An individual's brain development and aging – and brain health - are influenced by context. Outside the immediate family, many impactful contextual factors influencing our everyday lives act where we live and work. Area-level characteristics of physical, built and social environments shape our brains from conception onwards. But not all areas are the same: inequalities exist in multiple inter-twinned dimensions.

Social, economic and political conditions produce health inequalities within and across countries (Stuart and Soulsby, 2011; Scambler, 2012; Metzl and Hansen, 2018). For example, in high-income countries, individuals are more likely to experience poor mental health if growing up in households with low income (Bjorkenstam et al., 2017) or affluence (Rajmil et al., 2014; Elgar et al., 2015), living in areas with high deprivation (Kivimaki et al., 2020) or experiencing inequalities in income distribution (Mangalore et al., 2007). Certain communities are disadvantaged more than others (Waldron, 2018). This is especially true for Indigenous (Ogilvie et al., 2021) and racialized (Castro-Ramirez et al., 2021) communities, with a higher risk of mental-health problems and – at the same time – lower likelihood of receiving evidence-based treatment (Castro-Ramirez et al., 2021). At the area level, our physical, built and social environments combine to create ecosystems in which we live and work. Together, these ecosystems – and the structures and systems that produce them – contribute to what has been termed “social and structural determinants of health” (Vandenbroucke, 1990; Diderichsen et al., 2001).

As we have described elsewhere (Paus, 2016), there are countless permutations of physical, built and social environments that surround us in space and time. We both “receive” and “create” our environments (Kendler et al., 2003), thus co-determining what air we breathe, how many steps we take, how hot or cold we are, what and who we see, hear and interact with during our commutes. Together with our genes, these “external exposures” contribute to “internal” environments that exist in our body: on body surfaces (e.g., microbes on our skin and in the gut), in the lungs (e.g., particulate matter), circulating blood (e.g., toxins, micronutrients, inflammatory molecules) and the brain (e.g., stress- and reward-related neurotransmitters, cumulative engagement of specific neural circuits).

Assessing the “external” environment – rather than its biological markers in biospecimen – is challenging. Asking a series of questions using a standard survey is a common way of collecting information about the individual's physical, built and social environment. The PhenX Toolkit, for example, contains standardized protocols (including surveys) one can use to collect information about social determinants of health (www.phenxtoolkit.org/index.php). Although valuable, there are two main disadvantages of a survey-based approach: (1) participant's time (many hours required to cover multiple domains); and (2) self-reported nature of the collected information and, therefore, possible reporting biases. Furthermore, with the exception of longitudinal (birth) cohorts, questionnaire-based approaches provide only a snapshot of environments encountered by the individual at one (or a few) timepoints.

Here we describe an alternative approach, namely the use of aggregate-level (spatial) data, produced for multiple locations and time points, to characterize physical, built and social environments. We will then provide a brief overview about the linkage of such aggregate-level data with individual-level information about person's health in general, and brain health in particular.

Geospatial science and related tools enable spatial analysis and visualization of external environments in which we spend considerable amount of our lives (e.g., our residence, place of work, school, recreation or a commute path) and, in turn, evaluation of their impact on the individual's health. Datasets can be created at different levels of spatial granularity matching the goals of a given study and availability of relevant data. In Canada, for example, geographic units include six-digit postal codes, Canadian Census geographic units such as dissemination areas (DA; 400 to 700 persons) and census tracts (CT; 2,500 to 8,000 persons), or larger areas such as city districts. The spatial unit used to link geospatial datasets to health data varies; depending on the study and measures taken to protect confidentiality of study participants, this can be as precise as the exact street address or a postal code (half of a city block in dense urban area), or as coarse as a city district, a county, a province/state or a country. The temporal dimension depends on the type of data; it may range from data sampled monthly (e.g., air quality), annually (e.g., public transportation) or up to every 5 years (e.g., the Canadian Census).

The spatio-temporal datasets can be created using existing tools and databases provided by large GIS-based (Geographic Information Systems) organizations and companies, such as ESRI (www.esri.com), DMTI Spatial (www.dmtispatial.com), Google Earth Engine (https://earthengine.google.com), as well as open sources (e.g., www.openstreetmap.org), government sources (e.g., Statistics Canada), and academic organizations. The raw (initial) datasets are typically curated by data specialists and GIS technicians. In Canada, we have acquired, curated and disseminated geospatially coded information about physical and built environment through the Canadian Urban Environmental Health Research Consortium, CANUE (www.canuedata.ca/metadata.php) (Brook et al., 2018). Metrics derived from different sources can be combined to ask, for example, questions about the relationship between socio-economic indicators (e.g., household income) and built environment (e.g., access to parks), and thereby used to assess inequity in the spatial distribution of environmental good or hazards (Doiron et al., 2020). Figure 1 illustrates inequality in the access to parks and recreation (derived from Open Street Map data) across areas with the same (high) level of material deprivation [derived from Canadian Census data (Pampalon et al., 2012)].

