Work

On the Limits of Efficiency in AI-Based Air Quality Monitoring for Health Risk Assessment

Year: 2026
Type: article
Abstract: Air pollution remains a leading global health risk, with chronic low-level toxicity documented below current regulatory limits. This has prompted calls to tighten air quality standards, yet the monitoring systems underlying risk assessment were not designed to operate at such low concentrations. Drawing on epidemiological evidence from the ELAPSE project on long-term exposure to major air pollutants, we develop a quantitative model of how AI-based air quality monitoring (AQM) quality affects the hazard ratio from Cox proportional hazards models. We show that AQM error systematically attenuates the estimated hazard ratio toward the null, and that this attenuation intensifies as concentrations decline. The reliability of the risk estimate is bounded above by the validation accuracy of the AI exposure model. Because the deep learning component of AQM error has an irreducible floor, increasing model complexity cannot eliminate the attenuation. We derive a quantitative sufficiency criterion that identifies the AQM accuracy beyond which further improvement yields no toxicologically meaningful gain and accumulates technical debt. The analysis identifies the epidemiological design, not the monitoring system, as the binding constraint on the accuracy of toxicity assessment at low concentrations, and provides a quantitative basis for deciding whether further investment in AQM is justified.
Source: Toxics
Language: English
Open Access status: gold
Estimated APC paid: $3,137
OpenAlex ID: W7236645936
FWCI:
Cites: 45
Cited by:
Related to: 10