999; Gao 2013; Goodman 1995; Jenum 2006; Nafziger 2001; Nguyen 2012; Osler 1993). Three other studies also reported the percentage of participants attaining a certain level of physical activity but prescribed that this had to have taken place during leisure time (Kumpusalo 1996; Luepker 1994; Nishtar 2007).

Time spent being physically active during leisure time (for example as hours per week) was also reported as a continuous outcome in three studies (De Cocker 2007; Simon 2008; Wendel‐Vos 2009). Other continuous outcomes of physical activity reported in the included studies included walking (Brownson 2004; Brownson 2005; De Cocker 2007; Wendel‐Vos 2009), energy expenditure (Kloek 2006; Phillips 2014; Sarrafzadegan 2009; Solomon 2014) and minutes in moderate‐vigorous physical activity each day (Wilson 2014).

Most of the included studies also measured other behaviours and health outcomes related to chronic disease. Behaviours measured included smoking, alcohol consumption, fruit and vegetable intake, fat and junk food intake and BMI. Other studies included speciality activity measures such as percentage of persons cycling. Knowledge and attitudes towards physical activity and health knowledge were reported in some studies. Health outcomes measured included chronic disease such as diabetes and hypertension, obesity and laboratory measures such as vitamin C, plasma and cholesterol levels. Reviewing the findings of these measures was not the objective of this review and so they have not been explored here.

Excluded studies

The Excluded studies table lists the studies that were excluded and the determined reasons. In several cases the studies were excluded for more than one reason. The predominant reasons for studies being excluded at this stage of the selection process were the study design (n = 84) or the intervention (n = 83) not meeting the inclusion criteria. In 42 cases the study was not designed in a way which could target the entire community, and in 28 cases the population sampled was not inclusive. In one case the study described the intervention without providing any results, in one case the report was inadequate and in five the measurement of physical activity was absent (deemed not likely to be the result of selective reporting of outcomes bias).

Risk of bias in included studies

The update has noted the increased use of randomisation in the allocation procedure and a significant improvement in the study design methodology from earlier studies. Earlier, all of the included studies were described as controlled before and after studies with the exception of one controlled ITS study (Luepker 1994) and one cluster cohort study (O'Loughlin 1999). Although the original review contained only one cluster RCT (Simon 2008), the updated review now includes an additional four RCTs: three cluster randomised studies (Kamada 2013; Phillips 2014; Wilson 2014) and one stepped wedge cluster randomised trial (Solomon 2014). This should be clearly understood as a change in the methodological approach of evaluation of community wide interventions. Each of these studies used a random selection of participants (representative sample) from the communities to participate in the measurement of outcomes.

All included studies were assessed for their risk of bias. Graphical presentation of the results of the risk of bias assessments of the individual studies and of the overall body of evidence are found in Figure 3 and Figure 4. In the earlier review no studies were identified as low risk of bias, however in this update four of the eight studies have been identified as low risk (Kamada 2013; Phillips 2014; Solomon 2014; Wilson 2014). Overall, 19 studies were identified as being at a high risk of bias (Brown 2006; Brownson 2004; De Cocker 2007; Gao 2013; Gu 2006; Guo 2006; Jenum 2006; Kumpusalo 1996; Lupton 2003; Mead 2013; Nguyen 2012; NSW Health 2002; O'Loughlin 1999; Osler 1993; Reger‐Nash 2005; Simon 2008; Wendel‐Vos 2009; Young 1996; Zhang 2003). Ten studies were found to have an unclear risk of bias (Brownson 2005; Eaton 1999; Goodman 1995; Jiang 2008; Kloek 2006; Luepker 1994; Nafziger 2001; Nishtar 2007; Rissel 2010; Sarrafzadegan 2009). Of those studies judged as at either high or unclear risk of bias only one of the studies was randomised, thus selection bias was a major risk for these studies. This was exacerbated as many of these studies only included one measurement point pre‐intervention and one post‐intervention, and in a number of the studies there were differences in important baseline characteristics between the study groups. We observed minor methodological deviations such as a change in the method of application of the survey questions from baseline to follow‐up (for example Phillips 2014). Where a singular minor methodological issue occurred which was deemed unlikely to change interpretation of the findings, we determined that an overall downgrading of the study to high risk was unwarranted.

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 4

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

For the studies deemed to be low risk, allocation to the intervention and control occurred by randomisation (for example cluster RCT) rather than by purposeful allocation of the intervention community to communities which had the capacity to undertake the intervention rather than those which did not, such as Gao 2013. Non‐randomised controlled trials could also have been assessed as lower risk if the measurement was repeated pre and post‐intervention (to determine whether the changes were a result of trends toward the mean or the result of imprecision of the outcome measures). Low risk studies used measurement metrics that were both valid and reliable for population level interventions, avoided subjective self‐report assessment, and typically made over more than one day. Further, the individuals sampled should be representative of the population and include those difficult to reach. Studies at low risk of bias should, in the publication of results, include all of the measures stated in the study protocol and all of those reported in the initial publication of the study.

Selection bias

Selection bias was a major concern in the earlier review as only one study used randomisation to allocate communities (Simon 2008). Previously, no studies were judged as being at low risk of selection bias, although 19 studies were considered to have an unclear risk of bias (if the groups were comparable at baseline for important potential confounders; and if the assessors judged that if the communities were reversed it was likely that the same outcome would be achieved) (Brownson 2004; Brownson 2005; De Cocker 2007; Eaton 1999; Goodman 1995; Gu 2006; Guo 2006; Jenum 2006; Jiang 2008; Kloek 2006; Luepker 1994; Nafziger 2001; Nishtar 2007; NSW Health 2002; O'Loughlin 1999; Osler 1993; Sarrafzadegan 2009; Simon 2008; Zhang 2003). In this update, two of the new studies were identified as being at high risk of selection bias (Gao 2013; Nguyen 2012) and three unclear (Mead 2013; Phillips 2014; Rissel 2010). Four new randomised studies were considered to be at low risk of selection bias (Kamada 2013; Solomon 2014; Wilson 2014).

Performance bias

Collectively, 15 studies were judged as having a low risk of performance bias (Brownson 2005; Eaton 1999; Guo 2006; Jiang 2008; Kamada 2013; Kloek 2006; Luepker 1994; Nafziger 2001; Nishtar 2007; O'Loughlin 1999; Phillips 2014; Reger‐Nash 2005; Sarrafzadegan 2009; Wendel‐Vos 2009; Wilson 2014). While information on the blinding of communities was rare, these studies were judged as being at low risk of contamination and provided evidence of good integrity in the delivery of the intervention even though in some circumstances the intervention was clearly weak.

Attrition bias

Nineteen studies were assessed as being at low risk of attrition bias (Brown 2006; De Cocker 2007; Eaton 1999; Gao 2013; Goodman 1995; Jiang 2008; Kamada 2013; Luepker 1994; Mead 2013; Nafziger 2001; Nguyen 2012; Nishtar 2007; Phillips 2014; Rissel 2010; Sarrafzadegan 2009; Simon 2008; Solomon 2014; Wilson 2014; Zhang 2003). Potential for attrition bias was often not applicable through the cross‐sectional sampling of different individuals as representatives of the same population rather than following specific individuals through time. Some cohort studies had very high completion rates possibly related to recruitment intention of being resident in the community for the duration of the study (Mead 2013; Rissel 2010). There were no cases of communities withdrawing from the studies.

Detection bias

Twelve studies had a high risk of detection bias, 14 an unclear risk and 7 were low risk (Gao 2013; Kamada 2013; Kloek 2006; Nishtar 2007; Rissel 2010; Sarrafzadegan 2009; Solomon 2014). Assessment of detection bias included an assessment of the validity of the measurement tools and the quality of the outcome measures. In this update, one study used accelerometers to objectively measure physical activity.

Reporting bias

Four studies had a high risk of reporting bias (Brown 2006; Gu 2006; Jenum 2006; Mead 2013), with three assessed as being unclear (Guo 2006; Nafziger 2001; Zhang 2003) and 26 as low risk of bias. In the studies judged as having a high risk of reporting bias, there was evidence to indicate that outcomes important to the study were collected but not reported (as confirmed through communication with the authors). Ideally, access to study protocols would help with the process of accessing reporting bias, however in most cases this was not possible. Some studies did publish papers describing the intervention and evaluation methods prior to the final evaluation of the study thus enabling some scrutiny of reporting bias. Some studies with negative findings provided limited reporting of the outcomes and a preference towards the higher quality measurement instruments (for example Phillips 2014; Wilson 2014); however, with no likely impact upon the conclusions we determined them low risk for reporting bias.

Other bias

One study was judged as being at high risk of other bias (Brownson 2004), having had a 'head‐start' with several years of preparation in the intervention community prior to the program start, which was deemed to provide it with an advantage. The effect of this bias was unpredictable as it could have resulted in a null effect or been an effect modifier.

Effects of interventions

Physical activity, dichotomous outcomes

Twenty‐seven studies reported physical activity as some form of dichotomous measure.

Fourteen studies reported physical activity measured as the attainment of a predefined amount of physical activity (Brown 2006; Brownson 2005; Gao 2013; Jiang 2008; Kamada 2013; Kloek 2006; Lupton 2003; NSW Health 2002; Phillips 2014; Reger‐Nash 2005; Rissel 2010; Sarrafzadegan 2009; Solomon 2014; Wendel‐Vos 2009) (Table 5; Figure 5). Only two of these studies, both based in China, found the intervention to be collectively effective across the whole population, in an intense intervention in urban Beijing (Jiang 2008) and Hangzhou China (Gao 2013). Lupton 2003 and Brown 2006 found the interventions to be effective in the male and female populations of the targeted communities respectively. The remaining studies found no evidence of effect.

