The Canadian Urban Environmental Health Research Consortium – a protocol for building a national environmental exposure data platform for integrated analyses of urban form and health

CANUE’s vision is to increase scientific understanding of the interactions among the physical features of the urban environment and health. This understanding will lead to cost-effective actions that promote healthy childhood development and aging, reduce the burden of chronic disease, and minimize the impact of changing environments. To achieve this vision, CANUE is establishing and implementing a protocol for compiling environmental measures or metrics that: quantify exposure, behaviour patterns or effect modifiers; vary spatially on the urban and suburban scale; can be obtained for multiple urbanized regions in Canada and; could be modified through changes in policy or individual behaviour to benefit health. While urban areas are the focus, exposures across rural Canada are also being compiled. CANUE’s main data-related products are: (1) multiple spatial maps or surfaces (Fig. 1), one for each environmental exposure metric, that can linked to individuals included in health studies based upon the locations where they spend time (e.g., home); (2) computational tools to derive new exposure metrics; (3) documentation of the methods and data, how they were derived, by whom and key information relevant to data usage (e.g., potential limitations) and; (4) facilitation of exposure data extraction for individual-level data merging with major Canadian health databases (Table 1), including consideration of residential mobility.

CANUE currently focuses on collating and generating exposure metrics in six domains: Air Pollution, Noise, Greenness, Weather and Climate, Transportation, and Neighbourhood Factors, which include land-use, urban design and social determinants. These factors are grouped together, recognizing that much of our health and well-being begins at the neighbourhood level and there has been a great deal of theoretical guidance as to which factors at this scale are paramount, influencing key behaviours such physical activity and diet [25, 26]. Also, it is at this scale that patterns in socioeconomic factors manifest, creating a backdrop of individual susceptibility that must be considered in the context of public health. Active within CANUE are domain-specific working groups assessing the state of knowledge and research nationally and internationally, identifying critical gaps and conducting strategic research to improve the available exposure metrics. Fig. 2 places the six domains in the context of key external forces influencing urban form: population growth, economic growth, and weather/climate which includes factors such as extreme heat and cold events and longer-term climate change. In general, the main public responses to these forces are land-use planning and transportation infrastructure decisions; in turn, this leads to individual options around housing, employment and education locations. Choices made based upon these options or constraints subsequently impact an individual’s access to or interaction with urban features of health relevance and dictates individual behaviour such as time spent commuting and working or time available for leisure and family. All ultimately impact the magnitude of a range of harmful or beneficial exposures and thus individual and public health.

The exposure data or metrics being compiled in CANUE are georeferenced at the six-digit postal code level (or other geographic level as appropriate) facilitating linkage with health research cohorts and administrative health databases. Changes in the geographic distribution of exposure over time are important to consider given the potential time windows over which environmental factors can contribute to adverse health outcomes and chronic disease development. The temporal resolution required and the number of years back in time for which exposures can be estimated varies across the domains based upon the rate of change over time and available data sources. Accurately accounting for short and long term exposure time windows represents a considerable challenge and will be discussed below.

Within each domain, existing data are being centralized to improve accessibility for researchers and subsequent integration with Canadian health data platforms. Work with these initial datasets (Table 2) is also facilitating development of CANUE’s infrastructure for data transfers, storage, manipulation into analysis-ready formats and documentation, including terms of use that assure requirements of the data originators are respected. This initial phase is helping to identify challenges related to harmonization of environmental data within and between domains and with similar efforts internationally. Several of these existing exposure datasets have been previously used in health research, nationally or regionally within Canada and are thus in analysis-ready format; while others require further processing and quality assessment before linkage with health data.

Three air pollutants have been used most often in recent epidemiological research in Canada; fine particulate matter (PM_2.5), nitrogen dioxide (NO₂) and ozone (O₃). National coverage for PM_2.5 is derived from the recently developed 1 × 1 km satellite-derived PM_2.5 surface [27]. Exposures for NO₂ are estimated empirically from a national Land-Use Regression (LUR) model [28] and finer scale spatial patterns in NO₂ are available from LUR models for 10 cities in Canada [29]. Exposures to O₃ have been derived from a combination of observations and output from the chemical transport model developed by Environment and Climate Change Canada for air quality prediction and used in recent epidemiological studies [30, 31]. A national surface for sulphur dioxide (SO₂) is also available based upon recent progress in satellite detection and extrapolation to surface concentrations [32]. Temporal coverage of CANUE national and urban-level air pollutant exposure surfaces will initially extend from 2000 to the present.

