INTERACT: A comprehensive approach to assess urban form interventions through natural experiments

The INTERACT project will examine the potential health impacts of interventions in four Canadian cities.

Victoria (All Ages and Abilities (AAA) Cycling Network), a commitment of $14.3 M to implement an All Ages and Abilities (AAA) Cycling Network, starting with the implementation of a 5.4 km grid in the downtown core. Protected bike lanes, off-street paths, and low-volume shared-use routes will connect every neighbourhood and regional trails [26].

Vancouver (Arbutus Greenway), a 9 km rail corridor the City of Vancouver purchased in 2016 for $55 M, which is being developed into a continuous walking and cycling corridor through prime real estate connecting South Vancouver to False Creek [27].

Saskatoon (Bus Rapid Transit (BRT)), a $66 M investment in a 22 km Saskatoon BRT system along three major roadways. BRT is an enhanced bus system, operating on bus lanes or other transit ways to combine the flexibility of buses with the efficiency of rail [28].

Montreal (Sustainable Development Plan 2016–2020), reflects a total investment over $100 M, is a comprehensive sustainability plan which includes urban form changes initiated by the city and approximately 230 partnering organizations that will be implemented. We will specifically evaluate how traffic calming measures, changes in transportation infrastructure, place making, and greening program influence our outcomes of interest [29].

We require at a minimum that participants complete an online survey, and in addition they may choose to install an app on their mobile device, wear a mobile sensing device, participate in qualitative interviews, or any combination of the above. For participants who may be interested in the study but do not have internet access and a computer we will provide the opportunity to complete the surveys in person or over the phone with a research assistant. Figure 4 shows the flow diagram for potential participants in the INTERACT studies. Ethics approval has been received from the ethics boards of Simon Fraser University, the University of Saskatchewan, the Centre de Recherche du Centre hospitalier de l’Université de Montréal, and Memorial University of Newfoundland.

We estimate that INTERACT will collect approximately 100 terabyte (TB) of raw data during the study period. The volume and variety of data make data security, storage, and processing a crucial consideration for the INTERACT study. We plan on partnering with both the Centre for Health Information and Analytics at Memorial University and Compute Canada for data infrastructure requirements. There are considerable training requirements for public health researchers and students in both data handling and machine learning methods [25]. We will develop training modules related to working with large data infrastructure, programming in Python and R, and methods for analyzing large spatio-temporal datasets.

The INTERACT program is guided by a framework for sustained impact [76], which highlights the importance of early and sustained engagement with non-government organizations and policy makers nationally and locally. The meaningful engagement of intersectoral stakeholders throughout the research process will be guided by a governance model that prioritizes consultation and collective decision-making, and the equitable participation of knowledge users. For example, city decision-makers are integrated within intervention teams, positioning them to provide input on local research questions and methods, and to facilitate evidence-informed decision-making. We have a diversity of knowledge translation end products planned, including both applied and scholarly outputs relevant to municipalities, regional governments, health authorities, non-governmental organizations, academics, and citizens. One example is that we have adopted a process of rapid reviews that can inform research or policy at the local level. We also aim to enhance training and capacity for urban form research through open and wide dissemination of INTERACT’s evidence, tools, and methods to both practitioner and research communities.

There are ethical considerations for INTERACT, specifically related to the high precision location data collected from participants [77]. In natural experiment studies using mobile sensing it is important to recognize risks that are part of the intervention, and risks that arise through unintended circumstances during the study. Risks related to the intervention are unlikely in a natural experiment study because researchers do not control the intervention [77]. There are risks related to using mobile sensing data for health research including consent, privacy and confidentiality, mitigating risk, and vulnerable populations. We will mitigate potential risks by using plain language consent forms, having clear Frequently Asked Questions (FAQ) support for participants, updating participants about the data they are contributing, using secure protocols for data storage and access, and training research trainees and staff on data security and access [78, 79].

Making causal claims about intervention effects is the primary objective of epidemiologic studies [80]. Regression discontinuity, difference in differences, and interrupted-times series are methods that can permit causal claims to be made about intervention effects in observational research, provided that the assumptions of the methods are met. A challenge of evaluating the impact of urban form interventions is that interventions and individual level exposures are changing in both time and space [81]. In contrast to strict interventions, which can be implemented from one day to the next, urban form interventions are implemented over longer time periods and often in phases. The nature of urban form interventions will require us to test multiple methods for defining exposure both temporally and spatially, and cautiously interpret our results. The on-going nature of the intervention implementation and confounding by other urban form changes will be a challenge for causal attribution [82].

INTERACT faces limitations related to sampling bias, control groups, intervention specificity, and context. Sampling bias may take two forms. First, vulnerable populations are less likely to participate in research in general and may be even less likely to participate in research involving mobile sensing, either because they may not have these devices or because the detailed location data collection may be considered additionally intrusive to these populations [78]. Second, we have not a priori identified control groups, either in the form of comparable cities or comparable unexposed groups, for our interventions. However, for some research questions, spatial exposure can serve to identify levels of exposure within the sample. Third, INTERACT will evaluate four different urban form interventions in four different cities. Both the conceptual and statistical identification of the hypothesized and actual intervention effects and which aspects of the intervention we are evaluating is challenging. There are several potentially relevant contextual features to consider including: inequities, public consultations, media campaigns, and political decision making. With only four cities included, we are using an ad hoc approach to consider how these and other contextual factors should be considered in our analysis.

INTERACT will harness big data to deliver timely public health evidence on the influence of real-world urban form interventions on physical activity, well-being, and social participation, as well as social inequities in these outcomes. We aim to understand context, measure change, analyze impact, and mobilize knowledge about how the urban form can influence health.