Participation in neighborhood walking groups was associated with a small-but-sustained increase in minutes engaged in brisk walking, and in level of walking efficacy over the intervention period. Contrary to our hypothesized relation between the intervention and potential mediators, the intervention was not associated with a positive change in social cohesion or in neighborhood perceptions. In fact, the intervention was associated with a significant increase in perceived neighborhood problems. In single- and multiple-mediator models, increased walking efficacy appeared to mediate the intervention effects on brisk walking at 6 months. The single-mediator results provide some evidence that perceived neighborhood problems at 6 months (but not 3 months) suppress the intervention effects on brisk walking at 6 months. The absence of a finding for perceived neighborhood problems in the multiple mediator models suggests an overlap in mediated effects of walking-efficacy and perceived neighborhood problems.
Previous analyses examining the direct effect of physical-activity interventions among adults reported mixed results . In the context of this lay-led walking intervention for sedentary older adults, we hypothesized that the intervention would increase walking efficacy. Our findings, consistent with many other studies, found significant correlations between efficacy and physical-activity behavior in older adults [40–44]. This finding suggests that increasing walking efficacy is an important component of any successful future replication of the SHAPE intervention in other populations. It should be noted that in the current study, walking efficacy decreased slightly over the intervention period in the intervention group, although to a lesser degree than it decreased in the control group. The decline in the intervention group may reflect the influence of regression to the mean given that walking efficacy was fairly high (mean 8.0, SD 1.4) at baseline. After controlling for baseline level, walking efficacy was higher at 6-months in the intervention group compared to the control group.
Contrary to our hypotheses, perceptions of neighborhood problems at 6-months (but not 3-months) significantly increased in the intervention group from baseline to follow-up, relative to the control group. Although we are not aware of any previous studies assessing perceptions of neighborhood problems as a potential mediator of an activity-intervention effect, many studies have reported a significant inverse correlation between perception of neighborhood problems and physical activity or physical function in older adults [45–47]. In contrast, King and colleagues  reported that greater levels of neighborhood walking were correlated with higher reports of neighborhood problems, and hypothesized that regular walkers are more familiar with problems in the neighborhood and thus more likely to report these problems. Thus, our finding of increased reports of neighborhood problems within the intervention group at 6-months is consistent with King's results. The lack of a result at 3-months may suggest that it takes longer than 3-months of walking in the neighborhood to become familiar enough to register significant neighborhood problems.
Social cohesion can be defined as a "collective dimension of society external to the individual ," operationalized as strong social ties, mutual trust, and reciprocity. Social cohesion is associated with higher levels of self-rated health and lower morbidity and mortality [50–52]. social-ecological theory suggests that lay-led neighborhood walking groups may increase social cohesion by creating opportunities for neighbors to meet, to interact with non-walking group neighbors who are also outdoors, and to learn about neighborhood resources and facilities [53, 54]. Level of social cohesion is independently associated with health outcomes among older adults . However, based on the findings from this study, changes in social cohesion were not causally related to higher amounts of brisk walking.
The effects of the walking intervention were not significantly moderated by neighborhood walkability. Previous studies in older adults have reported positive associations between neighborhood built-environment characteristics and physical activity [25, 26, 56–58]. Our data suggest that neighborhood walkability may not be an important concern when targeting neighborhood-based walking interventions. Other studies should further assess whether this pattern holds true.
Our study is the first to evaluate change in social-ecological process variables within a multilevel RCT, and to apply a formal statistical test of mediation. This study lays out a method for future researchers seeking to identify individual mediators and moderators, and to quantify their influence on intervention effects. One reason our study was able to perform these analyses is that when the SHAPE trial was implemented, it was done so with the idea that social-ecological theory could be used subsequently to perform just such analyses. This is important, both from the point of view of initial intervention design, and because without an understanding of which elements of an intervention are producing the effect, interventions will not be able to be successfully reproduced in other communities and populations. Thus, analyses such as we present here may, when mediators and moderators are identified that do explain intervention effects, can afford enormous cost-savings for interventionists attempting to adapt existing intervention frameworks to a variety of communities.
Although SHAPE was designed to support analyses such as the ones described here, this area of research is in its infancy. Therefore, it should not be surprising or discouraging that initial attempts to identify mediators and moderators may not initially hit upon all the mediators and moderators contributing the most to the intervention effect. It is only through systematic analyses of individual mediators and moderators that we will be able to, eventually, identify the ones making the most significant contributions to intervention effects, and use this knowledge to refine future interventions built upon the platforms of existing ones.
We of course cannot entirely dismiss the idea that our analyses did not show that all the mediators and moderator in this study had no significant influence on the effect due to the study's lack of sufficient statistical power or model misspecification, or due to confounders such as reciprocal effects . Reciprocal effects are a potential explanation because of the correlative nature of change in brisk walking and mediators in this study. We attempted to address the correlative nature of the data by also evaluating mediators after three months in relation to six-month change in brisk walking. However, future studies should consider whether initial changes in mediators resulted in maintenance of increased brisk walking beyond the intervention.
There is a possibility that the observed differences in baseline characteristics were not due to random chance, in which case, adjusting for baseline values could introduce regression to the mean biases [31, 35]. To investigate possible the influence of baseline differences, we conducted a secondary analysis using observed changes in walking unadjusted for baseline values as the outcome. In this secondary analysis, the action theory tests indicate that there are intervention effects on change in social cohesion and no intervention effects on walking efficacy; however, the conclusions about mediating effects of social cohesion and perceived neighborhood problems remained unchanged. Of note, when using this analytical approach, walking efficacy no longer significantly mediated the intervention effect. This change should be noted when interpreting the finding regarding walking efficacy, as should the importance of obtaining balanced groups at baseline when designing future RCTs.
The analyses were performed using data gathered from RCT participants who completed the study, and therefore the results are generalizable to the subjects who completed the study. As previously reported, attrition was higher in the intervention group; only 62% of the intervention group (n = 159) completed the final 6-month assessment, compared to 95% of the control group (n = 265) . The attrition in the intervention group suggests that additional strategies geared toward retaining participants would be useful in future neighborhood-based lay-led walking interventions in older adults. Since most of the drop-out occurred by 3 months, use of traditional techniques such as imputation is limited because the only data available on the subjects are baseline data. We note that those with complete follow-up data are similar to the group without complete follow-up data. The two groups did not differ in terms of gender (p = 0.55), education (p = 0.61), self-rated health (p = 0.16), baseline walking (p = 0.66) or social cohesion (p = 0.66). In addition, if we assume that those who dropped out had no change in walking behavior and reanalyze the data, the results are consistent with our primary analysis.
Finally, measurement of walking in this study was self-reported, and thus subject to misclassification. The use of instruments to objectively measure walking, such as pedometers, is desirable for future research seeking to expand on these results .