Linking the Urban Environment and Health: An Innovative Methodology for Measuring Individual-Level Environmental Exposures

The paper addresses how built environment characteristics influence health and well-being, noting evidence for associations with air pollution, noise, green space/greenness, public transport, walkability/street centrality, food environments, and indoor qualities. Existing large-scale indicators often use aggregate geographies (e.g., postcodes) that miss street-level variation and individual exposure, potentially underestimating real-world exposure. There is inconsistency in how neighbourhood characteristics are measured and conceptualised, leading to contrasting findings (e.g., food environment associations with obesity, urbanicity with schizophrenia). The authors propose that precise, standardised, longitudinal measures of built environment features linked to individual-level longitudinal health and socio-economic data can improve causal inference and inform targeted planning and policy. They hypothesise that linking meaningful built environment features to whole-population databases will improve methods for correlating the built environment with health and social outcomes, generate new knowledge on environmental contributions to disease, and clarify how and when the built environment shapes long-term health outcomes, particularly for children growing up in unhealthy environments.

The paper reviews evidence linking environmental exposures to health outcomes: air pollution and noise with adverse health; green space/greenness with positive health effects; public transport and walkability with physical activity and related health benefits; mixed evidence for food environments and obesity; and the importance of indoor dwelling qualities. It highlights limitations in prior research, including reliance on aggregate spatial units (modifiable areal unit problem), street-level underestimation of exposures in national models, and the role of urban morphology (e.g., street canyons) in modulating exposure. It cites meta-narratives noting inconsistent definitions and measurements of neighbourhood characteristics in health studies, contributing to heterogeneous findings. The authors extend calls for standardised urbanicity measures to propose comprehensive, detailed, standardised longitudinal measures of built environment features. They emphasise the need to integrate nuanced urban planning metrics (e.g., street centrality, built form, land-use mix) into public health research and to consider social, cultural, and economic confounders.

Study area and data linkage: Bradford (population ~546,400) is among the 15 most deprived English districts. The Connected Bradford Whole System Data Linkage project pseudonymises NHS numbers and links longitudinal health, education, social care, housing, benefits, and crime data for ~800,000 citizens. The authors linked environmental indicators (EIs) to individuals using residential Unique Property Reference Numbers (UPRNs) via a five-step process: (1) derive EIs for each residential UPRN; (2) receive identifiable historic address data with pseudonymised NHS numbers; (3) georeference addresses and match to UPRNs using OS AddressBase Premium and adapted open-source R code (addressMatchR); (4) pseudonymise UPRNs and remove identifiers; (5) supply Connected Bradford with pseudonymised NHS numbers, UPRNs, and EIs. Historic address data (1950–2022) included >22.5 million address rows for 1,110,000 unique pseudonymised NHS numbers.

Geospatial base data: High-resolution Ordnance Survey (OS) datasets (MasterMap Topography, MasterMap Highways Roads and Path, AddressBase, Points of Interest), OS Open Greenspace, NaPTAN, Landsat 8–9, building height data, Energy Performance Certificates (EPC), local traffic (Trafficmaster) and air quality modelling data. A comprehensive pedestrian street network including roads and footpaths was constructed and segmented into 20 m segments.

Exposure construction framework: Nine spatial relationships representing static and dynamic exposures: (a) at/within the residence; (b) at the residential street; (c) at the urban block; (d) to the closest environmental feature; (e) to all features of a type; (f) features within a catchment; (g) circular buffers around the residence; (h) along routes to features; (i) properties of routes within a catchment. Four distance types: (1) straight-line (metric) distance; (2) network metric distance; (3) network angular distance; (4) network distance with exponential distance decay D = exp(−k·d), with varying decay parameters to reflect different user groups. PostgreSQL/PostGIS enabled large-scale computations.

Environmental domains and indicators (11 domains):

• Air quality: High-resolution (1×1 m) 2018 modelled concentrations of PM10, PM2.5, and NOx from the City of Bradford (Ricardo-AEA Rapid-Air), supplemented with UK emissions grids for rural areas. Calculated average and maximum values within buffers and catchments at multiple radii.
• Road traffic: Trafficmaster GPS-based data (since 2019) providing counts, average and free-flow speeds; congestion ratio C = 1 if x ≥ a, else x/a, where a is free-flow speed and x is average speed. Computed annual average/maximum bidirectional counts and congestion for peak/off-peak weekday periods at address street and 300 m catchments.
• Greenness and greenspace: NDVI from Landsat 8 (May 2020), averaged within varying radii/walking distances. Greenspace accessibility via OS Open Greenspace entrance points and classes; counts within guideline distances, distance-decay weighted counts, and size-weighted access.
• Public transport: NaPTAN stops/entrances (bus, coach, metro, rail, airports); walking distances to closest and counts within radii, distance-decay weighted counts.
• Walkability and land-use intensity: Pedestrian demand model combining land-use intensity (Shannon’s Diversity Index), transport accessibility, street centrality, and population density to generate a 25 m raster (2018). Aggregated average/maximum walkability and separate land-use intensity within varying radii.
• Street centrality: Angular closeness (integration) and betweenness (choice) centralities computed on the OS Highways + urban paths network at multiple radii; measured at the residential segment and within 300 m catchments.
• Built form: Building footprint, height, volume, floor area; block-level floor-space index (FSI), ground-space index (GSI), open-space ratio (OSR), and average layers of floors; EPC-derived dwelling/build form classes, construction age bands, storeys, and tenure.
• Indoor qualities: EPC-derived proxies for indoor environment (energy consumption/potential, lighting/heating/hot water costs, glazed area, floor area/height, number of heated/habitable rooms, extensions).
• Food environments: OS POI with text keywords to identify fast-food outlets; walking distance accessibility and distance-decay weighted counts; ratios of fast-food to all food outlets; measures at varying radii and along routes.