Figure 5

Forest plot of dichotomous outcomes of meeting a criteria of being physically active ‐ mixed measures and study designs by risk of bias.

Table 5. Dichotomous outcomes ‐ physical activity

Study	Overall bias	Measure	Definition	Net % change	Unadjusted RD	Adjusted RD (95% CI)	Unadjusted RR (95% CI)	Adjusted RR (95% CI)	Baseline
Brown 2006	High risk of bias	% physically active	150 minutes of activity in at least 5 separate sessions in the last week	15.40	0.9	7.33 (‐23.48 ‐ 38.13)	1.02	1.18 (0.60 ‐ 2.35)	41.9
Gao 2013	High risk of bias	% Moderate or high physically active	Categories on IPAQ	3.34	7.4	2.50 (1.17 ‐ 3.83)	1.10	1.03 (1.01 ‐ 1.05)	70.5
Jiang 2008	Unclear risk of bias	Regular physical activity	Not provided	18.12	6.38	10.75 (5.23 ‐ 16.27)	1.24	1.20 (1.09 ‐ 1.31)	60.39
Kamada 2013	Low risk of bias	% physically active	Engaging in 150mins/week or more of walking, engaging in daily flexibility or engaging in 2 or more days a week of in muscle strengthening activities (All groups vs. control)	‐0.17	‐1.6	0.00 (0.0‐0.0)	0.973	1.00 (0.99‐1.00)	63.0
	Low risk of bias	% physically active	Engaging in 150mins/week or more of walking, engaging in daily flexibility or engaging in 2 or more days a week of in muscle strengthening activities (Aerobic exercise group vs. control)	‐2.80	0.000	‐2.0	1.00	0.97	66.6
	Low risk of bias	% physically active	Engaging in 150mins/week or more of walking, engaging in daily flexibility or engaging in 2 or more days a week of in muscle strengthening activities (Aerobic exercise and strengthening group vs. control)	0.41	‐0.3	0.30 (‐4.56 ‐ 5.16)	1.00	1.00 (0.94 ‐ 1.08)	64.0
Kloek 2006	Unclear risk of bias	% physically active	At least 30 minutes of moderate‐intensity physical activity on at least 5 days a week	‐7.36	‐1	‐3.97 (5.02 ‐ ‐12.95)	1.04	0.93 (0.79 ‐1.10)	59.0
Lupton 2003	High risk of bias	% physically active	Minimum of four hours of weekly moderate PA during the last year	9.84	8.3	6.87 (‐13.04 ‐ 26.78)	0.98	1.10 (0.84 ‐ 1.43)	72.5
NSW Health 2002	High risk of bias	% physically active	Engaged in at least 150 minutes and five sessions of moderate activity or three sessions of vigorous activity per week	7.14	‐0.2	3.39 (‐0.29 ‐ 7.08)	1.14	1.08 (0.99 ‐ 1.17)	49.2
Phillips 2014	Low risk of bias	% meeting	Physical activity: 5x30 minutes per week	7.89	1.9	5.00 (‐2.879 ‐ 12.879)	1.029	1.079 (0.957 ‐ 1.216)	63.4
Reger‐Nash 2005	High risk of bias	% physically active	Moderate activity at least 30 minutes for at least 5 days per week or vigorous activity at least 20 minutes for at least 3 days per week	0.36	1.2	0.38 (‐0.06 ‐ 0.82)	1.15	1.01 (0.10 ‐ 1.01)	46.9
Rissel 2010	Unclear risk of bias	% physically active	undertaking 150 min/week	‐5.55	‐5.0	‐2.8(‐6.47 ‐ 0.873)	0.907	0.951 ( 0.891 ‐1.015)	44.9
Sarrafzadegan 2009	Unclear risk of bias	% physically active	Individuals with >= 30 minutes/day of moderate or vigorous activity	4.17	2.1	1.89 (‐0.23 ‐ 4.02)	1.07	1.06 (1.00 ‐ 1.14)	47.0
Solomon 2014	Low risk of bias	% physically active	Did sufficient physical activity to meet the current United Kingdom physical activity guidelines (at least 150 minutes of moderate‐intensity activity per week in bouts of 10 minutes or more, or at least 75 minutes of vigorous intensity activity per week	1.03	NA	NA	NA	1.02 (0.88 ‐ 1.17)†	66.9
Wendel‐Vos 2009	High risk of bias	% physically active	150 min/week and at least 5 sessions per week, and physically active at least 30 min/day at least 5 days a week	‐3.50	‐0.7	‐1.60 (‐0.10 ‐ ‐3.10)	0.86	0.97 (0.93 ‐ 1.00)	42.8

RD = Risk difference

RR = Relative Risk

† Data as presented by the study authors. Odds ratio of adjusted comparison (Intervention minus control in stepped wedge cluster randomised controlled design, p‐value = 0.80, ICC 0.008. Baseline represents baseline for all..

Jiang 2008 reported an increase in regular physical activity (we calculated an adjusted RR 1.20, 95% CI 1.09 to 1.31) for an intervention involving intensive contact with individuals in urban communities in Beijing. The intervention had very substantial penetration into the community with quarterly door‐to‐door distribution of handouts, counselling by health practitioners, and the identification of those within the community with high risk factors through an intensive individual screening campaign in which 73% of the community participated. Gao 2013 also reported a small but statistically significant increase (adjusted RR 1.03, 95% CI 1.01 to 1.05). This intervention was a multi‐component high intensity intervention and the study was at high risk of bias as the authors allocated communities to the control arm which did not have the capacity to support the intervention.

The Finnmark Intervention study (Lupton 2003) aimed at improving cardiovascular health in a small arctic community in Norway, and reported a significant increase (P = 0.047) in males being physically active, as defined as accruing a minimum of four hours of moderate physical activity over a week during the last year. This was measured six years after the initial baseline measurement and commencement of an intervention which involved the engagement of the community largely through activities run by sporting clubs and associations. Unfortunately, no significant change was found in the female population (P = 0.151) as reported by the authors and the calculated adjusted RR for the entire population was non‐significant (RR 1.10, 95% CI 0.84 to 1.43).

Conversely, the Rockhampton 10,000 Steps Project conducted in a regional Australian community found an increase in the proportion of physically active females (achieving 150 minutes of activity in at least five separate sessions over the last week) but not males (Brown 2006). The interpretation of these findings was complicated as the control community was significantly more active than the comparison community at baseline (OR 0.77, 95% CI 0.65 to 0.93). At follow‐up, two years later, there was no longer a significant difference with the percentage of the comparison community categorised as being active decreasing by 6.4% while the intervention community increased 0.9%. Combined, there was once again no difference between the two populations (adjusted RR 1.18, 95% CI 0.60 to 2.35).

No evidence of effectiveness was found in the three studies at low risk of bias. Phillips 2014 found no increase in the percentage of people meeting the target of 5 x 30 minutes per week (adjusted RR 1.03, 95% CI 0.96 to 1.22) and, similarly, Solomon 2014 did not find an increase in the percentage meeting the UK recommendation of at least 150 minutes of moderate‐intensity activity per week in bouts of 10 minutes or more, or at least 75 minutes of vigorous‐intensity activity per week (RR 1.02, 95% CI 0.88 to 1.17). Further, in Japan Kamada 2013 in three comparisons, controlled versus muscle strengthening versus aerobic activity versus combined, found no statistical increases in either arm of the intervention analysed (adjusted RR 1.00, 95% CI 0.99 to 1.00; RR 0.97 (confidence interval could not be calculated); RR 1.00 95% CI 0.94 to 1.10).

The Isfahan Healthy Heart program aimed to improve the health of a large population (> two million) through a multi‐strategic, large scale intervention (Sarrafzadegan 2009). The adjusted RR of 1.06 (95% CI 0.99 to 1.14) suggested a small increase in the percentage of the population with greater than, or equal to, 30 minutes per day of moderate or vigorous activity, although this was not found to be statistically significant. This result needs to be understood in the context of a decreasing trend in physical activity in both the intervention and comparison groups. Further, for the continuous outcome energy expenditure, a decrease was observed.

Wendel‐Vos 2009 reported no effect on the percentage of participants meeting the study's target of 150 minutes per week and at least five sessions per week in the Maastricht region of the Netherlands, following a large five‐year project aiming to improve individuals' chronic disease risk factors (adjusted RR 0.97, 95% CI 0.93 to 1.0). Also, targeting several health‐related behaviours, Kloek 2006 reported on an intervention targeting socioeconomically deprived neighbourhoods in Eindhoven, the Netherlands. No effect was found on the proportion of the population attaining at least 30 minutes of moderate‐intensity physical activity on at least five days in a week (adjusted RR 0.93, 95% CI 0.79 to 1.10).

In investigating a mass media dominated intervention aimed at increasing walking behaviour, Reger‐Nash 2005 found no effect on moderate activity of at least 30 minutes for at least five days per week or on vigorous activity for at least 20 minutes on at least three days per week (adjusted RR 1.00, 95% CI 1.00 to 1.01).

NSW Health 2002 reported no statistically significant effects on physical activity, defined as those individuals engaged in at least 150 minutes and five sessions of moderate activity or three sessions of vigorous activity per week, for a short intervention aimed at increasing the use of parks and walking. The calculated adjusted RR was 1.08 (95% CI 0.99 to 1.17) with the interpretation of this finding complicated by a decrease in physical activity attainment in both the intervention and the comparison communities. This was demonstrated with the risk difference (RD) for the intervention being ‐0.2. Similarly, Rissel 2010, with an emphasis on cycling, used the same outcome measures and found no increase (adjusted RR 0.95, 95% CI 0.89 to 1.02).

