Structural Equation Modeling in Exposure Assessment
M Davis, T Smith, Francine Laden, J Hart, LOUISE M. RYAN, Eric Garshick · Epidemiology · 2006
P-618 Abstract: Multi-tiered sampling approaches are common in environmental and occupational exposure modeling, where exposures at a given location are often modeled based on simultaneous measurements at multiple indoor and outdoor sites. The monitoring data from such studies is hierarchical by design, imposing a complex covariance structure that must be accounted for in order to obtain unbiased estimates of exposure. Multi-stage methods such as structural equation modeling (SEM) provide a useful alternative to simple linear regression in these cases. We test the SEM approach using data from a large exposure assessment of diesel particle exposure in the US trucking industry. The exposure assessment includes data from 36 different trucking terminals across the United States sampled between 2001 and 2005. Particle measurements were taken for elemental carbon, organic carbon, and PM2.5 at multiple indoor work locations, outdoor “background” locations, and by personal monitoring. Using the SEM method, we predict personal exposures as a function of work related exposure and smoking status. At the same time, work related exposure is predicted as a function of terminal characteristics, indoor ventilation, job location, and background exposure conditions. Finally, these background exposure conditions are predicted by weather, nearby source pollution, and other regional differences across terminal sites. The primary advantage of SEMs in this setting is that we can simultaneously predict exposures at each of these locations, while accounting for the complex correlation structure among the measurements. The significant results of this approach provide evidence to its usefulness as a modeling tool in these settings, with relatively high R2 values and significance levels. The outcome of applying the simple linear approach is quite different by comparison, which essentially ignores all of the data important in characterizing exposure to diesel particles at these locations, favoring only the first equation of personal exposure in a backwards elimination model. This is not surprising since applying this method requires the assumption that the sampling components are independent (can be broken down into their individual effects). For the epidemiologic study of lung cancer in the trucking industry, it is essential to understand the underlying factors that elevate diesel particles in these work locations in order to predict exposures at terminals not visited during the exposure assessment. SEMs provide a better alternative for modeling these types of exposure settings and we are further exploring these methods for use in similar datasets, including VOCs and on-road driver exposures.