Lianne Sheppard’s research while affiliated with University of Mary Washington and other places

Introduction: Exposure to ambient air pollution may increase the risk of cardiovascular disease events and mortality, but prior publications have primarily included administrative cohorts with outcomes that have not been individually reviewed and with air pollution estimates created without cohort-specific exposure monitoring. Multi-Ethnic Study of Atherosclerosis (MESA) is a multi-site cohort study designed specifically to prospectively collect and adjudicate cardiovascular disease (CVD) events. MESA Air recruited additional participants into sub-cohorts for enhanced air pollution variation and sample size. Research Question: The aim of this analysis was to characterize the relationship between long-term exposure to nitrogen dioxide (NO 2 ) and fine particulate matter (PM 2.5 ) and all-cause mortality and CVD events. Methods: Air pollution exposure was assessed using address history with a purpose-built exposure model incorporating cohort-specific monitoring including measurement and validation at participant homes. We used Cox models to assess the risk of rolling 2-year average exposures on all cause-mortality and on a composite CVD endpoint (definite angina, probable angina with revascularization, myocardial infarction, atherosclerosis or other CVD death, resuscitated cardiac arrest, and stroke). Models were stratified for baseline hazard by age, sub-cohort, and recruitment year and were additionally adjusted for age, sex, race/ethnicity, field center, smoking/second-hand smoke, pack-years, physical activity, education, income, neighborhood socioeconomic status, and statin use. Results: MESA Air participants were aged 44-87 years at enrollment between 2000 and 2007; follow-up averaged 14 years. 6,915 participants had follow-up for events, NO 2 exposure, and covariate information. We observed 1,442 deaths and 985 CVD events. The interquartile range over all 2-year averages was 10.5-23.1 ppb for NO 2 and 10.1-14.9 µ/m ³ for PM 2.5 . The adjusted hazard ratio (aHR) for a 10 ppb increment in NO 2 was 1.38 (95% CI: 1.17, 1.64) for all-cause mortality and 1.16 (95% CI: 0.95, 1.42) for incident CVD events. The aHR for a 5 µg/m ³ increment in PM 2.5 was 1.20 (95% CI: 0.99, 1.46) for all-cause mortality and 1.15 (95% CI: 0.95, 1.39) for incident CVD events Conclusions: These results add to growing literature demonstrating an association between air pollution exposure, mortality, and CVD in a cohort with well-characterized clinical endpoints and cohort-specific exposure assessment.

Growing evidence links ultrafine particles (UFP) to neurotoxicity, but human studies remain limited. Various mobile monitoring approaches have been used to collect repeated short-term air pollution samples and develop human exposure models. However, whether design choices impact epidemiologic inferences, including for UFP and cognitive function, remains unclear. We evaluated the adjusted association between UFP number concentration (PNC) and late-life cognitive function (Cognitive Abilities Screening Instrument – Item Response Theory [CASI-IRT]) in the Adult Changes in Thought cohort (N=5,283) by leveraging an extensive roadside mobile monitoring campaign specifically designed for epidemiologic application. To assess the impact of common, reduced monitoring approaches on this association, we repeatedly subsampled UFP measures from the extensive campaign, developed exposure models, and evaluated the degree to which associations were impacted. In a reduced adjustment model, the mean baseline CASI-IRT score decreased by 0.020 (95% CI: -0.036, -0.004) per 1,900 pt/cm³ increase in PNC. Associations were consistent across most sampling designs, including fewer visits (median point estimate -0.019, IQR: -0.022, -0.016), fewer seasons (-0.019, IQR: -0.021, -0.016), and unbalanced sampling (-0.018, IQR: -0.022, -0.016), with very unbalanced designs yielding more variable estimates. In the primary adjustment model, the CASI-IRT score increased by 0.002 (95% CI: -0.016, 0.020), with similar estimates across fewer visits (0.002, IQR: -0.001, 0.004), fewer seasons (0.000, IQR: -0.001, 0.003), and unbalanced sampling (0.001, IQR: -0.001, 0.004). Rush hour designs were more similar (0.002, IQR: 0.000, 0.003) than business hour designs (0.006, IQR: 0.005, 0.007), but the opposite was true when temporal adjustments were applied (rush: -0.003, IQR: -0.005, -0.001; business: 0.002, IQR: 0.001, 0.004). We observed similar trends in sensitivity and secondary analyses. We found no link between UFP exposures and late-life cognition. While various monitoring approaches may be used to capture epidemiologic inferences, extending beyond weekday business and rush hours is likely important.