A further study did report on the number of people involved in physical exercise, however we could not obtain a definition of physical exercise (Guo 2006). Given this, interpretation of the results of this study conducted in rural villages in China was difficult (and this study was not included in Table 5). This was further complicated as the villages were not comparable at baseline for the number of people undertaking physical activity (34.6%, 95% CI 29.7 to 40.2; 6.2%, 95% CI 12.2 to 20.8). The study did conclude there was a significant difference in the number of people undertaking physical exercise between the intervention and control villages over the period of the study (change of 27%; P value not found).

Three studies reported the measure of leisure time physical activity (Kumpusalo 1996; Luepker 1994; Nishtar 2007) (Table 6; Figure 6). Two studies, one set across a large region in Pakistan (Nishtar 2007) and the other in Finnish villages (Kumpusalo 1996), found no evidence of effect. One of these studies, the Minnesota Heart Health Program, found some evidence of effectiveness although this was not consistent across the different sampling methods used in the study nor over the time span of data collection (Luepker 1994).

Figure 6

Forest plot of dichotomous outcomes of meeting a criteria of being physically active during leisure time ‐ mixed measures and study designs.

Table 6. Dichotomous outcomes ‐ physical activity during leisure time

Study	Overall bias	Measure	Definition	Net % change	Unadjusted RD	Adjusted RD (95% CI)	Unadjusted RR	Adjusted RR (95% CI)	Baseline
Luepker 1994	Unclear risk of bias	PA during leisure time	Regularly active during leisure time	^a11.26	8.5	5.35 (‐3.32 ‐ 14.02)	1.08	1.11 (0.94 ‐ 1.30)	48.6
Luepker 1994	Unclear risk of bias	PA during leisure time	Regularly active during leisure time	^b9.4	4.3	4.70 (‐1.64 ‐ 11.04)	1.09	1.08 (0.97 ‐ 1.20)	49.4
Kumpusalo 1996	High risk of bias	PA during leisure time	Undertaking physical activity during leisure time > 3 times weekly	‐1.76	0.6	‐0.64 (‐8.24 ‐ 6.96)	1.02	0.98 (0.80 ‐ 1.21)	39.0
Nishtar 2007	Unclear risk of bias	PA during leisure time	Not provided	‐25.58	2.5	0.52 (‐0.04 ‐ 1.08)	2.41	0.88 (0.77 ‐ 1.02)	3.0

^adata from independent surveys

^bdata from cohort surveys

RD = Risk difference

RR = Relative Risk

Luepker 1994 reported the findings of a large scale, high intensity, long‐term cardiovascular disease prevention intervention called the Minnesota Heart Health Program. In this study, six communities were matched, with one community of each pair non‐randomly selected to receive this large scale, five to six‐year intervention. Independent cross‐sectional samples of 300 to 500 randomly selected adults were surveyed periodically, including multiple measurements during the 16‐month baseline period and then at one, three, five and six years post‐implementation. Concurrently, a cohort randomly selected from the pre‐intervention cross‐sectional surveys (n = 7097) were re‐surveyed at baseline, two, four and seven years post‐intervention (end of study follow‐up 67.1%), although alternate halves of the cohort group were surveyed at two and four years. The authors presented the pooled data at the various measurement points adjusted for age, gender and education. They reported that the cross‐sectional surveys found the intervention communities to have a significantly greater proportion of the population being physically active during leisure time at one and three years; at five and six years there was no longer a statistically significant difference despite trending higher (P values not provided). The cohort data found no significant differences at two and four years, however there was a statistically significant difference at seven years post‐intervention (P values not provided). The adjusted RR calculated using data extracted from year zero and the final year of measurement was 1.11 (95% CI 0.94 to 1.30) for the cross‐sectional data and 1.08 (95% CI 0.97 to 1.20) for the cohort data, respectively.

Nishtar 2007 reported on the Heartfile Lodhran CVD project aimed at cardiovascular disease prevention in Pakistan. The authors reported no change in leisure time physical activity (adjusted RR 0.84, 95% CI 0.70 to 1.02).

In a study set in Finnish villages (Finnish Healthy Village Study), Kumpusalo 1996 found that the intervention was not associated with improvements in the physical activity patterns of people living in rural villages. The adjusted RR was 0.98 (95% CI 0.80 to 1.21).

An additional study reported on the effectiveness of an intense community intervention in Shandong, China for the similar outcome of non‐occupational physical activity (Zhang 2003). This study found no difference in the relative proportion of the intervention community found to be physically active pre and post‐measurement (P > 0.05), although over the same time the authors reported a significant reduction in the proportion of the control community who were physically active (P < 0.05).

Seven studies reported a dichotomous measure of physical inactivity, that is the proportion of people who failed to attain a defined level of activity (Eaton 1999; Gao 2013; Goodman 1995; Jenum 2006; Nafziger 2001; Nguyen 2012; Osler 1993) (Table 7; Figure 7). Of the remaining studies, the Romsas in motion study showed some evidence that the three‐year, multi‐strategic intervention was effective at decreasing the proportion of a population in a low socio‐economic district in Oslo, Norway not engaging in heavy physical activity (Jenum 2006). Eaton 1999, Nafziger 2001, Osler 1993 and Goodman 1995 all found the community wide interventions that they investigated not to be effective.

Figure 7

Forest plot of dichotomous outcomes of meeting a criteria of being physically inactive ‐ mixed measures and study.

Table 7. Dichotomous outcomes ‐ physically inactive or sedentary

Study	Overall bias	Measure	Definition	Net % change	Unadjusted RD	Adjusted RD (95% CI)	Unadjusted RR	Adjusted RR (95% CI)	Baseline
Eaton 1999	High risk of bias	Physically inactive (%)	used different criteria for physical inactivity combined questions XS1, XS2 (exercise <1 times per week on average), XS5 & XS6 (=0 days of sweat related physical activity) (not validated measures). Men <=35 yo	1.77	‐20.3	0	1.08	1.03	50.2
			Men >35	0.09	1.9	0	0.98	1.00	53.6
			Women <=35 yo	‐0.18	‐8.2	0	0.99	1.00	56.2
			Women >35 yo	‐0.09	‐6.1	0	0.99	1.00	62.6
Gao 2013	High risk of bias	Low physical activity (%)	IPAQ category	‐8.76	‐7.3	‐2.30 (‐3.39 ‐ ‐1.01))	0.77	0.91 (0.86 ‐0.95)	29.4
Goodman 1995	Unclear risk of bias	Physically inactive (%)	Physically inactive was defined as engaging in no physical activity or exercise during the last month	‐1.82	‐1.7	‐1.02 (‐3.03 ‐ 0.99)	0.97	0.99 (0.96 ‐ 1.01)	44.6
Jenum 2006	High risk of bias	Physically inactive	No heavy physical activity in leisure time or commuting (%)	20.04	12.9	8.13 (5.25 ‐ 10.99)	1.36	1.20 (1.12 ‐1.28)	40.5
Nafziger 2001	Unclear risk of bias	Physically inactive (%)	Described as self‐reported sedentary life‐style unless they were involved in a physical activity strenuous enough to work up a sweat <3 times/week	‐15.85	‐7.8	‐11.43 (‐23.06 ‐ 0.21)	0.89	0.84 (0.71 ‐ 1.00)	72.5
Nguyen 2012	High risk of bias	Physically inactive (%)	Total physical activity less than 3,000 MET‐minutes per week (all)	71.26	4.9	5.63 (2.64 ‐ 8.62)	1.51	1.65 (1.26 ‐ 2.16)	8.0
			Total physical activity less than 3,000 MET‐minutes per week (men)	45.24	5.8	4.82 (0.99 ‐ 8.65)	1.50	1.35 (1.06 ‐ 1.72)	10.0
			Total physical activity less than 3,000 MET‐minutes per week (women)	93.44	4.44	6.43 (1.85 ‐ 11.02)	1.52	1.98 (1.21 ‐ 3.24)	6.8
Osler 1993	High risk of bias	Physically inactive (%)	No details provided	20.51	0.1	2.07 (‐125.30 ‐ 129.45)	1.00	1.16 (extreme 95% CI crossing 1)^a	13.0

This table contains data where % inactive, not undertaking sufficient PA, or classed as achieving a level of PA defined as low or inactive.

Eaton 1999 did not provide a sufficient breakdown of the respective sample sizes to calculate 95% CI. Author was contacted for further details, no reply was provided.

RD ‐ Risk difference

RR ‐ Relative Risk

a Due to small numbers, the confidence interval for the adjusted 95% using Wald‐test yields extreme values.

The Romsas in motion study was a controlled before and after study with a cohort follow‐up panel (Jenum 2006). After a three‐year follow‐up it reported that the percentage of respondents not achieving heavy physical activity sufficient to make them sweat and feel out of breath was significantly smaller in the intervention population, with a pre‐post reduction during the study period in the intervention district of 8.1% (95% CI 2.4 to 13.8; P = 0.005). However, the calculated adjusted RR for the whole study was 0.8 (95% CI 0.59 to 1.08). As has been the case with other studies, these findings were complicated by the differences between the two communities at baseline. In this situation, the intervention community had a 5% higher baseline inactivity proportion as compared to the control community.

