Examining the association between cycling infrastructure exposure and physical activity: a natural experiment study in Victoria, Canada

Background: Most Canadians are not meeting physical activity guidelines. Physical inactivity is an epidemic related to alarming rates of preventable disease and economic burden. Health benefits ensue from modest increases in physical activity (i.e., 5 minutes daily). Despite the health benefits, there is limited evidence about interventions to increase physical activity at the population level. Built environment interventions, like bicycle infrastructure, are one understudied intervention with the potential to increase total population physical activity. The purpose of this study was to examine the physical activity impacts of bicycle infrastructure. Intervention: The intervention studied is the All Ages and Abilities (AAA) Cycling Network in Victoria, BC. This network of bicycle infrastructure is a $7.75M commitment. In 2018, a 5.4km grid of protected cycle tracks was built downtown. Research Design: Adults were recruited in Victoria who ride bicycles at least monthly. Baseline activity data were recorded in 2017 (n = 281) using surveys, Global Positioning System (GPS), and accelerometer data, with a follow-up data collection in 2019 (n = 315). The primary outcome was moderate to vigorous physical activity (MVPA), assessed via accelerometer data. Hypothesis: Exposure to new active transportation infrastructure will be associated with increased MVPA over time. Analysis: I calculated exposure measures to the AAA Cycling Network using GPS and Geographic Information System using intersections. I used regression models to examine the associations between exposure to new infrastructure and total location-based physical activity levels, controlling for confounders. Results: In the multilevel models analyzing the interaction between exposure and wave with covariates, wave two, compared to wave one, saw a non-statistically significant increase in MVPA of 0.93 minutes per week [CI = -5.28, 7.14]. Comparing the models with and without covariates suggests that the wave one and wave two comparisons were highly confounded by individual and weather covariates