Spatial linear Land-Use Regression (LUR) is commonly used for long-term modeling of air pollution in support of exposure and epidemiological assessments. Machine Learning (ML) methods in conjunction with spatiotemporal modeling can provide more flexible exposure-relevant metrics and have been studied using different model structures. There is however a lack of comparisons of methods available within these two modeling frameworks, that can guide model/algorithm selection in air quality epidemiology. The present study compares thirteen algorithms for spatial/spatiotemporal modeling applied for daily maxima of 8-hour running averages of ambient ozone concentrations at spatial resolutions corresponding to census tracts, to support estimation of annual ozone design values across the contiguous US. These algorithms were selected from nine representative categories and trained using predictors that included chemistry-transport model predictions, meteorological factors, land use and land cover, and stationary and mobile emissions. To obtain the best predictive performance, model structures were optimized through a repeated coarse/fine grid search with expert knowledge. Six target-oriented validation strategies were used to prevent overfitting and avoid over-optimistic model evaluation results. In order to take full advantage of the power of different algorithms, we introduced tuning sample weights in spatiotemporal modeling to ensure predictive accuracy of peak concentrations, that is crucial for exposure assessments. In spatial modeling, four interpretation and visualization tools were introduced to explain predictions from different algorithms. Nonlinear ML methods achieved higher prediction accuracy than linear LUR, and the improvements were more significant for spatiotemporal modeling (nearly 10%-40% decrease of predicted RMSE). By tuning the sample weights, spatiotemporal models can predict concentrations used to calculate ozone design values that are comparable or even better than spatial models (nearly 30% decrease of cross-validated RMSE). We visualized the underlying nonlinear relationships, heterogeneous associations and complex interactions from the two best performing ML algorithms, i.e., Random Forest and Extreme Gradient Boosting, and found that the complex patterns were relatively less significant with respect to model accuracy for spatial modeling. Machine Learning can provide estimates that are actually more interpretable and practical than linear regression to improve accuracy in modeling human exposures. A careful design of hyperparameter tuning and flexible data splitting and validations is crucial to obtain reliable and stable results. Desirable/successful nonlinear models are expected to capture similar nonlinear patterns and interactions using different ML algorithms.
A conceptual/computational framework for exposure reconstruction from biomarker data combined with auxiliary exposure-related data is presented, evaluated with example applications, and examined in the context of future needs and opportunities. This framework employs physiologically based toxicokinetic (PBTK) modeling in conjunction with numerical “inversion” techniques. To quantify the value of different types of exposure data “accompanying” biomarker data, a study was conducted focusing on reconstructing exposures to chlorpyrifos, from measurements of its metabolite levels in urine. The study employed biomarker data as well as supporting exposure-related information from the National Human Exposure Assessment Survey (NHEXAS), Maryland, while the MENTOR-3P system (Modeling ENvironment for TOtal Risk with Physiologically based Pharmacokinetic modeling for Populations) was used for PBTK modeling. Recently proposed, simple numerical reconstruction methods were applied in this study, in conjunction with PBTK models. Two types of reconstructions were studied using (a) just the available biomarker and supporting exposure data and (b) synthetic data developed via augmenting available observations. Reconstruction using only available data resulted in a wide range of variation in estimated exposures. Reconstruction using synthetic data facilitated evaluation of numerical inversion methods and characterization of the value of additional information, such as study-specific data that can be collected in conjunction with the biomarker data. Although the NHEXAS data set provides a significant amount of supporting exposure-related information, especially when compared to national studies such as the National Health and Nutrition Examination Survey (NHANES), this information is still not adequate for detailed reconstruction of exposures under several conditions, as demonstrated here. The analysis presented here provides a starting point for introducing improved designs for future biomonitoring studies, from the perspective of exposure reconstruction; identifies specific limitations in existing exposure reconstruction methods that can be applied to population biomarker data; and suggests potential approaches for addressing exposure reconstruction from such data.
1 January 2006 in Journal of Toxicology and Environmental Health, Part B
Georgopoulos and Lioy (1994) Georgopoulos, P. G. and Lioy, P. J. 1994. Conceptual and theoretical aspects of human exposure and dose assessment. J. Expo. Anal. Environ. Epidemiol, 4: 253–285. [CSA] [Google Scholar] presented a theoretical framework for exposure analysis, incorporating multiple levels of empirical and mechanistic information while characterizing/reducing uncertainties. The present review summarizes efforts towards implementing that framework, through the development of a mechanistic source-to-dose Modeling ENvironment for TOtal Risks studies (MENTOR), a computational toolbox that provides various modeling and data analysis tools to facilitate assessment of cumulative and aggregate (multipathway) exposures to contaminant mixtures. MENTOR adopts a “Person Oriented Modeling” (POM) approach that can be applied to either specific individuals or to populations/subpopulations of interest; the latter is accomplished by defining samples of “virtual” individuals that statistically reproduce the physiological, demographic, etc., attributes of the populations studied. MENTOR implementations currently incorporate and expand USEPA's SHEDS (Stochastic Human Exposure and Dose Simulation) approach and consider multiple exposure routes (inhalation, food, drinking water intake; non-dietary ingestion; dermal absorption). Typically, simulations involve: (1) characterizing background levels of contaminants by combining model predictions and measurement studies; (2) characterizing multimedia levels and temporal profiles of contaminants in various residential and occupational microenvironments; (3) selecting sample populations that statistically reproduce essential demographics (age, gender, race, occupation, education) of relevant population units (e.g., census tracts); (4) developing activity event sequences for each member of the sample by matching attributes to entries of USEPA's Consolidated Human Activity Database (CHAD); (5) calculating intake rates for the sample population members, reflecting physiological attributes and activities pursued; (6) combining intake rates from multiple routes to assess exposures; (7) estimating target tissue doses with physiologically based dosimetry/toxicokinetic modeling.