In addition to sourcing and creating data about physical and built environments from existing databases [see Table 1 in (Paus, 2016)], one can also derive relevant metrics from new data streams such as high-resolution satellite and street-level imagery combined with machine-learning techniques (Weichenthal et al., 2019). For example, Google Street View allows investigators to assess different features of built environment using panoramic street-level images taken mostly by camera-equipped cars, while recent satellite technology provides daily coverage of most inhabited areas on Earth at a resolution of only a few meters. These geo-coded images can be rated for various features, such as signs of physical disorder (e.g., litter, graffiti), physical decay (e.g., poor conditions of sidewalks), type of stores, traffic, or street walkability (Odgers et al., 2012; Less et al., 2015); there are some limitations of this approach, however (Curtis et al., 2013). In turn, these computer vision and machine learning algorithms can exploit these image data to generate indirect indices of social environment (e.g., psychosocial stress) and physical environment (e.g. air or noise pollution) in a manner similar to that used by others to derive metrics such as living environment, health and crime (Suel et al., 2019).

As summarized in Table 1 (Social environment), there is a wealth of data that speak to basic – often self-reported - measures of socio-economic factors, such as education, employment, immigration, household spending habits or volunteering and giving, that are collected through governmental agencies (e.g., census) and national surveys. But one can also use data from digital streams, such as search-engine usage and Twitter, to generate new metrics of social environment relevant, for example, for attitudes vis-à-vis health and health interventions (e.g., vaccination), as well as social cohesion, social support and role models and, most recently, for the emerging issues related to environmental anxiety (Hickman et al., 2021; To et al., 2021; Soutar and Wand, 2022; Usher, 2022). The key advantage of this approach is its reliance on behavior rather than self-reports; as such, it is akin to ethology, a study of animal behavior by observing it in the “wild”. Twitter – for example - provides a rich source of information about the interests of its users; a short message (a tweet, up to 140 characters in length) sent by a user can be tagged by a hashtag identifying explicit topics of the message. One can retrieve all tweets posted or retweeted from a particular location during a 12-month period from Twitter's full firehose, using PowerTrack filtering language (), and: (1) map all Twitter users into geospatial units; (2) count the number of times each user posted or retweeted a tweet; (3) extract the tweet time-of-day; (4) determine the physical distance between each Twitter user and the geographic origin of a retweeted tweet; and (5) categorize tweets into various social themes, such as Economy, Politics, Health, Employment and Spending. The latter approach has been developed by Peng and colleagues in a study of interplay between real events and Twitter activity (Peng et al., 2017) whereby tweets are categorized using ~2,000 keywords related to the five social issues, followed by a machine-learning approach to filter out all irrelevant tweets (Peng et al., 2017). Facebook – an on-line social network – provides another window into social activities of its users. Various Facebook activities, such as wall postings, likes or status messages, can be used to classify users into different categories with regards to their interest, such as watching TV (Chunara et al., 2013), or their individual characteristics, such as the Big Five personality traits (Park et al., 2015). These can be, in turn, related to the geographic locations of Facebook users, thus creating additional layers of information about social features of a given geospatial unit. The above examples illustrate the power of digital ethology, an objective way of assessing social environment by measuring behavior through the individual's use of digital tools.

Once properly curated, all aggregate-level data (see Table 1 for examples) should be described using comprehensive metadata and indexed to different geographic units (postal codes, dissemination areas and other census geographies), as we have done previously (https://canuedata.ca/).

Ultimately, one is interested in linking aggregate-level “exposures” described above to the individual-level “outcomes”. Here we provide two examples of how one can achieve this goal by using: (1) administrative health databases; and (2) data acquired in research cohorts.

The above approach that combines aggregate-level assessments of “exposures” with individual-level information on “outcomes” reflects principles of “Big Data” and “Open Science”. We offer that no single study, however large, can achieve what is possible using this approach: (1) concurrent quantification of multiple standardized measures of physical, built and social environment; (2) large geospatial coverage (e.g., an entire country) implemented with a high geographic granularity; (3) extraction of the same measures at multiple timepoints, going back 20+ years; and (4) linkage of aggregate-level to individual-level data, the latter collected through administrative health databases and research cohorts. Although the availability of different types of data varies across space (geospatial granularity), time (frequency of sampling) and countries (monitoring tools, access), such limitations can be addressed – for example - by modeling (e.g., land-use regression to estimate air pollution) or through targeted surveys and other data collections. Overall, we suggest that this approach brings together two large communities of scholars and their trainees, namely health geographers with population neuroscientists, thus leveraging their respective knowledge and expertise to uncover pathways leading from our environment to brain health. It is clear that structural inequalities lead to health inequalities but the pathways mediating or moderating this relationship are largely unknown. A detailed, and concurrent, assessment of the physical, built and social environment on a large scale may allow one to approach the complexity of the dynamic relationship between “environment” and “external” and the human brain.

TP conceived this mini-review and wrote the first draft. JB and DD contributed to the first draft and reviewed critically the entire manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.