The Ostego‐Schoharie health heart program targeted the prevention of cardiovascular disease in rural USA through a hospital based intervention. This study collected both cross‐sectional data and cohort data at baseline and at five‐year follow‐up (Nafziger 2001). The cross‐sectional data were reported as a non‐significant reduction in self‐reported physically inactive lifestyle in the intervention population. Our analysis of the extracted results found an adjusted RR of 0.84 (95% CI 0.71 to 1.00). The cohort data also reported no evidence of effect with both the intervention and control communities decreasing in the proportion found to be inactive (P > 0.05).

The Osler 1993 study reported an increase in physical inactivity in both the intervention and control communities of rural municipalities in Denmark. The calculated adjusted RR of 1.16 (having an extreme 95% CI crossing 1), which suggested the intervention group was more physically inactive after the intervention as compared to the control group, was not statistically significant. Nguyen 2012, in rural Vietnam, found a significant failure of the intervention at a population level (adjusted RR 1.65, 95% CI 1.16 to 2.16), less so for men (adjusted RR 1.35, 95% 1.06 to 1.72) and most detrimental for females (adjusted RR 1.98, 95% CI 1.21 to 3.24), from relatively low levels of physical activity. The intervention fared worse than the control in advancement of inactivity within the community.

Goodman 1995 also found no difference between the intervention and control groups for physical inactivity in a chronic disease prevention project in an urban US setting (adjusted RR 0.99, 95% CI 0.96 to 1.01).

Three studies reported leisure time physical activity (Kumpusalo 1996; Nishtar 2007; O'Loughlin 1999). None demonstrated evidence of effectiveness.

Nishtar 2007 investigated an intervention aimed at increasing the physical activity levels in a large regional population in Pakistan. The investigators found no difference between the intervention and comparison populations in recreation or leisure time physical inactivity (P values not reported). Similarly, Kumpusalo 1996 reported no difference in leisure time physical inactivity in the Finnish Healthy Village study (P > 0.05) and O'Loughlin 1999 found no difference in an intervention targeting a low income, inner city neighbourhood in Montreal, Canada (P = 0.063).

Two studies reported the attainment of vigorous activity (NSW Health 2002; Young 1996).

The Stanford five‐city project, based in California, found inconsistent and limited intervention effects between intervention cities and control cities for behavioural measures of physical activity (Young 1996). In this study, independent cross‐sectional surveys were conducted at baseline, 25, 51 and 73 months (n = 1800 to 2500 participants). Those who participated at baseline also comprised a cohort who were sampled at 17, 39 and 60 months (n = 907). The percentage of men who regularly engaged in at least one vigorous activity did significantly differ over time between the treatment and control cities (P < 0.004), although this increase was not found in the cohort sample (P = 0.068) nor in an independent (P = 0.237) or cohort sample of women (P = 0.842).

The NSW Health study also reported the percentage of people engaging in physical activity and found no difference between the intervention and treatment groups (P = 0.077) (NSW Health 2002).

Physical activity, continuous outcomes

Eleven of the included studies reported continuous measures of physical activity.

Three studies reported leisure time physical activity measured by time (De Cocker 2007; Simon 2008; Wendel‐Vos 2009) (Table 8) with each of the three studies showing some evidence of effectiveness, however only Simon 2008 reported an increase in physical activity levels.

Table 8. Continuous outcomes ‐ leisure time physical activity

Study	Measure	Subgroup	Post mean difference	Adjusted mean difference	Adjusted % change relative to the control mean	Baseline value	Timeline
De Cocker 2007	Leisure time PA (hours/week)	No subgroup‡	0	0.53	25.60	2.33	1 year
Simon 2008	Supervised leisure time physical activity (hours/week)	Measured only in children†	0.9	1.1	43.14	2.5	4 years
Wendel‐Vos 2009	Leisure time PA (hours/week)	Men (NS)	‐0.2	‐0.4	‐2.06	19.8	5 years
Wendel‐Vos 2009	Leisure time PA (hours/week)	Women‡	‐4.4	2.2	14.01	11.7	5 years

† authors reported a statistically significant increase (P < 0.05) in favour of the community

‡authors reported a statistically significant difference ( P < 0.05), however there was no observed increase in PA

NS ‐ no statistically significant difference

Wendel‐Vos 2009 reported on a regional cardiovascular disease prevention program in Limburg, Netherlands. Total leisure time physical activity was reported for both males and females. Both groups decreased their leisure time physical activity between baseline and follow‐up at five years, with no difference between the intervention and control groups for men. In women, however, the reduction in leisure time physical activity in the intervention group was significantly less then in the control group (P < 0.05).

Leisure time physical activity also decreased from baseline to follow‐up in both the intervention and control communities in the Ghent 10,000 steps study (De Cocker 2007). Importantly, this reduction was significantly greater in the control group than the intervention group (P ≤ 0.05) with the adjusted percentage change calculated as 25.60%. The authors reported that in addition to leisure time physical activity there were significant intervention effects for a range of physical activity outcomes including moderate physical activity (minutes per week) and work‐related physical activity (minutes per week) but not vigorous physical activity, transport‐related physical activity and household physical activity.

Simon 2008 reported the results of a cluster RCT of an intervention based predominantly in a school setting. It reported an adjusted change in supervised leisure time physical activity of 43% in adolescents and an adjusted mean difference of 1.1 hour per week (95% CI 0.56 to 1.63) in leisure time physical activity at four years post‐baseline. This was a statistically significant difference between the intervention and control groups (P < 0.0001).

Four studies reported a continuous measure of walking (Brownson 2004; Brownson 2005; De Cocker 2008; Wendel‐Vos 2009) (Table 9). Two of the studies (De Cocker 2007; Wendel‐Vos 2009) reported some evidence of effectiveness although two that were conducted in the same population in Missouri, USA found no evidence of increased time spent walking (Brownson 2004; Brownson 2005).

Table 9. Continuous outcomes ‐ walking

Study	Measure	Sub group	Post mean difference	Adjusted mean difference	Adjusted % change relative to the control mean	Baseline value	Timeline
Brownson 2004	7 day total walking (mean min/week) (NS)	N/A	‐5.3	‐1.4	‐1.38	97.2	2 years
Brownson 2004	7 day walking for exercise (mean min/week) (NS)	N/A	‐0.1	‐5.6	‐17.61	37.3	2 years
Brownson 2005	Walking (mean min/week) (NS)	N/A	‐0.8	5.2	4.75	97	1 year
De Cocker 2007	Pedometer‐determined (steps/day)†	N/A	957	1030	10.80	9597	1 year
De Cocker 2007	Walking (min/week)†	N/A	34	47	17.34	288	1 year
Wendel‐Vos 2009	Walking (hours/week)‡	Male (NS)	‐0.4	‐1.1	‐12.09	8.5	5 years
Wendel‐Vos 2009	Walking (hours/week)‡	Women	2.4	2.6	38.24	8.9	5 years

NS ‐ authors reported no statistically significant difference

† authors reported a statistically significant increase (P < 0.001) in favour of the intervention community

‡ authors reported as a statistically significant difference ( P < 0.05), however there was no observed increase in PA for the intervention community

In an evaluation of a large, expensive five‐year intervention in a region in the Netherlands, Wendel‐Vos 2009 reported a small decrease in walking time per week in males in the intervention group compared to the comparison group (adjusted change ‐12.09%), however this was not found to be statistically significant (P > 0.05). Despite a reduction in walking hours per week in women from both groups, there was a larger reduction in the control community than the intervention community (adjusted change 29.41%) with the intervention group found to be statistically significantly different (or having less of a reduction) than the control community (P ≤ 0.05).

The Ghent 10,000 steps study reported a statistically significant increase in walking measured with a pedometer (steps per day) (P < 0.01) and self‐reported walking (minutes per week) (P < 0.01). The adjusted changes were 10.8% and 17.34%, respectively (De Cocker 2007).

Two studies conducted in a rural area of Missouri reported measures of walking. Brownson 2004 found no difference between the communities in seven‐day total walking (P = 0.91) and seven‐day walking for exercise (P = 0.37). A later study reported on the mean rates of walking per week and found that the intervention and control communities were not statistically significantly different (P value not reported) (Brownson 2005).

Five studies reported continuous measures of energy expenditure (Gao 2013; Kloek 2006; Phillips 2014; Sarrafzadegan 2009; Solomon 2014) (Table 10).

Table 10. Continuous outcomes ‐ energy expenditure

Study	Measure	Post mean difference	Adjusted mean difference	Adjusted % change relative to the control mean	Baseline value	Timeline
Gao 2013	Physical activity MET ‐ min/week†	462	176	38.09	1204	2 years
Kloek 2006	METs/week (NS)	81	‐241	‐3.54	7253	2 years
Phillips 2014	METS/week score (NS)	4.2	‐113 (95% CI ‐ 847‐ 621)*	n/a	2626 (95% CI 1978 to 3279)	3‐4 years
Sarrafzadegan 2009	Total daily PA (MET‐m/week ± SD) (P<0.01)^b	‐65	‐51	‐8.46	606	3 years
Sarrafzadegan 2009	Leisure time PA (MET‐min/week) (P<0.01)^b	‐2	‐3	‐2.83	85	3 years
Solomon 2014	METs/week (NS)	155	171 (95%CI: ‐16 ‐ 358) p=0.07^a	n/a	2561 (SD=2977)	Stepped wedge design ˜ 12 months follow up

*Adjusted analysis adjusted for age, gender, ethnicity, education, employment, appropriate baseline values

a Adjusted analysis adjusted for period, gender, age and area as reported by the authors

b When compared to the control, the authors reported a statistically significant decrease in physical activity for the intervention group.

