This systematic review protocol was developed according to the preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) checklist.

Eligibility criteria 1.

Published and peer-reviewed empirical studies in the English language.

2.

Studies must have been conducted in high-income countries (according to The World Bank [30] classification).

3.

Empirical results based on longitudinal, repeated cross-sectional, natural-experiment and quasi-experiment study designs (cross-sectional designs are excluded).

4.

Must have studied and estimated the BE effects on transportation or recreational walking.

5.

Must have studied the effects either on working-aged adults (aged between 18 and 64 years) or/and older adults separately (older than 64 years). Children or adolescents are excluded.

6.

BE attributes must be objectively measured.

7.

Studies must have been conducted in urban areas (functional urban areas will be identified using the European Commission [31] classification).

8.

Must have studied the effects on the general population and not on specific population subgroups (other than age and/or sex) such as people with specific medical conditions.

Information sources

We will search articles on the BE and walking behaviour in health, behavioural sciences, sport, environmental social sciences, transportation, urban design, physical activity, and multi-disciplinary databases, based on guidance from a team of researchers and librarians to sufficiently cover the topic. The selected databases are: Scopus, PubMed, SPORTDiscus with Full Text (EBSCO), Business Source Complete (EBSCO), Art and Architecture Archive (Proquest), Avery Index to Architectural Periodicals (ProQuest) and Art, Design & Architecture Collection (ProQuest).

Search strategy

The search will be conducted using the search terms listed in Table 1 in the Appendix, which is divided into four sections for BE, walking, purpose of activity (transportation or recreational) and study design. The search terms for each section were identified by referring to those used in other relevant reviews [20, 21, 32, 33]. The terms listed in each of these four sections will be searched in titles and abstracts of studies using OR Boolean operator, and then the results of these four searches will be combined using AND Boolean operator. An example of the search queries used for the Scopus database registry is given in Table 2 in the Appendix. Finally, systematic reviews are excluded, and the studies will be filtered to ensure they are peer-reviewed and published in the English language. The search was carried out for titles and abstracts of articles published until 30th June 2022 and the resulted number of studies is given in Table 3 in the Appendix.

Study records

Once the search is completed duplicate studies will be removed using the EndNote reference manager, and titles and abstracts screened based on the inclusion criteria following the method described by Bramer et al. [34]. Lastly, full texts will be reviewed. Two reviewers (TB and CH) will independently screen all studies identified in the search and a third reviewer will decide on discrepancies. Once the studies are selected, the reviewers will be given a form (Table 4 in the Appendix) to extract the required data. Reviewers will independently record the data from all selected studies and will confirm discrepancies by discussion until consensus is reached.

Data items

We will extract the type of BE attributes or interventions as exposures, exposure measurement details, outcomes (walking for transportation and/or recreation), measurement details of the outcome, covariates used, statistical approach, findings (estimates of the associations with confidence intervals, significance status and sample sizes), study characteristics (authors, year, study design, follow-up period, location of the study etc.) and participants’ characteristics (age, sex, etc.).

Outcomes and prioritization

The primary outcomes of this systematic review are walking for recreation and walking for transportation.

Risk of bias in individual studies (methodological quality assessment)

The quality of the included articles will be assessed using 14 criteria mirroring those used in a previous relatable systematic review [32], these criteria were developed based on those used by Cerin et al. [20]. Items and the score allocations of the quality criteria tool are summarized in Table 5 in the Appendix. As described by Chandrabose et al., a score of 1 or 0 will be assigned for each assessment item depending on whether each item meets the quality criteria or not respectively. A score of 0.5 will be assigned if an item is at an acceptable level of its criteria. A score of 0 or 1/3 will be assigned for items 6, 7 and 11 to avoid overstating the importance of statistical aspects of the articles. An additional score will be given for sample size for each study as described by Chandrabose et al., [32] and Cerin et al. [20] in their reviews. The scores for sample sizes are as follows: 0.25 for sample size ≤ 100; 0.5 for sample size 101–300; 1 for sample size 301–500; 1.25 for sample size 501–1000; 1.5 for sample size 1001–2500; and 1.75 for sample size ≥ 2500. The sum of the quality assessment score and sample size score will be assigned to each included study to assess the strength of evidence towards the research questions. The total quality scores will be used as weights when synthesising the data as described in the Data synthesis section.

Data synthesis

We will summarize the reported evidence to answer our research questions. All reported unique associations between BE attributes and walking will be summarized as number of positive (i.e., BE attributes that support walking, for example, high walkability that encourages walking) and negative (i.e., BE attributes that discourage walking) associations found into Table 6 in the Appendix. In the case of finding non-linear relationships, we will report these findings in the tables and discuss them in detail in the discussion.

The effects of the BE on transportation and recreational walking will be summarized separately for males and females aged 18–64 years, as well as for older adults aged 65 years and older. For example, if an included study looked at the effects only for males aged 18–45 years, this evidence will be used to synthesise evidence for younger male adults. Evidence from studies that have not been separated by sex will be considered to assess the effects for younger and older adults, to compare the effects between younger and older adults irrespective of sex. As a sensitivity analysis, we will exclude evidence from low-quality studies.

Meta-bias(es)

When a protocol is available for a study, we will compare whether all the planned outcomes in the protocol are reported in the published study to identify the existence of selective outcome reporting bias. This bias occurs when study outcomes are reported according to the researchers’ choice of significance, magnitude or direction of the outcomes [35], instead of reporting all intended study outcomes.

Confidence in cumulative evidence

Quality of the evidence (also known as certainty of the evidence) for all outcomes will be graded using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) tool [36]. The quality of the evidence will be graded into four levels: very low (the true effect is probably noticeably different to the estimated effect), low (the true effect might be noticeably different to the estimated effect), moderate (the true effect is probably close to the estimated effect) and high (the true effect is very likely to be the estimated effect). As described by GRADE, the rating for the quality of outcomes for randomized controlled trials is set as high-quality, whilst observational studies are set as low-quality. Then the quality of outcomes is rated down for five categories; (i) risk of bias – limitations in the study design, (ii) inconsistency of results – heterogeneity or variability in results (treatment effect) across studies, (iii) indirectness of evidence – not directly comparing the intervention of interest for the population of interest, (iv) imprecision – having wide 95% confidence interval (CI) around the estimate of the effect when relatively small sample size is used, (v) publication bias – bias due to selective publication of studies. Likewise, the quality of outcomes is rated up for three categories; (i) large magnitude of effect, (ii) exposure-response gradient - increasing level of exposure is associated with either an increasing or a decreasing of outcome, (iii) residual confounding – adjusting for prognostic factors that may relate to the outcome of interest.

The causal effect of an intervention can only be inferred through a randomised allocation of exposure [37], therefore randomised studies are rated as high-quality studies. However, it is likely that a greater number of observational BE-behaviour studies will be included in our review due to the ethical and implementation challenges of randomized controlled trials for evaluating the behavioural impact of BE interventions. Natural experiments, which are observational studies may support stronger causal inferences. They are increasingly common in public health and their strength of evidence is higher than in other observational studies [37]. Also, a conceptual article from the GRADE Public Health Group identifies this challenge and suggests that natural experiments (i.e., interrupted time series or regression discontinuity studies) should be rated as moderate quality [38]. Therefore, the quality of outcomes from natural experiment studies will be rated as moderate quality.