The LUR method has also been used to model spatial surfaces of urban environmental noise exposure in two Canadian cities, Montreal [33] and Toronto [34], with field monitoring also conducted in other cities (e.g., Vancouver, Ottawa, and Halifax [35]). Vancouver noise maps for 2003 have been generated using the deterministic propagation model CadnaA [36]. A similar model is currently being run for Montreal for 2008. CANUE is documenting these noise exposure surfaces to make them more-widely available for epidemiological research.

The Normalized Difference Vegetation Index (NDVI), which is derived from satellite measurements of near infrared and visible red radiation reflected by vegetation, is readily available as an indicator of greenness and the exposures this metric may represent. This includes already developed annual and peak growing season NDVI products using Landsat 5 and Landsat 8 [37, 38], the Moderate Resolution Imaging Spectroradiometer (MODIS) [39] or the Advanced Very High Resolution Radiometer (AVHRR) data which provides annual coverage and potentially seasonal variations as far back as 1981 [40, 41].

To facilitate research on how extreme weather and climate relate to the incidence of chronic disease, CANUE is including climate data. At present, the spatial resolution available across Canada is limited and originates from interpolation of the available, largely routine observations and/or from re-analysis products combining models and observations. As such, an observation-based dataset of daily maximum and minimum temperatures and precipitation produced by the Canadian Forest Service and Environment and Climate Change Canada is available at 10 × 10 km [42] and raw data can also be accessed by station [43] to derive proximity-based metrics of weather and climate (i.e., summary statistics based on nearest stations). The Climate Forecast System Reanalysis [44] or the Japanese 55-year Reanalysis [45] are comparable, while higher resolution observed gridded data, such as the ~800 m data covering British Columbia through the Parameter-elevation Regressions on Independent Slopes Model (PRISM) [46], are expected to become available nationally in the future.

Geographic Information Systems (GIS) provide the tools for calculation of a variety of exposure metrics at a fine scale across urban areas and within neighbourhoods. Walkability, for which multiple measures have been developed [47, 48], will be included early in the CANUE data holdings. The Canadian Census data includes socioeconomic data for the country from which several indices can be computed and mapped. The Canadian Marginalization Index (CanMarg) [49] and the Pampalon Index [50] have been or are being determined for multiple cycles of the Canadian census from the 1980s to the present. Light at night, which is derived from satellite observations with 1 km resolution, is also available and is listed as part of the neighbourhood factors domain [51].

Limitations associated with the exposure measures currently available for each domain are being addressed by the CANUE Working Groups. This involves the initiation of research projects and/or targeted workshops to guide future projects. Priorities for this work were developed at a national workshop held in December 2016 (www.​canue.​ca/​workshop). Clearly, CANUE will not be able to address all limitations within five years. In Table 3, selected key exposure metric advances planned for this time period (i.e., through ~2021) are summarized and through new partnerships CANUE will be able to further expand the amount and type of new exposure data available for health research.

Key challenges for large environmental health studies, particularly those aiming to implement an exposome-based approach, continue to be enrichment of cohorts with individual-level exposures, harmonization across cohorts and, and ultimately identification of modifiable risk factors leading to interventions that have benefits on population health. To help meet these challenges Stingone et al. [83] suggested that exposome studies would be well-served by centralized support and coordination to ensure that potential exposure assessment strategies are rigorously evaluated. CANUE represents an attempt to meet these challenges with respect to exogenous factors and, while CANUE is the largest coordinated effort in Canada around environmental exposure data, many challenges remain.