Dimension reduction is often the first step in statistical modeling or prediction of multivariate spatial data. However, most existing dimension reduction techniques do not account for the spatial correlation between observations and do not take the downstream modeling task into consideration when finding the lower-dimensional representation. We formalize the closeness of approximation to the original data and the utility of lower-dimensional scores for downstream modeling as two complementary, sometimes conflicting, metrics for dimension reduction. We illustrate how existing methodologies fall into this framework and propose a flexible dimension reduction algorithm that achieves the optimal trade-off. We derive a computationally simple form for our algorithm and illustrate its performance through simulation studies, as well as two applications in air pollution modeling and spatial transcriptomics.

Exposure assessment is fundamental to air pollution cohort studies. The objective is to predict air pollution exposures for study subjects at locations without data in order to optimize our ability to learn about health effects of air pollution. In addition to generating accurate predictions to minimize exposure measurement error, understanding the mechanism captured by the model is another crucial aspect that may not always be straightforward due to the complex nature of machine learning methods, as well as the lack of unifying notions of variable importance. This is further complicated in air pollution modeling by the presence of spatial correlation. We tackle these challenges in two datasets: sulfur (S) from regulatory United States national PM2.5 sub-species data and ultrafine particles (UFP) from a new Seattle-area traffic-related air pollution dataset. Our key contribution is a leave-one-out approach for variable importance that leads to interpretable and comparable measures for a broad class of models with separable mean and covariance components. We illustrate our approach with several spatial machine learning models, and it clearly highlights the difference in model mechanisms, even for those producing similar predictions. We leverage insights from this variable importance measure to assess the relative utilities of two exposure models for S and UFP that have similar out-of-sample prediction accuracies but appear to draw on different types of spatial information to make predictions.

Background Statistical models of air pollution enable intra-urban characterization of pollutant concentrations, benefiting exposure assessment for environmental epidemiology. The new generation of low-cost sensors facilitate the deployment of dense monitoring networks and can potentially be used to improve intra-urban models of air pollution. Objective Develop and evaluate a spatiotemporal model for nitrogen dioxide (NO2) in the Puget Sound region of WA, USA for the Adult Changes in Thought Air Pollution (ACT-AP) study and assess the contribution of low-cost sensor data to the model’s performance through cross-validation. Methods We developed a spatiotemporal NO2 model for the study region incorporating data from 11 agency locations, 364 supplementary monitoring locations, and 117 low-cost sensor (LCS) locations for the 1996–2020 time period. Model features included long-term time trends and dimension-reduced land use regression. We evaluated the contribution of LCS network data by comparing models fit with and without sensor data using cross-validated (CV) summary performance statistics. Results The best performing model had one time trend and geographic covariates summarized into three partial least squares components. The model, fit with LCS data, performed as well as other recent studies (agency cross-validation: CV- root mean square error (RMSE) = 2.5 ppb NO2; CV- coefficient of determination (R2${R}^{2}$) = 0.85). Predictions of NO2 concentrations developed with LCS were higher at residential locations compared to a model without LCS, especially in recent years. While LCS did not provide a strong performance gain at agency sites (CV-RMSE = 2.8 ppb NO2; CV-R2${R}^{2}$ = 0.82 without LCS), at residential locations, the improvement was substantial, with RMSE = 3.8 ppb NO2 and R2${R}^{2}$ = 0.08 (without LCS), compared to CV-RMSE = 2.8 ppb NO2 and CV-R2${R}^{2}$ = 0.51 (with LCS). Impact We developed a spatiotemporal model for nitrogen dioxide (NO2) pollution in Washington’s Puget Sound region for epidemiologic exposure assessment for the Adult Changes in Thought Air Pollution study. We examined the impact of including low-cost sensor data in the NO2 model and found the additional spatial information the sensors provided predicted NO2 concentrations that were higher than without low-cost sensors, particularly in recent years. We did not observe a clear, substantial improvement in cross-validation performance over a similar model fit without low-cost sensor data; however, the prediction improvement with low-cost sensors at residential locations was substantial. The performance gains from low-cost sensors may have been attenuated due to spatial information provided by other supplementary monitoring data.

