A computational framework for agent-based assessment of multiple environmental exposures

Agent-based assessment of long-term personal exposure to environmental factors accounts for spatio-temporal variation in exposures along daily activity tracks of individuals. Application up to nationwide study populations requires integration of large data sets on environmental factors, personal behavior, and socio-economic status, as well as propagating uncertainties in these inputs to personal exposure values.

To develop and illustrate a methodology and software framework for agent-based personal exposure assessment for large cohorts, including uncertainty assessment.

We design an agent-based methodology that addresses the sparse information on individual activity patterns available in large cohorts. This methodology was implemented in a Python-based open-source and reusable framework, which was subsequently applied to assess exposure to air pollution and noise for 626,381 residential addresses in the province of Utrecht, the Netherlands. Air pollution exposures were also assessed across all addresses in Switzerland and the EPIC-NL cohort in the Netherlands.

The designed framework aggregates time by divisions marked by a particular pattern in individual movement (e.g., weekdays, weekend days). Movement over a division is represented by a sequence of activities, each with a duration and spatial context, i.e., the geographical area where the activity takes place. Several activity types are included, each with a methodology to assess the spatial context, for instance, the route from home to work location. Uncertainty in inputs is defined by probability distributions constrained by observational data, if available, like statistics on origin and destination of trips, and propagated to calculated personal exposures through Monte Carlo simulation. The exposures assessed through our framework result in minor to moderate differences with those calculated using home-based exposure (for Utrecht an r² of 0.79 for noise and 0.98 for nitrogen dioxide (NO₂) and particulate matter with aerodynamic diameters of 2.5 microns or smaller (PM_2.5), respectively), in particular leading to reduced contrast across the population in exposures.

Epidemiological studies on long-term effects of air pollution typically use a residential-based exposure assessment. However, it fails to account for individual mobility and spatial contrasts in environmental concentrations. While there is thus a need to investigate activity-based methods, their implementation is constrained by the lack of conceptual frameworks and software, particularly for large cohorts, which present unique demands regarding data inputs and computation. To address this gap, we introduce general concepts and a reusable, open-source software framework, designed for cluster computing, that can be applied consistently across a wide array of environmental factors and cohort studies.