Grantee Research Project Results

During Year 6 of the Center, each of the five Research Projects, the Air Pollution Core, and the Administrative Unit have continued to make progress on their objectives. In the remainder of this section of the report, we will discuss achievements of the Center. These will include discussion the added Center objective particle radioactivity, a report of our data management activities and human subject activities, as well as Center milestones. In later

sections of this report, each project is discussed separately. For each project, we summarize the objectives and present the progress and achievements, publications and presentations, inter- Center collaborative efforts, and changes (if any) to key personnel during the sixth year of the ACE Center.

During Year 6, we have continued our regular working group Research Meetings. These meetings feature presentations on their research by Center investigators, or by others whose research is relevant to Center objectives. These meetings also feature discussion of Center activities and serve as a forum for issues which may impact Center research. The Steering Committee (SC) has continued to work together to discuss project updates and data fusion. In addition, we have inaugurated research meetings focused on fostering interactions between researchers at Harvard and MIT, particularly among the post-doctoral fellows and doctoral students. These meetings are held quarterly.

Our Center researchers presented two seminars in the ACE Center webinar series during Year 6. On Thursday May 6, 2021, post-doctoral researcher Longxiang Li presented his research, focusing on a model to estimate spatiotemporally resolved residential exposure to radon in the Greater Boston area. Radon is the lead cause of lung cancer in non-smokers of the United States. Previous radon-related health studies relied on a county-level invariant radon model (aka Berkeley model) based on radon measurements collected in the 1980s. Massive short-term radon measurements provided by a local radon sampler manufacturer were leveraged to predict the monthly ZIP Code- level radon level with an ensemble learning method. The presentation also discussed the potential exposure misclassification caused by locating short-term radon samplers in the basement instead of in the upstairs living spaces, and a national model to estimate exposure to atmospheric particulate radioactivity, which is mostly sourced from long-term radon progeny. On Thursday November 4, 2021, doctoral student Barrak Alahmad presented his research on “Extreme Temperatures & Cardiovascular Health: Insights from a Multi-Country Multi-City Study.” The presentation focused on analysis of cardiovascular causes of death using a multinational database from 567 cities in 27 countries to investigate associations with extreme hot and cold temperatures. Using case-crossover models in each location and a mixed-effects meta-analytic framework to pool individual city estimates, the study found that extreme heat and cold increased the risk of dying from all cardiovascular causes, ischemic heart disease, stroke, heart failure and arrhythmia. Heart failure was associated with the highest risk of mortality.

During Year 6, we requested and were granted a second year of no-cost extension (NCE) for the Center for reasons detailed in the NCE request. In addition, the current global Covid-19 pandemic has also impacted the Center’s work. Although Harvard University resumed on-campus research activities starting in May 2020, this continues to be on a very limited basis and at minimal occupancy levels. Most of our faculty, staff, and student researchers have continued working remotely. Analysis of samples collected by the Air Pollution Core at our PM Supersite has resumed but remains delayed.

Project 1. Regional Air Pollution: Mixtures Characterization, Emission Inventories, Pollutant Trends, and Climate Impacts

During Year 6 of the Center, as part of Project 1, our research on Objective 5, particle radioactivity and radon exposures and their health effects, continued to grow during Year 6. Additionally, we have included investigations of the effects of solar activity and geomagnetic disturbances on air pollution and health outcomes. These studies have involved investigators from Project 1 and other center Projects. Our productivity is reflected by the 8 new accepted papers on these topics we have included in our publication list. In addition, we have submitted several additional manuscripts for publication, which are currently under review or revision. Published studies on solar activity and geomagnetic disturbances have shown that solar activity has a significant impact on daily ambient ultrafine particle concentrations that may affect the Earth’s climate system and human health (Vieira and Koutrakis 2021). Wang et al. (2021) found that solar activity and solar-driven geomagnetic disturbances were positively associated with BP in a cohort of elderly men, and Vieira et al. (2021) observed that solar and geomagnetic activity enhance the effects of air pollution on the risk of atrial fibrillation in people with implanted dual-chamber cardioverter defibrillators.