Temporal scope and generalisability: Core OS and related datasets are available annually since 2007, enabling longitudinal EI construction (2007–2022+). Methods are designed to generalise to other datasets with individual address information (e.g., postcodes/ZIP codes) and to other cities/countries, subject to governance approvals.

• Developed and linked a comprehensive suite of >500 environmental indicators across 11 domains to individual residential addresses for the entire Bradford population, enabling address-level precision of exposures and route-based measures.
• Successfully georeferenced >22.5 million historic address rows for 1,110,000 pseudonymised NHS numbers and linked to UPRNs for EI attachment, supporting longitudinal analyses within Connected Bradford (~800,000 citizens covered).
• Demonstrated substantial intra-urban spatial variation in exposures (e.g., fast-food outlet exposure higher in central urbanised areas; suburban/rural areas lower). Visualisations show address-level heterogeneity.
• Example “synthetic person” (Samina) illustrates combined exposures typical of Bradford residents: average PM2.5 during a short morning walk ≈ 9.01 µg/m³; exposure to 182 vehicles in the morning; average greenery 0.22 (NDVI), walkability 0.84, shop diversity 0.06; closest bus stop at 182.99 m; closer to fast-food outlets (561.89 m) than to public parks/gardens (889.26 m). Citywide summary statistics (Table 2) include: median distance to closest bus stop 182.99 m; median distance to closest fast-food outlet 561.89 m; median PM2.5 during short walk 9.01 µg/m³; median building height 5.70 m; median home size 78 m².
• In an applied test linking BiB cohort data with fast-food exposure, increased spatial precision in quantifying fast-food exposure around the home did not change associations with childhood obesity, challenging prior findings and underscoring definitional and confounding issues in food environment research.

The study addresses methodological and measurement gaps by moving from aggregate area-based indicators to individual-level, street- and route-based exposure measures that better capture lived environments. The framework’s nine spatial relationships and four distance types (including exponential distance decay) model potential interactions and accessibility, reducing reliance on arbitrary buffers and mitigating modifiable areal unit bias. Findings illustrate the feasibility and value of high-resolution, longitudinal environmental data linked to whole-population health records, enabling nuanced analyses of environmental determinants of health and interactions among domains (e.g., walkability, air quality, traffic). The null effect of increased precision on fast-food exposure–childhood obesity associations highlights the importance of consistent outlet definitions, comprehensive confounder control (e.g., deprivation, behaviour, genetics, car ownership), and careful operationalisation. The approach supports observational association studies and opens opportunities for causal inference (e.g., leveraging involuntary moves). The generalisability of the methods allows replication in other contexts and linkage to national longitudinal cohorts, potentially accelerating discoveries in urban health and informing targeted interventions and planning policies.

The paper presents a scalable methodology to construct and longitudinally link detailed, individual-level environmental exposure indicators to health and administrative datasets for an entire city. The resulting dataset fills a critical gap by providing street- and building-level measures across 11 domains, supporting research on associations and causal links between the built environment and health. The work offers a template for nationwide replication and aligns with WHO priorities to build city-level evidence on urban exposures across the life course. Future research should extend longitudinal EI construction, harmonise definitions (e.g., fast-food classifications), incorporate additional locally relevant exposures (e.g., surface temperature in hot climates), and exploit natural experiments (e.g., residential moves, policy changes) for causal inference. Integration with multiple longitudinal cohorts can further elucidate pathways and timing of environmental impacts on health.

• Data availability and comparability over time: Historic datasets vary in quality, resolution, completeness, and classification; e.g., consistent OS AddressBase address classifications are only available since 2013, requiring inference from alternative codes for earlier years, which may not align perfectly.
• Measurement and definitional issues: Neighbourhood definitions are inherently variable across individuals; despite distance decay and network measures, operational choices can influence results. Definitions of fast-food outlets vary across studies, affecting comparability and associations.
• Exposure modelling assumptions: Accessibility and proximity are proxies for interaction; actual behaviours, preferences, cultural norms, and socio-economic factors (e.g., car ownership) are not directly observed and may confound associations.
• Coverage and modelling limitations: National air pollution models may underrepresent street-level variation; local models were used where available, but rural areas relied on coarser emissions data. Traffic data temporal granularity and estimation for local roads can introduce uncertainty.
• Generalisability and governance: Application to other contexts depends on availability of analogous datasets, ethical approvals, and address-level linkages; privacy constraints limit direct data sharing.