† authors reported a statistically significant increase (P < 0.05) in favour of the intervention

MET‐ metabolic equivalent of task

The Isfahan Healthy Heart program aimed to improve the health of a large population (> two million) through a multi‐strategic, large‐scale intervention (Sarrafzadegan 2009). This study reported total daily physical activity as well as leisure time physical activity, expressed as metabolic equivalent of task (MET), in minutes per week. The MET is commonly used as a means of expressing the energy cost of physical activity as the ratio of the metabolic rate of any activity to the metabolic rate at rest. The total daily physical activity (MET) decreased in both the intervention and comparison areas over the three years of evaluation. This decrease was significantly greater in the comparison area then the intervention area (‐114 versus ‐68 MET minutes per week; P < 0.05). The intervention and control areas did increase for leisure time physical activity (MET) with the difference at the final evaluation being significantly different (P < 0.01) with an adjusted change of 12.26%.

Kloek 2006 reported on an intervention targeting deprived neighbourhoods in Eindhoven, Netherlands. The study found no evidence of an increase in energy expenditure in the intervention group as compared to the comparison groups at two years post‐baseline (P = 0.95). In the UK, both Phillips 2014, using an intervention targeting socio‐economically disadvantaged neighbourhoods of London, and Solomon 2014, in rural villages of Devon, found no evidence of an effect. However, Gao 2013 using a two‐year intervention in China with communities selected on the basis of their capacity to support the intervention against a control with no capacity found an adjusted mean difference of 176 MET minutes/week.

One study (Wilson 2014) reported the average daily minutes of moderate to vigorous physical activity (MVPA) (Table 11). As a study at low risk of bias, Wilson 2014 reported on an intervention which focused primarily upon an environmental intervention with social marketing emphasising walking and access to walking trails in underserved African American communities. The study measured the average daily minutes of moderate to vigorous physical activity using two methods: with accelerometry, and four‐month recall. Data on the individual level accelerometry, noted by the authors as representing the program effects upon individuals who were representative of the community, were analysed with a mixed model ANCOVA. They examined potential differences between the community estimates that differed across the communities. The analysis revealed no significant differences by communities from baseline, months 12, 18 and 24 for MVPA, indicating that the intervention did not have a broader effect at a population level. However, the authors did observe a more immediate intervention or program level impact of the walking programs using attendance and stationary observations of walking. The multi‐strategy community increased from 40 to 400 walkers per month by 9 months, and the intervention program level effects were sustained with over 200 walkers per month on average. The two samples analysed were not linked.

Table 11. Continuous outcomes ‐ physical activity

Study

Measure

Estimate of effect for full intervention

Baseline value

Timeline

Wilson 2014

Average daily minutes of moderate to vigorous PA

Measured by acceleratory (7‐day estimates) (NS)

0.69 (SE 0.39); 95% CI: ‐0.14 ‐ 1.39)

35.96 (SE 4.17)

Measured 6, 12 & 24 months

Wilson 2014

Average daily minutes of moderate to vigorous PA

self‐reported 4 month recall

data unavailable, only more robust measurement of PA reported

SE ‐ standard error

NS ‐ authors reported no statistically significant difference

Average daily minutes of MVPA is on the square root scale. Analysis as reported by the author using a mixed model ANCOVA implemented to examine difference between communities in accelerometer‐assessed MVPA, random effects for individuals over time.

We noted that the self‐reported measures of MVPA were not included in the published reports. It seems the authors opted to publish only the more valid accelerometry measures, which were neither clinically meaningful nor statistically significant.

More intense studies

Ten of the studies included in the review were classified as being of high intensity based upon the subjective assessment described in the methods section (Brown 2006; Eaton 1999; Gu 2006; Jiang 2008; Luepker 1994; Lupton 2003; Nafziger 2001; Wendel‐Vos 2009; Wilson 2014; Zhang 2003).

Several of these studies reported some improved physical activity outcomes (Brown 2006; Jiang 2008; Luepker 1994; Lupton 2003; Zhang 2003) however this finding was inconsistent, with several studies finding no effect (Eaton 1999; Nafziger 2001; Wendel‐Vos 2009; Wilson 2014) and one study selectively not reporting the study outcome of physical activity (Gu 2006).

Higher quality studies

Four newly published studies were deemed to be high quality studies (Kamada 2013; Phillips 2014; Solomon 2014; Wilson 2014), however none reported evidence of effect upon community levels of physical activity. Wilson 2014 reported an immediate program level effect of more walking.

Eleven studies were assessed as having unclear risk of bias (Brownson 2005; Eaton 1999; Goodman 1995; Jiang 2008; Kloek 2006; Luepker 1994; Nafziger 2001; Nishtar 2007; Rissel 2010; Sarrafzadegan 2009; Simon 2008). Of the 11 studies with unclear risk of bias, only three studies reported some evidence of effect (Jiang 2008; Luepker 1994; Simon 2008).

Equity pointers

In the data extraction we sought to identify studies which had conducted analyses of outcome measures by subgroups of socio‐economic disadvantage such as income, education, occupation, ethnicity and other proxy measures of economic status. Brownson 2004 presented results stratified by whether respondents had a high school certificate or less, whether they had household incomes ≤ USD 20,000 or were African American respondents. In no instance was the net intervention effect statistically significant within these strata for the two outcomes measured in the study (seven‐day total walking, seven‐day walking for exercise). Wendel‐Vos 2009 reported the outcomes of time spent in leisure time physical activity and walking (adjusted for age) for communities stratified into low educational level (intermediate secondary education or less) and moderate or high educational level (higher secondary educational, and higher vocational education or university). In this analysis differences between the intervention and comparison communities were not significant except in walking hours per week in males where the intervention community stayed constant while the control community significantly decreased (P ≤ 0.05) over the period of the study (P ≤ 0.05) as reported by the authors. No other studies had analyses by socio‐economic subgroups that we could identify, although a number of interventions were set or were targeted at areas of deprivation, disadvantage or low socio‐economic status (Brownson 2004; Brownson 2005; Eaton 1999; Jenum 2006; Kloek 2006; Kumpusalo 1996; Lupton 2003; Nafziger 2001; Mead 2013; O'Loughlin 1999; Phillips 2014; Reger‐Nash 2005; Wendel‐Vos 2009; Wilson 2014). Eight of the included studies were also undertaken in low middle or high middle income countries (Gao 2013; Gu 2006; Guo 2006; Jiang 2008; Nishtar 2007; Nguyen 2012; Sarrafzadegan 2009; Zhang 2003).

Several studies did provide results analysed by gender (Brown 2006; Eaton 1999; Kumpusalo 1996; Lupton 2003; Wendel‐Vos 2009; Young 1996). Eaton 1999 presented results grouped by age (< 35 and > 35 years, categories described by the authors) and by sex, with significant differences between age (P = 0.001) and sex (P = 0.001) being identified for physical inactivity. Over the course of the study, men under the age of 35 years decreased physical activity significantly more than men over 35 years and women (both age groups), although there was no difference between the intervention and comparison cities. As already outlined above, time spent in leisure time physical activity and walking (adjusted for age and educational level) as reported by Wendel‐Vos 2009 decreased in both the control and intervention communities over the period of the study, however there was significantly less reduction in the intervention community compared to the control community in females (P ≤ 0.05) than in males (P ≥ 0.05). Brown 2006 provided data on the proportion of the population of the intervention and control communities being physically active, for males and females. The investigators concluded that there was a different pattern between the sexes with the proportion of males in the intervention community categorised as being physically active decreasing by 4.2% (95% CI ‐10.1 to 1.7) compared to females where the proportion increased by 5% (95% CI ‐0.6 to 10.6). In a fishing village in Northern Norway, Lupton 2003 investigated the efficacy of an intervention aimed at improving the risk factor profile of the population. The proportion of males and females in the intervention group increased over the three‐year study as compared to the control population, however this was only statistically significant in the male population (P = 0.047). In the Stanford Five City Project, Young 1996 presented the results of each of the intervention and control cities by men and women. Intervention effects of behavioural improvement were limited and not always consistent between intervention cities, however the percentage of men who regularly engaged in vigorous activity was significantly different over time between the intervention and comparison cities (P < 0.004) in the independent sample (there was also a cohort sample). Kumpusalo 1996 provided results analysed by male and female and for the participating villages. No significant differences were found in any group between the baseline and follow‐up measurements (P > 0.05).

Reach

To be an included study (see Types of participants) each intervention was required to show an intent to be comprehensive in reaching the targeted community. Although intent of reach was required, it was hypothesised using the logic model (Figure 1) that reach (both intended and actual) would differ between the studies and could affect the outcome. There is evidence from some process evaluations that in many community wide interventions not everyone was able to be reached. Goodman 1995 found that African Americans perceived the intervention explored in their study as 'upper class'. Further, there was evidence in the Brown 2006 study, based in Rockhampton Australia, that the intervention was less attractive to men, or that “It didn't speak to men”, a finding that was borne out in the gender differential in the outcomes. Similar findings were also present in Wendel‐Vos 2009. The approach of Simon 2008 was extremely limited in reach as it used 12‐year olds as the target of the intervention and therefore was unlikely to penetrate much beyond the school community. Very few studies described how they recruited participants to events or the intervention strategies. Solomon 2014 found that very few residents were even aware of and participated in the intervention and, not surprisingly, no evidence of an increased prevalence of activity was observed. These studies have accordingly described community levels of physical activity by drawing an evaluation sample using a sampling frame from the community, rather than the actual participants in the events, unlike traditional RCTs.

The absence of reporting by subgroups and process evaluation made the assessment of reach difficult for most studies. Furthermore, as reach is also a component of the assessment of intensity, it was not possible to undertake further interpretation due to the inconsistency of findings when overall intensity was assessed.

Adverse events

None of the included studies reported the occurrence of adverse events.