We have continued to measure particle gross alpha and beta activities from already collected/archived filters samples from our Harvard Supersite and indoor studies. This will make it possible to examine spatiotemporal patterns of ambient particle radioactivity levels and indoor outdoor relationships. A recent paper on ambient particle radioactivity by Yu et al. (2021) investigated the PM2.5 sources that influence ambient particle radioactivity, finding that regional sources have greatest impact, followed by local sources. Using gross alpha and beta activity of archived filters from indoor samples, we have manuscripts in preparation relating exposure to particle radioactivity with pulmonary function and with biomarkers of cardiovascular health and inflammatory response.

We have continued acquiring additional Radon exposure data, in New England and other states. We have published papers investigating the factors that influence indoor concentrations of radon in the midwestern U.S., showing that topography, geology, and soil composition were most influential (Carrion-Matta et al. 2021). In addition, because most radon measurements are made in basements rather than in living areas, we investigated the spatiotemporal gradients in ratios between the radon concentrations in the upstairs and basements in the northeastern and midwestern U.S. (Li et al. 2022). We have published a Radon prediction model (Li et al., 2021) which used the New England data set to identify predictors of radon indoor exposures and develop a radon exposure prediction model. Our New England Radon dataset and model results are being used to investigate the effects of radon in several cohorts by ACE Center researchers and students. In a presentation at the American Academy of Asthma, Allergy and Immunology Conference (Banzon et al. 2021) we showed a positive association of FeNO for asthmatic children with higher indoor Rn exposure, suggesting Rn exposure may be an important environmental risk factor for airway inflammation. From the same cohort of asthmatic children, we have a manuscript in preparation communicating a positive association between asthma symptoms and elevated radon exposures.

Project 1 has been supporting the development of a continuous Particle Alpha Radioactivity Sampler (PARS) that measures radioactivity from short-lived alpha-emitting (SLA) radon decay progeny. PARS provides time-resolved measurement of SLA at environmentally relevant levels indoors and outdoors. A manuscript is in preparation.

Publications during Year 6 supporting Project 1’s other objectives include a paper on the effects of climate change on indoor air pollution (Liang et al. 2021), an assessment of particulate air pollution and sources in Kuwait (Alahmad et al. 2021), estimation of ambient PM2.5 in Iraq and Kuwait between 2001 and 2018 using satellite remote sensing (Li J et al. 2021a), and the impacts of El Nino and La Nina on global surface dust levels between 1982 and 2019 (Li J et al. 2021b).

In addition, we also published a paper in which we found evidence of a statistically significant higher mortality risk among Medicare beneficiaries associated with living in proximity to and downwind of unconventional oil and gas wells, suggesting that primary air pollutants sourced from unconventional oil and gas exploration can be a major exposure pathway with adverse health effects in the elderly (LI L et al. 2022).

During Year 6, we also extended the 16-year GEOS-Chem simulation previously used as a predictor in the machine learning studies of Di et al. (2019a,b) and Requia et al. (2020). We generated a large array of model output variables for 2018-2020 at 0.5° x 0.625° horizontal resolution (about 50 km2), nested within a global simulation at 2° x 2.5° resolution. The output is currently stored in Google Drive and is in use by the Schwartz research group to update their machine learning models of US air quality. The link is here: https://drive.google.com/drive/folders/1dDU8jKHoy1LHL_fng3Y1Ju0Feql5hjBl

In addition, during Year 6, in support of Objectives 3 and 4, we finalized and published work on:

•            Response of dust emissions in southwestern North America to 21st century trends in climate, CO2 fertilization, and land use. [This work was accepted in late 2020, published in early 2021].