Background: Mobile monitoring campaigns are frequently used to develop air pollution exposure models to be used in health studies. Monitoring designs vary substantially, however, and it is unclear how design features impact exposure assessment models or health inferences. Methods: We conducted a case study of the impact of mobile monitoring study design on ultrafine particle (UFP) exposure assessment and the estimated association between UFP and late-life cognitive function. We leveraged UFP measures from an extensive mobile monitoring campaign consisting of 309 temporary roadside stationary sites, each with ~29 temporally balanced visits over a year. We subsampled the data following common field designs: fewer visits per site (4-12); shorter campaign durations (1-4 seasons); business or rush hours (unadjusted and temporally adjusted); and an unbalanced number of visits where high variability sites received more or less visits than low variability sites. We developed annual average UFP exposure models with the resulting data and ran health analyses to estimate the adjusted association between five-year UFP exposure and baseline cognitive function (Cognitive Abilities Screening Instrument – Item Response Theory [CASI-IRT]) in the Adult Changes in Thought (ACT) cohort (N=5,409). Results: The reference UFP all-data exposure model (R2=0.65) estimated that the adjusted mean CASI-IRT was lower by 0.020 (95% CI: -0.036, -0.004) per each 1,900 pt/cm3. More restricted designs generally produced poorer performing exposure models (median R2: 0.40-0.61), with business hours (R2: 0.40-0.45), one-season (R2: 0.43), and unbalanced visits (R2: 0.48) performing worst. Health inferences were broadly consistent with those from the all-data exposure model with just fewer visits per location, but they had more bias and/or were inconsistent across campaigns with fewer seasons, business or rush hours, or unbalanced visits. Business and rush hour designs had the most biased and attenuated health estimates. Conclusions: Thoughtful monitoring design can improve exposure models and subsequent health inferences.

Background: In contrast to fine particles, less is known of the inflammatory and coagulation impacts of coarse particulate matter (PM10-2.5, particulate matter with aerodynamic diameter ≤10μm and>2.5μm). Toxicological research suggests that these pathways might be important processes by which PM10-2.5 impacts health, but there are relatively few epidemiological studies due to a lack of a national PM10-2.5 monitoring network. Objectives: We used new spatiotemporal exposure models to examine associations of both 1-y and 1-month average PM10-2.5 concentrations with markers of inflammation and coagulation. Methods: We leveraged data from 7,071 Multi-Ethnic Study of Atherosclerosis and ancillary study participants 45-84 y of age who had repeated plasma measures of inflammatory and coagulation biomarkers. We estimated PM10-2.5 at participant addresses 1 y and 1 month before each of up to four exams (2000-2012) using spatiotemporal models that incorporated satellite, regulatory monitoring, and local geographic data and accounted for spatial correlation. We used random effects models to estimate associations with interleukin-6 (IL-6), C-reactive protein (CRP), fibrinogen, and D-dimer, controlling for potential confounders. Results: Increases in PM10-2.5 were not associated with greater levels of inflammation or coagulation. A 10-μg/m3 increase in annual average PM10-2.5 was associated with a 2.5% decrease in CRP [95% confidence interval (CI): -5.5, 0.6]. We saw no association between annual average PM10-2.5 and the other markers (IL-6: -0.7%, 95% CI: -2.6, 1.2; fibrinogen: -0.3%, 95% CI: -0.9, 0.3; D-dimer: -0.2%, 95% CI: -2.6, 2.4). Associations consistently showed that a 10-μg/m3 increase in 1-month average PM10-2.5 was associated with reduced inflammation and coagulation, though none were distinguishable from no association (IL-6: -1.2%, 95% CI: -3.0 , 0.5; CRP: -2.5%, 95% CI: -5.3, 0.4; fibrinogen: -0.4%, 95% CI: -1.0, 0.1; D-dimer: -2.0%, 95% CI: -4.3, 0.3). Discussion: We found no evidence that PM10-2.5 is associated with higher inflammation or coagulation levels. More research is needed to determine whether the inflammation and coagulation pathways are as important in explaining observed PM10-2.5 health impacts in humans as they have been shown to be in toxicology studies or whether PM10-2.5 might impact human health through alternative biological mechanisms. https://doi.org/10.1289/EHP12972.