Discussion

Summary of main results

We updated our previous review with eight new studies, an increase of one‐third in total. Four new studies were at low risk of bias in that the allocation of the intervention was randomly assigned. None of these four low risk studies reported an increase in population levels of physical activity for the community wide intervention. Overall, we still found no consistent evidence to support the effectiveness of multi‐component community wide interventions to increase population levels of physical activity, with the weight of the evidence indicating no increase in physical activity levels. There was still considerable heterogeneity between intervention approaches, intensity of actions delivered, the outcomes assessed and the comparison communities. The overall quality of the remaining studies was poor with the majority assessed as having a high risk of bias. The high risk of bias was largely due to studies with no randomisation to control and comparator groups, the selection and retention of participants, and the use of non‐validated outcome measures. Even amongst the studies at high risk of bias we consistently did not observe positive results. As a group, the interventions undertaken in China appeared to have the greatest possibility of success through high participation rates and may indicate that culturally China retains the potential to increase population levels of physical activity through community interventions, However, in Vietnam where advances in technology and urbanisation are leading to decreased physical activity, the one included study (Nguyen 2012) (at high risk of bias) found the community wide intervention to be problematic, having reported a statistically significant decline for the intervention group. Some of the studies such as Wilson 2014 and Rissel 2010 found a measurable increase in the use of trails and pathways indicating that some people were reached by the program, however increased trail use did not translate into increased population levels of physical activity. Wilson 2014 provides some evidence that social marketing and environmental enhancements together lead to an increase trail usage. Selective outcome reporting bias, identified in Mead 2013 and Gu 2006, may lead to an understatement of the evidence of ineffectiveness or of potential harm of some community wide approaches.

Overall completeness and applicability of evidence

Our review was able to draw upon the best available evidence from studies across the globe, conducted in high and low and high middle income countries. We were also able to successfully obtain additional information and data from study authors. The review shows that the hypothesis that multi‐component community wide interventions effectively increase population levels of physical activity continues to be unsupported by current evidence. Although we found differences in the mix of intervention components deployed by the included studies, one common approach was applicable across most studies. Almost all of the interventions included a component of building partnerships with local governments or NGOs (29 studies). Many also employed some form of individual counselling by health professionals (18 studies), mass media (15 studies) or other forms of communication (18 studies). Fewer studies worked in specific settings (11 studies) or used environmental change strategies (seven studies). Despite some common principles and approaches, of the 10 studies assessed as being of unclear risk of bias only three studies reported some evidence of effect. This finding is also consistent with the finding of no effect in all four low risk of bias studies. There is the potential that selective outcome reporting bias exists in this body of research as two authors appear not to have published all of the outcomes available from the measurement tools they used when the primary finding was negative, and that this co‐exists with publication bias of other studies.

Quality of the evidence

The overall quality of the studies has improved in the past three years, with four assessed as having a low risk of bias by virtue of improved design. All of the designs of the studies were controlled before and after studies with the exception of one controlled interrupted time series (Luepker 1994), one cluster cohort study (O'Loughlin 1999), and four cluster randomised controlled trials (Kamada 2013; Phillips 2014; Simon 2008; Wilson 2014). More sophisticated study designs emerged, including the stepped wedge cluster randomised trial, to accommodate the complexity of the intervention delivery undertaken (Solomon 2014). Selection bias was a main concern as only five studies were randomised. Many studies only had one measurement point pre‐intervention and one post‐intervention, and a number of the control groups had different baseline characteristics compared to the intervention groups. The other common problem related to detection bias as few studies reported the validity of their measurement tools. Validity of the measurement tools is particularly important given the small differences in physical activity reported by some studies. Many studies also relied on self‐reported physical activity measures as these are the most feasible way of collecting data from a large population. However, improvements are underway as in one recent study (Wilson 2014) individual accelerometer estimates were undertaken at baseline, 12, 18 and 24 months. Collectively the newer studies provide evidence that a more robust approach to health promoting interventions is possible. However, some studies failed to report primary outcomes measured post‐intervention.

Potential biases in the review process

One limitation of this review remains, potential publication bias. Other studies may exist but have not been submitted or accepted for publication and therefore were not identified through our searching efforts. The likelihood of this is difficult to judge. Through the new stricter requirements by journals and broader definitions for trial registration, we found evidence of an increasing registration of trials as we were able to use trial registries to determine whether a study had been completed but not published.

Our inclusion criteria required studies to have at least two intervention strategies and this excluded a number of large‐scale mass media interventions. It is possible that these mass media only studies may have included other strategies as part of their approaches but have not reported these activities formally. However, our objective to examine the effects of community interventions that deployed multiple strategies rather than a single strategy approach meant that without evidence of multiple strategies studies were excluded from our review.

Agreements and disagreements with other studies or reviews

The recent Lancet series on physical activity, published in London in 2012, also examined review level evidence for a range of global physical activity interventions in studies published between 2000 and 2011 (Heath 2012) but came to a slightly different conclusion on effectiveness. This review adopted a more mixed approach to typologies of interventions and concluded that the evidence of effectiveness of community interventions was "inconsistent, especially in communities in countries of low to middle income". Heath 2012 presented a reason for the inconsistent interpretation of effectiveness, because they found that more rigorous reviews (including the earlier version of this review) had not included in their reviews "observational studies or investigations with insufficient evidence (not necessarily ineffective)". This observation is indeed correct as design biases (such as an absence of a suitable comparison) would drive more inconstant results and fail to provide substantial evidence for causality as defined by GRADE (Schünemann 2011) and also by the Bradford‐Hill criteria for causality. Our findings differ as we included only studies with an element of controlled design and not those with pre‐post measures only. Our decision to maintain a higher design quality for included studies could be justified not only on the grounds of genuine assessment of impact but also it may in part have contributed to improvements since our last review in the design quality of newly published studies. This has been seen in other Cochrane physical activity reviews where recommendations on study quality are reflected in subsequent generations of studies, for example with longer follow‐up (Foster 2005). Brand 2014, a narrative overview of systematic reviews, found community interventions to be inconclusive in their ability to increase physical activity.

An earlier review by Yang 2010 examined the effectiveness of abroad range of interventions to promote cycling. This review found small positive effects to promote cycling in two city level community intervention studies (Yang 2010). For example, The English Cycle Demonstration Towns programme reported increases in cycling across six towns between 2005 and 2008. Towns opted for different strategies to promote cycling, ranging from mass media campaigns, travel planning, cycle training services and improvements to local cycling infrastructures. Yang 2010 mirrored our findings in the conclusions of their review, as they were also limited by the quality of study design, measures and data analysis. This is a consistent finding with systematic reviews of physical activity interventions; that the limitations of study design and measures probably mask any possible effects of such interventions (Foster 2005; NICE 2008; Ogilvie 2007; Richards 2013). Some of the measures may be useful for surveillance but may not be sensitive to change in intervention studies.

Often cited is Kahn 2002, a systematic review conducted on the effectiveness of a range of interventions intended to increase physical activity, including community wide campaigns. This review found that there was strong evidence that community wide campaigns are likely to be effective in increasing physical activity in the population, assuming that they are modified to target the populations in which they are implemented (Kahn 2002). The systematic review upon which these conclusions were based does not, however, include the latest studies (studies published since the year 2000) and six of the 10 studies that were included in the Kahn 2002 review (Jason 1991; Malmgren 1986; Meyer 1980; Owen 1987; Tudor‐Smith 1998; Wimbush 1998) were excluded from our systematic review for reasons outlined in the excluded studies table (Characteristics of excluded studies). Twelve years later, this Cochrane review presents evidence from recently published and in press studies at low risk of bias, previously unincorporated into any other systematic review. Collectively, the newer studies have trended towards more robust design and also conclusions of an absence of effectiveness for community wide interventions.

The absence of an effect from the most recently published suite of higher quality studies could be explained by the attributes of the intervention, their design, measures and reach. The apparent failure or potential reach of studies and their penetration into their target communities has also been described in a number of recent reviews on recruitment and potential impact of studies on inequalities. The lack of reporting of recruitment and marketing approaches in our studies were also highlighted by Foster et al's reviews of walking interventions (Foster 2011). More worrisome is the lack of data exploring the potential impacts of these community interventions upon specific subgroups, particularly those groups whose physical activity participation is socially patterned (Humphreys 2013). In theory, any potential impact in one group might be masked by a decline in another, and there might be the potential for differential effects of such interventions. In the absence of adequate reporting the impact of population level physical activity interventions on social inequality effects will remain unknown (Humphreys 2013).

Figure 1

Logic Model for Community Wide Interventions for Increasing Physical Activity.

Figure 2

PRISMA diagram based upon Moher 2009.

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 4

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figure 5

Forest plot of dichotomous outcomes of meeting a criteria of being physically active ‐ mixed measures and study designs by risk of bias.

Figure 6

Forest plot of dichotomous outcomes of meeting a criteria of being physically active during leisure time ‐ mixed measures and study designs.

Figure 7

Forest plot of dichotomous outcomes of meeting a criteria of being physically inactive ‐ mixed measures and study.