Climate models predict a shift toward warmer and drier environments in southwestern North America. The consequences of such a shift for dust mobilization and dust concentration are unknown, but they could have large implications for human health, given the connections between dust inhalation and disease. For this project, we link a dynamic vegetation model (LPJ-LMfire) to a chemical transport model (GEOSChem) to assess the impacts of future changes in three factors – climate, CO2 fertilization, and land use practices – on vegetation in this region. From there, we investigate the impacts of changing vegetation on dust mobilization and assess the net effect on fine dust concentration (defined as dust particles less than 2.5 μm in diameter) on surface air quality. In the most extreme warming scenario (RCP8.5), we find that surface temperatures in southwestern North America during the season of greatest dust emissions (March April, May) warm by 3.3 K and precipitation decreases by nearly 40% by 2100. These conditions lead to vegetation dieback and an increase in dust-producing bare ground. Enhanced CO2 fertilization, however, offsets the modeled effects of warming temperatures and rainfall deficit on vegetation in some areas of the southwestern United States. Considering all three factors in RCP8.5 scenario, dust concentrations decrease over Arizona and New Mexico in spring by the late 21st century due to greater CO2 fertilization and a more densely vegetated environment, which inhibits dust mobilization. Along Mexico's northern border, dust concentrations increase as a result of land use intensification. In contrast, when CO2 fertilization is not considered in the RCP8.5 scenario, vegetation cover declines significantly across most of the domain by 2100, leading to widespread increases in fine dust concentrations, especially in southeastern New Mexico (up to ~2.0 µg m-3 relative to the present day) and along the border between New Mexico and Mexico (up to ~2.5 µg m-3). Our results have implications for human health, especially for the health of the indigenous people who make up a large percentage of the population in this region. (Li Y et al., 2021)

•            Excess of COVID-19 cases and deaths due to fine particulate matter exposure during the 2020 wildfires in the United States.

The year 2020 brought unimaginable challenges in public health, with the confluence of the COVID-19 pandemic and wildfires across the western United States. Wildfires produce high levels of fine particulate matter (PM2.5). Recent studies reported that short-term exposure to PM2.5 is associated with increased risk of COVID-19 cases and deaths. We acquired and linked publicly available daily data on PM2.5, the number of COVID-19 cases and deaths, and other confounders for 92 western U.S. counties that were affected by the 2020 wildfires. We estimated the association between short-term exposure to PM2.5 during the wildfires and the epidemiological dynamics of COVID-19 cases and deaths. We adjusted for several time-varying confounding factors (e.g., weather, seasonality, long-term trends, mobility, and population size). We found strong evidence that wildfires amplified the effect of short-term exposure to PM2.5 on COVID-19 cases and deaths, although with substantial heterogeneity across counties. (Zhou et al., 2021)

•            Estimating the health effects of environmental mixtures using principal stratification.

The control of ambient air quality in the United States has been a major public health success since the passing of the Clean Air Act, with particulate matter (PM) reductions resulting in an estimated 160,000 premature deaths prevented in 2010 alone. Currently, public policy is oriented around lowering the levels of individual pollutants and this focus has driven the nature of much epidemiological research. Recently, attention has been given to viewing air pollution as a complex mixture and to developing a multi-pollutant approach to controlling ambient concentrations. We present a statistical approach for estimating the health impacts of complex environmental mixtures using a mixture-altering contrast, which is any comparison, intervention, policy, or natural experiment that changes a mixture's composition. We combine the notion of mixture-altering contrasts with sliced inverse regression, propensity score matching, and principal stratification to assess the health effects of different air pollution chemical mixtures. We demonstrate the application of this approach in an analysis of the health effects of wildfire particulate matter air pollution in the Western United States. (Peng et al., provisionally accepted, November 2021).

