Personalized Air Pollution Exposure Assessment and Its Association with Male Reproductive Health

Abstract

Fine particulate matter (PM2.5) is a leading environmental risk factor for human health, yet most epidemiologic studies estimate exposure using outdoor concentrations near participants' residential addresses. This residence-based approach assumes that individuals remain near home and that outdoor concentrations adequately represent indoor exposures, despite evidence that people spend approximately 90% of their time indoors and that indoor environments host a range of PM2.5 sources – including cooking, combustion-related heating, and other activities – that ambient monitoring networks do not capture. Reliance on ambient-only exposure assignment may therefore systematically mischaracterize personal PM2.5 exposure and may contribute to inconsistent findings in studies of PM2.5 and health outcomes. This has particular relevance for male reproductive health, where evidence linking PM2.5 to semen quality has been inconsistent despite biologically plausible mechanisms. This dissertation develops, evaluates, and applies enhanced approaches to personalized PM2.5 exposure assessment, drawing on data from two cohorts: the Reproductive Effects of Chemicals and Air Pollution (RECAP) study, a nationwide cohort of men nested within the Growing Up Today Study, and the Home-based Observation and Monitoring Exposure (HOME) study in Chelsea and Dorchester, Massachusetts. Chapter 1 develops and compares four daily PM2.5 exposure metrics among 161 participants in the RECAP study: a residential ambient metric based on the nearest U.S. Environmental Protection Agency (EPA) monitor (Home), a mobility-integrated ambient metric incorporating minute-level smartphone-based GPS data (GPS), a directly measured residential indoor metric from a real-time multipollutant monitoring platform (Indoor), and a personalized metric integrating indoor measurements with mobility-based ambient exposure (Personal). Home and GPS metrics were closely aligned across the exposure distribution (mean 8.36 vs 8.19 µg/m3; Spearman ρ = 0.91), indicating that incorporating mobility alone did not substantially change ambient exposure estimates. Indoor PM2.5 exhibited substantially greater variability and extreme upper-tail values (99th percentile 119.71 µg/m3), and the Personal metric shifted exposure distributions upward relative to ambient-only metrics, with the largest shifts observed in urban areas, high-density counties, and the Western and Northeastern United States. Chapter 2 applies the four exposure metrics from Chapter 1 to evaluate associations between short-term PM2.5 exposure and semen quality among 144 participants in the RECAP study. PM2.5 was averaged over the 0-90 days prior to semen sample collection, corresponding to the approximate duration of spermatogenesis, with supplementary analyses across biologically relevant sub-windows. Across all four exposure metrics, three covariate adjustment sets, and multiple exposure windows, associations between PM2.5 and sperm concentration, total sperm count, and motility were small in magnitude and imprecise, with confidence intervals consistently spanning the null. For example, in fully adjusted models, a 1 µg/m3 increase in residential ambient PM2.5 was associated with a 4.8% lower mean sperm concentration (95% CI: -11%, 1.9%). Findings were similar in sensitivity analyses restricted to participants with higher exposure data completeness. Chapter 3, previously published in Indoor Environments, addresses the accurate measurement of cooking activity – a primary driver of indoor PM2.5 peaks that is typically captured only through error-prone self-report. Using real-time temperature data collected near the stove and in the living room of 148 HOME study households, this chapter develops and validates an algorithm to identify cooking events at a 5-minute resolution. The algorithm identified substantially more cooking events than participants self-reported, with 65% of algorithm-detected events not reported in participants' Daily Activity Logs. In mixed-effects logistic regression models for elevated indoor PM2.5, the algorithm-derived cooking metric was much more strongly associated with peak PM2.5 (odds ratio: 2.85, 95% CI: 2.76, 2.95) than self-reported stove use (odds ratio: 1.22, 95% CI: 1.17, 1.27). Together, these studies demonstrate that residential indoor environments are a distinct and consequential domain of PM2.5 exposure, and that scalable, technology-enabled methods can meaningfully improve how personal PM2.5 exposure is characterized. The dissertation contributes methodological tools for personalized PM2.5 exposure assessment and highlights their particular relevance for research in communities disproportionately affected by air pollution.