Community wide interventions for promoting physical activity
Patient or population: whole communities (adults, adolescents and children) Settings: community based Intervention: multi‐component of at least two physical activity interventions targeting the whole community Comparison: existing programmes and infrastructure
Outcomes [duration of follow up]	Summary of effects	Number of communities (studies)	Quality of the evidence (GRADE)
Physical activity % Physically active Intervention compared to control adjusted pre/post cross‐sectional sampling (end of intervention to 6 years)	Typically no evidence of benefit	25 (10)	⊕⊕OO¹ Low
Physical activity % physically active Intervention compared to control adjusted pre‐post cross‐sectional sampling (end of intervention to 3 years, 4 months)	Typically no evidence of benefit	160 (3)	⊕⊕⊕⊕ High
Energy expenditure METS/week score, adjusted mean difference (follow up; end of intervention to 4 years)	Typically no evidence of effect Range: ‐241 to +176	156 (5)	⊕⊕OO¹ Low
Physical activity Average daily minutes of moderate to vigorous (24 months)	No evidence of effect from the baseline of 36 minutes per day	2 (1)	⊕⊕⊕O² Moderate
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Substantial heterogeneity between trials regarding type of interventions and measured outcomes; wide and overlapping range of effects ²Findings based on a single study in only two communities

Table 1. Search results for electronic databases

Database	Number of hits
ASSIA	1144
British Nursing Index (BNI)	105
CINAHL	2881
Chinese atabase:CAJ,CCND,CPCD,CJSS,CMFD,CDFD, http://www.global.cnki.net/grid20/index.htm	124
Cochrane Library	1841
Cochrane Public Health Group Specialized Register	31
EMBASE	4941
EPPI Centre DoPHER TRoPHI	38 200
ERIC	416
Health Management Information Consortium (HMIC)	308
LILACS	416
MEDLINE & MEDLINE In‐Process	5691
PsycINFO	1315
Sociological Abstracts	874
SPORTDiscus	365
Transport Database TRIS	49
Web of Science Science Citation Index, Social Sciences Citation Index and Conference Proceedings Citation Index	9108

Table 1. Search results for electronic databases

Table 2. Search results for websites

Web sites	Hits
EU Platform on Diet, Physical Activity and Health	0
http://health‐evidence.ca	5
IUHPE (International Union for Health Promotion and Education)	0
NCCHTA http://www.ncchta.org	1
NICE guidelines http://www.nice.org.uk	4
SIGN guidelines http://www.sign.ac.uk	0
US Centres for Disease Control and Prevention http://www.cdc.gov/	0
World Health Organisation http://www.who.int/en/	1

Table 2. Search results for websites

Table 3. Categories of strategies included in interventions

Study	Mass Media	Other communication	Individual	Partnerships	Settings	Environmental	Total
Brown 2006	X	X	X	X	X	X	6
Brownson 2004		X	X	X		X	4
Brownson 2005	X	X	X	X			4
De Cocker 2007	X	X		X	X	X	5
Eaton 1999		X		X	X	X	4
Gao 2013	X	X	X	X	X	X	6
Goodman 1995	X	X	X	X	X	X	6
Gu 2006		X	X				2
Guo 2006	X	X	X				3
Jenum 2006	X	X	X	X		X	5
Jiang 2008		X	X	X			3
Kamada 2013	X	X		X	X		4
Kloek 2006	X		X	X	X		4
Kumpusalo 1996		X	X	X	X		4
Luepker 1994	X	X	X	X	X	X	6
Lupton 2003	X		X	X	X		4
Mead 2013	X			X	X		3
Nafziger 2001	X	X		X	X		4
Nguyen 2012	X	X	X		X		4
Nishtar 2007	X		X	X			3
NSW Health 2002	X	X		X		X	4
O'Loughlin 1999		X	X	X		X	4
Osler 1993	X		X	X			3
Phillips 2014		X		X		X	3
Reger‐Nash 2005	X	X	X	X	X		5
Rissel 2010		X		X	X		3
Sarrafzadegan 2009	X		X	X			3
Simon 2008				X	X	X	3
Solomon 2014	X	X		X	X	X	5
Wendel‐Vos 2009	X	X		X	X		4
Wilson 2014	X	X		X		X	4
Young 1996	X	X		X			3
Zhang 2003		X	X				2
Total	23	26	20	29	18	14
2 components ‐2 studies; 3 components ‐ 10 studies; 4 components ‐ 13 studies; 5 components ‐ 4 studies; 6 components ‐ 4 studies. See Types of interventions for examples of suitable strategies which would be components of an integrated community wide strategy

Table 3. Categories of strategies included in interventions

Table 4. Assessment of intensity of the interventions

Study	High	Medium	Low	Unclear
Brown 2006	X
Brownson 2004		X
Brownson 2005		X
De Cocker 2007		X
Eaton 1999	X
Gao 2013		X
Goodman 1995			X
Gu 2006	X
Guo 2006		X
Jenum 2006		X
Jiang 2008	X
Kamada 2013			X
Kloek 2006			X
Kumpusalo 1996		X
Luepker 1994	X
Lupton 2003	X
Mead 2013		X
Nafziger 2001	X
Nguyen 2012		X
Nishtar 2007			X
NSW Health 2002			X
O'Loughlin 1999		X
Osler 1993			X
Phillips 2014		X
Rissel 2010			X
Reger‐Nash 2005		X
Sarrafzadegan 2009		X
Simon 2008			X
Solomon 2014			X
Wendel‐Vos 2009	X
Wilson 2014	X
Young 1996		X
Zhang 2003	X
Total	10	14	9	0
Intensity was assessed subjectively and independently based upon six characteristics as described in Data collection and analysis

Table 4. Assessment of intensity of the interventions

Table 5. Dichotomous outcomes ‐ physical activity

Study	Overall bias	Measure	Definition	Net % change	Unadjusted RD	Adjusted RD (95% CI)	Unadjusted RR (95% CI)	Adjusted RR (95% CI)	Baseline
Brown 2006	High risk of bias	% physically active	150 minutes of activity in at least 5 separate sessions in the last week	15.40	0.9	7.33 (‐23.48 ‐ 38.13)	1.02	1.18 (0.60 ‐ 2.35)	41.9
Gao 2013	High risk of bias	% Moderate or high physically active	Categories on IPAQ	3.34	7.4	2.50 (1.17 ‐ 3.83)	1.10	1.03 (1.01 ‐ 1.05)	70.5
Jiang 2008	Unclear risk of bias	Regular physical activity	Not provided	18.12	6.38	10.75 (5.23 ‐ 16.27)	1.24	1.20 (1.09 ‐ 1.31)	60.39
Kamada 2013	Low risk of bias	% physically active	Engaging in 150mins/week or more of walking, engaging in daily flexibility or engaging in 2 or more days a week of in muscle strengthening activities (All groups vs. control)	‐0.17	‐1.6	0.00 (0.0‐0.0)	0.973	1.00 (0.99‐1.00)	63.0
	Low risk of bias	% physically active	Engaging in 150mins/week or more of walking, engaging in daily flexibility or engaging in 2 or more days a week of in muscle strengthening activities (Aerobic exercise group vs. control)	‐2.80	0.000	‐2.0	1.00	0.97	66.6
	Low risk of bias	% physically active	Engaging in 150mins/week or more of walking, engaging in daily flexibility or engaging in 2 or more days a week of in muscle strengthening activities (Aerobic exercise and strengthening group vs. control)	0.41	‐0.3	0.30 (‐4.56 ‐ 5.16)	1.00	1.00 (0.94 ‐ 1.08)	64.0
Kloek 2006	Unclear risk of bias	% physically active	At least 30 minutes of moderate‐intensity physical activity on at least 5 days a week	‐7.36	‐1	‐3.97 (5.02 ‐ ‐12.95)	1.04	0.93 (0.79 ‐1.10)	59.0
Lupton 2003	High risk of bias	% physically active	Minimum of four hours of weekly moderate PA during the last year	9.84	8.3	6.87 (‐13.04 ‐ 26.78)	0.98	1.10 (0.84 ‐ 1.43)	72.5
NSW Health 2002	High risk of bias	% physically active	Engaged in at least 150 minutes and five sessions of moderate activity or three sessions of vigorous activity per week	7.14	‐0.2	3.39 (‐0.29 ‐ 7.08)	1.14	1.08 (0.99 ‐ 1.17)	49.2
Phillips 2014	Low risk of bias	% meeting	Physical activity: 5x30 minutes per week	7.89	1.9	5.00 (‐2.879 ‐ 12.879)	1.029	1.079 (0.957 ‐ 1.216)	63.4
Reger‐Nash 2005	High risk of bias	% physically active	Moderate activity at least 30 minutes for at least 5 days per week or vigorous activity at least 20 minutes for at least 3 days per week	0.36	1.2	0.38 (‐0.06 ‐ 0.82)	1.15	1.01 (0.10 ‐ 1.01)	46.9
Rissel 2010	Unclear risk of bias	% physically active	undertaking 150 min/week	‐5.55	‐5.0	‐2.8(‐6.47 ‐ 0.873)	0.907	0.951 ( 0.891 ‐1.015)	44.9
Sarrafzadegan 2009	Unclear risk of bias	% physically active	Individuals with >= 30 minutes/day of moderate or vigorous activity	4.17	2.1	1.89 (‐0.23 ‐ 4.02)	1.07	1.06 (1.00 ‐ 1.14)	47.0
Solomon 2014	Low risk of bias	% physically active	Did sufficient physical activity to meet the current United Kingdom physical activity guidelines (at least 150 minutes of moderate‐intensity activity per week in bouts of 10 minutes or more, or at least 75 minutes of vigorous intensity activity per week	1.03	NA	NA	NA	1.02 (0.88 ‐ 1.17)†	66.9
Wendel‐Vos 2009	High risk of bias	% physically active	150 min/week and at least 5 sessions per week, and physically active at least 30 min/day at least 5 days a week	‐3.50	‐0.7	‐1.60 (‐0.10 ‐ ‐3.10)	0.86	0.97 (0.93 ‐ 1.00)	42.8
RD = Risk difference RR = Relative Risk † Data as presented by the study authors. Odds ratio of adjusted comparison (Intervention minus control in stepped wedge cluster randomised controlled design, p‐value = 0.80, ICC 0.008. Baseline represents baseline for all..