Inter-Center Collaborations: During Year 6, Project 1 researchers concluded a collaborative study with Roger Peng (Johns Hopkins, SEARCH Center). The study presented a statistical approach for estimating the health impacts of complex environmental mixtures, using wildfires as a case study. The paper (Peng et al.) was provisionally accepted in November 2021.  During Year 6, Project 1 researchers continued collaborative study with Michelle Bell (Yale University, SEARCH Center) on the direct and indirect effects of solar activity on health outcomes. Preliminary findings indicate that solar activity may modify or mediate PM2.5-related risk of circulatory and respiratory hospitalizations; a manuscript from this collaboration is in preparation.

Project 2. Air Pollutant Mixtures in Eastern Massachusetts: Spatial Multi-resolution Analysis of Trends, Effects of Modifiable Factors, Climate, and Particle-induced Mortality

In addition to the papers published during the project period, Project 2 investigators made significant progress on multiple fronts, including continued progress on quantifying the health effects of early life outcomes and exposures. This work includes the following:

In 2019-2020, we introduced a new Bayesian nonparametric ensemble framework that adaptively combines models based on their predictive accuracy in the over space and time and also nonparametrically models the ensemble's predictive cumulative density function (CDF) so that the model's quantification of the predictive uncertainty is consistent with observed data. We showed the model can improve upon the predictive performance of an ensemble model with naive distribution assumptions when the data distribution is complex. We applied the method to data on fine particle levels in eastern Massachusetts, USA. In the current reporting period, we have extended the approach nationally, developing a fitting method known as Random Fourier Features (RFF) that scales up to big data. We have applied the approach to ensemble (i.e. integrate) results from four different models that yield predictions over the continental U.S., including models generated recently by Harvard, other ACE Centers (CACES), among several others. We have generated uncertainty factors across the U.S., including measures of disagreement across the models, and overall uncertainties of the ensembled predictions, and are currently identifying factors most associated with these different sources of error, including pollution levels, population density, distance from monitoring locations, elevation, and temporal factors such as season and year. This work was submitted as an abstract to AIStats, a meeting focusing on the intersection of artificial intelligence, machine learning, statistics, and related areas. Final model fits are being generated now and a paper describing the work is currently in progress.

In 2019-2020, as outlined in last year’s progress report, we examined which elemental components of PM2.5 are responsible for previously reported associations between PM2.5 and neonatal BP in 1131 mother-infant pairs in Project Viva, a Boston-area prebirth cohort. Our findings suggest that prenatal exposures to particulate matter components, and particularly nickel, may increase newborn BP. This paper was published in Journal of the American Heart Association after the funding year ended (Zanobetti et al. 2021). We have a new A-clustering sparse canonical correlation analysis (Aclust-sCCA) pipeline to investigate epigenetic mechanisms underlying these associations between infant outcomes and PM2.5 composition. We identified several genes in which Nickel is associated with DNA methylation, suggesting potential biological mechanism of the blood pressure findings reported by Zanobetti et al. (2021). We have developed an R package that implements the methods and are currently writing up the resulting manuscript.

In past work funded by this Center we have illustrated the usefulness of distributed lag models to identify critical windows of susceptibility to in utero air pollution (and temperature) exposures during pregnancy. In this project period, we have extended existing distributed lag models to accommodate effect heterogeneity and identify subpopulations experiencing the strongest pollution effects. That is, the model simultaneously identifies groups of subjects exhibiting the strongest associations with exposure and estimates the distributed lag function representing critical windows of exposure during pregnancy in this group. We used the approach to estimate the relationship between weekly exposures to fine particulate matter during gestation and birth weight in an administrative Colorado birth cohort. We identified maternal body mass index (BMI), age, Hispanic designation, and education as modifiers of the distributed lag effects and find non- Hispanics with increased BMI to be a susceptible population (Mork et al. 2021).

Project 3.  Causal Estimates of Effects of Regional and National Pollution Mixtures on Health: Providing Tools for Policy Makers

Year 6 has been a productive reporting period for Project 3. We continued our efforts in causal modeling, effects of acute and chronic exposures, and spatiotemporal modeling.

