If You Build it, Will They Come?... and Live Longer - A Natural Experiment of an Urban Trail Network Expansion and Cardiovascular Disease

Two different time series methods will be used to estimate the effect of an urban trail expansion (i.e. "intervention") that occured in WInnipeg, Manitoba Canada, between 2010 and 2012. The study is designed to determine if a reduction in Major Adverse Cardiovascular Events (MACE) was observed in neighbourhoods that received the intervention relative to trends among the control neighbourhoods that did not receive the intervention. First, a multi-group segmented regression of interrupted time series data will be used to assess the effect of the intervention on CVD incidence, both immediately (change in level) and over time (change in trend) by creating indicator variables . The level will be the base rate of CVD-related end-points at the beginning of the pre-intervention period (2000) and the value immediately following each change point at which successive segments join until 2010. The trend is the rate of change in MACE end-points (in other words, the slope) during a segment. Autoregressive errors will be modeled to account for correlated outcomes. Second, an autoregressive integrated moving average (ARIMA) model will be fitted for the CVD incidence time series by using the standard approach to identification, estimation, and checking. A trend and periodic seasonal terms will be applied to the entire study period (November 2000 to October 2019). A separate ARIMA model will also be built for the pre-intervention period to forecast CVD evolution of the treated neighbourhoods. The number of CVD end-points prevented by the intervention will be estimated by calculating the difference between the predicted number and the observed number of cases. Should there by difficulty fitting an ARIMA model to a relatively small dataset, exponential smoothing models or the Holt Winters Algorithm will be used. Although they require larger sample sizes, they are ideal for this project as (1) they permit a variety of different types of intervention effect to be modeled explicitly, and (2) they are well suited to forecasting future trends.