Flexible Multi-Level Models for Health Effects of Air Quality Actions: Results from the Children's Health Study

Abstract

Flexible Multi-Level Models for Health Effects of Air Quality Actions: Results from the Children's Health StudyAbstract Number:2892 Kiros Berhane*, Jim Gauderman, Chih-Chieh Chang, Robert Urmann, and Frank Gilliland Kiros Berhane* University of Southern California, United States, E-mail Address: [email protected] Search for more papers by this author , Jim Gauderman University of Southern California, United States, E-mail Address: [email protected] Search for more papers by this author , Chih-Chieh Chang University of Southern California, United States, E-mail Address: [email protected] Search for more papers by this author , Robert Urmann University of Southern California, United States, E-mail Address: [email protected] Search for more papers by this author , and Frank Gilliland University of Southern California, United States, E-mail Address: [email protected] Search for more papers by this author AbstractDetermination of the positive health impacts of declining air pollution levels is important in assessing the effectiveness (or lack thereof) of environmental regulatory policy. In data from children, the challenges in doing this properly include dealing with effects of puberty (e.g., non-linear growth trajectories over the entire childhood period), long term temporal trends due to unmeasured time-varying confounders, complexity in the potentially multi-pollutant dose response relationships and potentially synergistic effects acting at multiple levels of data aggregation (e.g., between communities, between individuals and across time). By using data on multiple cohorts of children spanning a long time period (1993-2012) from the Southern California Children’s Health Study as a motivation, we outline a flexible multi-level modeling framework that allows for the assessment of the healthimpacts of the recent declines in air pollution levels in Southern California. The focus will be on assessment of bronchitic symptoms (in children with or without asthma) and lung function development in children. The Generalized Linear Mixed Models setup will be used as a general modeling paradigm for all outcomes from the exponential family of distributions. For lung function outcomes, splines will be used to characterize growth trajectories for ‘homogenous’ sub-groups of children, with random effects used to capture subject-to-subject heterogeneity in growth trajectories. For bronchitic symptoms, the logistic setup will be used with a rich random effects structure to allow for heterogeneity of pollution effects across communities. In all cases, the effects of declining of pollution levels will be assessed using models of increasing complexity to deal with (a) non-constant effects across time,(b) non-linear dose-response relationships, (c) heterogeneity in pollution effects across multiple sites (e.g., communities, cities, regions, etc.), and/or (d) multi-pollutant health effects.