There is a long-standing need to better understand temporal change in spatial exposure patterns going back decades and how this contributes to exposure misclassification and subsequent epidemiological results [84, 85]. Detailed characterization of spatial patterns with high resolution that are indicative of chronic exposure is typically only accomplished for ‘snapshots’ in time because of the effort and expense required. It is therefore necessary to estimate temporal changes in these spatial exposure patterns by extrapolation of the spatial detail. This could include estimates covering longer time periods (i.e., decades) or particular months to years before or after the time of the ‘snapshot’. For air pollution a variety of extrapolation approaches have been used [84‐87]; however, in order to have reasonable confidence in the estimates it is necessary to have monitoring site data with temporal coverage for the time periods and pollutants of interest and ideally from multiple locations depending upon the size of the spatial domain modelled. This is problematic because long-term exposures over relatively large geographic areas require estimates much further back in time pre-dating the monitoring of some pollutants (e.g., PM_2.5). In these cases, there is likely much greater uncertainty in the exposure estimates [85], but they are difficult to quantify given lack of evaluation data.

The need for temporal extrapolation and uncertainty arising from lack of historical exposures are limitations impacting most of the exposure domains of interest to CANUE. The noise pollution maps are available for a limited number of cities and specific snapshots in time. New noise maps to be developed through CANUE will also face this limitation and their applicability to other time periods or longer time windows relies on the assumption of temporal stability. Given that a key source of noise is traffic and other transportation activities (e.g., airports) and the infrastructure for these is stable over relatively long periods, extrapolation is reasonable. However, road, air and train movements have changed overtime as well as emissions; the locations of many other noise sources can change more-rapidly; and even changes such as construction of noise barriers will alter exposure patterns. Further, fitting noise models to similar geospatial predictors as air pollution contributes to collinearity hindering attempts to isolate effects due to these two exposures [88].

Coordination through CANUE offers promise that some progress on these and other challenges can be achieved. Google’s Earth Engine, for example, is hypothesized to facilitate the analysis of big geospatial data with a temporal coverage that will be informative of changes in urban environment exposure metrics going back into the 1980s. CANUE provides the critical mass to explore this idea. Given high resolution surfaces of noise and air pollution, other health-relevant neighbourhood features and maps of local climate zones that indicate potential for heat islands, it may be possible, using local land use variables as model inputs, to develop algorithms that can relate land use classifications derived from the 30 m Landsat images. These algorithms, if robust and mechanistically-based, could then enable reliable estimation of a variety of urban form exposure variables back to 1984.

Residential mobility is also an important cause of misclassification when exposure assessment relies on geographic location. Often exposure is based upon a single home address, such as might be acquired at time of study recruitment or baseline. The potential for differential exposure misclassification has been demonstrated in birth cohorts [89], and can be expected to increase the longer the follow-up period or the longer the exposure time window of interest. Crouse et al. [30] reported that nearly 50% of the Canadian population moved at least once in the 5 year period from 2001 to 2006. They accounted for residential mobility during the 16 year of follow-up of Canadian Census Health and Environment Cohort and found that this led to larger hazard ratios compared to those determined using exposures assigned using a single baseline address. This attenuation in hazard ratio was greatest for NO₂, less for PM_2.5 and negligible for O₃.

Residential history on study individuals, if available, can be used to determine time-weighted exposures, assuming exposure data are available for the different addresses reported. Ideally, such information is obtained in prospective cohorts through questionnaires. In practice this is not always the case and/or the data are incomplete. Administrative data housed at the federal and provincial level represents a different option, taking the burden away from the subjects, while standardizing the approach. CANUE is working with Statistics Canada through the Social Data Linkage Environment [90] to obtain annual residential history data for individuals in some cohorts following the method used by Crouse et al. [30]. Provincial health care records also retain addresses and these data are being assessed for reconstruction of residential history.

Daily mobility and time spent indoors poses another challenge for exposure assessment. Accounting for time at work or school and proximate exposures is feasible given sufficient information and resources. While where a person lives plays a major role in their relationship with all of the urban form features related to CANUE’s exposure domains, all locations where significant time is spent, including in transit (i.e., commuting), are potentially important (Fig. 2). CANUE is aiming to provide exposure metrics for many potential locations allowing for additional time-weighting of outdoor exposures. However, reliable time activity behaviour at the individual level represents a key limitation.