Table 5. Dichotomous outcomes ‐ physical activity

Table 6. Dichotomous outcomes ‐ physical activity during leisure time

Study	Overall bias	Measure	Definition	Net % change	Unadjusted RD	Adjusted RD (95% CI)	Unadjusted RR	Adjusted RR (95% CI)	Baseline
Luepker 1994	Unclear risk of bias	PA during leisure time	Regularly active during leisure time	^a11.26	8.5	5.35 (‐3.32 ‐ 14.02)	1.08	1.11 (0.94 ‐ 1.30)	48.6
Luepker 1994	Unclear risk of bias	PA during leisure time	Regularly active during leisure time	^b9.4	4.3	4.70 (‐1.64 ‐ 11.04)	1.09	1.08 (0.97 ‐ 1.20)	49.4
Kumpusalo 1996	High risk of bias	PA during leisure time	Undertaking physical activity during leisure time > 3 times weekly	‐1.76	0.6	‐0.64 (‐8.24 ‐ 6.96)	1.02	0.98 (0.80 ‐ 1.21)	39.0
Nishtar 2007	Unclear risk of bias	PA during leisure time	Not provided	‐25.58	2.5	0.52 (‐0.04 ‐ 1.08)	2.41	0.88 (0.77 ‐ 1.02)	3.0
^adata from independent surveys ^bdata from cohort surveys RD = Risk difference RR = Relative Risk

Table 6. Dichotomous outcomes ‐ physical activity during leisure time

Table 7. Dichotomous outcomes ‐ physically inactive or sedentary

Study	Overall bias	Measure	Definition	Net % change	Unadjusted RD	Adjusted RD (95% CI)	Unadjusted RR	Adjusted RR (95% CI)	Baseline
Eaton 1999	High risk of bias	Physically inactive (%)	used different criteria for physical inactivity combined questions XS1, XS2 (exercise <1 times per week on average), XS5 & XS6 (=0 days of sweat related physical activity) (not validated measures). Men <=35 yo	1.77	‐20.3	0	1.08	1.03	50.2
			Men >35	0.09	1.9	0	0.98	1.00	53.6
			Women <=35 yo	‐0.18	‐8.2	0	0.99	1.00	56.2
			Women >35 yo	‐0.09	‐6.1	0	0.99	1.00	62.6
Gao 2013	High risk of bias	Low physical activity (%)	IPAQ category	‐8.76	‐7.3	‐2.30 (‐3.39 ‐ ‐1.01))	0.77	0.91 (0.86 ‐0.95)	29.4
Goodman 1995	Unclear risk of bias	Physically inactive (%)	Physically inactive was defined as engaging in no physical activity or exercise during the last month	‐1.82	‐1.7	‐1.02 (‐3.03 ‐ 0.99)	0.97	0.99 (0.96 ‐ 1.01)	44.6
Jenum 2006	High risk of bias	Physically inactive	No heavy physical activity in leisure time or commuting (%)	20.04	12.9	8.13 (5.25 ‐ 10.99)	1.36	1.20 (1.12 ‐1.28)	40.5
Nafziger 2001	Unclear risk of bias	Physically inactive (%)	Described as self‐reported sedentary life‐style unless they were involved in a physical activity strenuous enough to work up a sweat <3 times/week	‐15.85	‐7.8	‐11.43 (‐23.06 ‐ 0.21)	0.89	0.84 (0.71 ‐ 1.00)	72.5
Nguyen 2012	High risk of bias	Physically inactive (%)	Total physical activity less than 3,000 MET‐minutes per week (all)	71.26	4.9	5.63 (2.64 ‐ 8.62)	1.51	1.65 (1.26 ‐ 2.16)	8.0
			Total physical activity less than 3,000 MET‐minutes per week (men)	45.24	5.8	4.82 (0.99 ‐ 8.65)	1.50	1.35 (1.06 ‐ 1.72)	10.0
			Total physical activity less than 3,000 MET‐minutes per week (women)	93.44	4.44	6.43 (1.85 ‐ 11.02)	1.52	1.98 (1.21 ‐ 3.24)	6.8
Osler 1993	High risk of bias	Physically inactive (%)	No details provided	20.51	0.1	2.07 (‐125.30 ‐ 129.45)	1.00	1.16 (extreme 95% CI crossing 1)^a	13.0
This table contains data where % inactive, not undertaking sufficient PA, or classed as achieving a level of PA defined as low or inactive. Eaton 1999 did not provide a sufficient breakdown of the respective sample sizes to calculate 95% CI. Author was contacted for further details, no reply was provided. RD ‐ Risk difference RR ‐ Relative Risk a Due to small numbers, the confidence interval for the adjusted 95% using Wald‐test yields extreme values.

Table 7. Dichotomous outcomes ‐ physically inactive or sedentary

Table 8. Continuous outcomes ‐ leisure time physical activity

Study	Measure	Subgroup	Post mean difference	Adjusted mean difference	Adjusted % change relative to the control mean	Baseline value	Timeline
De Cocker 2007	Leisure time PA (hours/week)	No subgroup‡	0	0.53	25.60	2.33	1 year
Simon 2008	Supervised leisure time physical activity (hours/week)	Measured only in children†	0.9	1.1	43.14	2.5	4 years
Wendel‐Vos 2009	Leisure time PA (hours/week)	Men (NS)	‐0.2	‐0.4	‐2.06	19.8	5 years
Wendel‐Vos 2009	Leisure time PA (hours/week)	Women‡	‐4.4	2.2	14.01	11.7	5 years
† authors reported a statistically significant increase (P < 0.05) in favour of the community ‡authors reported a statistically significant difference ( P < 0.05), however there was no observed increase in PA NS ‐ no statistically significant difference

Table 8. Continuous outcomes ‐ leisure time physical activity

Table 9. Continuous outcomes ‐ walking

Study	Measure	Sub group	Post mean difference	Adjusted mean difference	Adjusted % change relative to the control mean	Baseline value	Timeline
Brownson 2004	7 day total walking (mean min/week) (NS)	N/A	‐5.3	‐1.4	‐1.38	97.2	2 years
Brownson 2004	7 day walking for exercise (mean min/week) (NS)	N/A	‐0.1	‐5.6	‐17.61	37.3	2 years
Brownson 2005	Walking (mean min/week) (NS)	N/A	‐0.8	5.2	4.75	97	1 year
De Cocker 2007	Pedometer‐determined (steps/day)†	N/A	957	1030	10.80	9597	1 year
De Cocker 2007	Walking (min/week)†	N/A	34	47	17.34	288	1 year
Wendel‐Vos 2009	Walking (hours/week)‡	Male (NS)	‐0.4	‐1.1	‐12.09	8.5	5 years
Wendel‐Vos 2009	Walking (hours/week)‡	Women	2.4	2.6	38.24	8.9	5 years
NS ‐ authors reported no statistically significant difference † authors reported a statistically significant increase (P < 0.001) in favour of the intervention community ‡ authors reported as a statistically significant difference ( P < 0.05), however there was no observed increase in PA for the intervention community

Table 9. Continuous outcomes ‐ walking

Table 10. Continuous outcomes ‐ energy expenditure

Study	Measure	Post mean difference	Adjusted mean difference	Adjusted % change relative to the control mean	Baseline value	Timeline
Gao 2013	Physical activity MET ‐ min/week†	462	176	38.09	1204	2 years
Kloek 2006	METs/week (NS)	81	‐241	‐3.54	7253	2 years
Phillips 2014	METS/week score (NS)	4.2	‐113 (95% CI ‐ 847‐ 621)*	n/a	2626 (95% CI 1978 to 3279)	3‐4 years
Sarrafzadegan 2009	Total daily PA (MET‐m/week ± SD) (P<0.01)^b	‐65	‐51	‐8.46	606	3 years
Sarrafzadegan 2009	Leisure time PA (MET‐min/week) (P<0.01)^b	‐2	‐3	‐2.83	85	3 years
Solomon 2014	METs/week (NS)	155	171 (95%CI: ‐16 ‐ 358) p=0.07^a	n/a	2561 (SD=2977)	Stepped wedge design ˜ 12 months follow up
*Adjusted analysis adjusted for age, gender, ethnicity, education, employment, appropriate baseline values a Adjusted analysis adjusted for period, gender, age and area as reported by the authors b When compared to the control, the authors reported a statistically significant decrease in physical activity for the intervention group. † authors reported a statistically significant increase (P < 0.05) in favour of the intervention MET‐ metabolic equivalent of task

Table 10. Continuous outcomes ‐ energy expenditure

Table 11. Continuous outcomes ‐ physical activity

Study	Measure	Estimate of effect for full intervention	Baseline value	Timeline
Wilson 2014	Average daily minutes of moderate to vigorous PA Measured by acceleratory (7‐day estimates) (NS)	0.69 (SE 0.39); 95% CI: ‐0.14 ‐ 1.39)	35.96 (SE 4.17)	Measured 6, 12 & 24 months
Wilson 2014	Average daily minutes of moderate to vigorous PA self‐reported 4 month recall	data unavailable, only more robust measurement of PA reported
SE ‐ standard error NS ‐ authors reported no statistically significant difference Average daily minutes of MVPA is on the square root scale. Analysis as reported by the author using a mixed model ANCOVA implemented to examine difference between communities in accelerometer‐assessed MVPA, random effects for individuals over time.

Table 11. Continuous outcomes ‐ physical activity