In Wei 2022 we reported on the association of PM2.5, NO2, and O3 with asthma hospital admissions in the National Medicaid population. Since Medicaid supplies health care for the medically indigent, this is a vulnerable population. We saw significant effects of all three pollutant, which was stronger in neighborhoods with higher deprivation. The effect size per unit exposure was greater at concentrations below the NAAQS for PM2.5 and NO2.

In Lu 2022 we examined the association of air pollutants with school standardized test scores at all school districts in the United States. We found that higher concentrations of PM2.5 and NO2 were associated with lower test scores for both math and English Language Arts, and O3 was associated with lower test scores for Math.

In Qiu 2022a we examined the association of air pollutants with psychiatric symptoms in the Normative Aging study, and found that NO2 and O3, but not PM2.5 were associated with those symptoms.

In Hood 2022 we examined the effects of both short and long term PM2.5 exposure on the metabolome in women undergoing fertility treatment. We found significant alterations for both exposures and identified 17 pathways associated with acute exposure 15 pathways with chronic exposure and 7 pathways involving inflammation and lipid metabolism associated with both.

In Qiu 2022b we found that exposure to PM2.5 and NO2, but not O3 was associated with hospital admissions for schizophrenia, bipolar disease, and depression in the national Medicare population.

In Li 2022 we examined the association of long-term exposure to PM2.5 components with the incidence of dementia in the Medicare population. We found the strongest significant effects for BC and SO4-2. This differs from previous Medicare studies which only examined exacerbations and not incidence.

In Vanoli 2022 we showed that PM2.5 exposure postnatally modified the growth trajectories of children in a cohort study in Boston.

In Rahman 2022 we showed that prenatal PM2.5 exposure, particularly in the first two trimesters, was associated with the incidence of autism in the Southern California Kaiser Health System.

In Huang 2021 we reported that both PM2.5 and NO2 were associated with an increased risk of placental abruption in a cohort of ~700,000 pregnancies in New York City.

In Nassan 2021a we examined the association of PM2.5 species and metabolomic profile in the Normative Aging Study. We found the most significant association for short-term exposure were with ultrafine particles, Ni, V, and K, while for long-term exposures the most significant exposures were to BC, V, Zn, and Ni. A pathway analysis indicated pertubance of the glycerophospholipid, sphingolipid, and glutathione pathways, which are involved in inflammation, oxidative stress, immunity, and nucleic acid damage and repair.

In Vodonos and Schwartz 2021 we showed that risk assessments using county level exposure and baseline rates underestimate the mortality effects of PM2.5 and demonstrated disparities in risk within individual cities using a 1km exposure model and baseline rates at the zipcode level.

In Wei 2021 we used a propensity score model fit with a random forest to estimate the causal effects of PM2.5, NO2, and O3 on mortality in persons aged 65 and older.

In Zanobetti 2021 we examined the association of PM2.5 components with newborn blood pressure in Project Viva and found a significant association with Ni.

In Nassan 2021b we examined long-term exposure to NO2, O3, and PM2.5 and its association with metabolomic changes. We found changes in 58 metabolites associated with PM2.5, 15 with NO2 and 6 with O3.

In Wang 2022 (in press) we have examined and found associations of PM2.5 components with changes in DNA methylation in an epigenome wide study.

In Danesh Yazdi 2021 we showed the air pollution concentrations below the NAAQS for PM2.5, NO2, and O3 were associated with increased risk of death.

Project 4.  A Causal Inference Framework to Support Policy Decisions by Evaluating the Effectiveness of Past Air Pollution Control Strategies for the Entire United States

Year 6 has seen considerable progress towards objectives 2, 3, and 4 with development across three overlapping domains: 1) reduced-complexity models, with extensions to sources other than power plants 2) statistical methods for causal inference, and 3) environmental justice and health disparities research. Several key studies are highlighted here to emphasize these areas, but are not a comprehensive list of research output for the project, which is compiled elsewhere.

Domain 1: Reduced-Complexity Models.

Against the backdrop of strengthened validation of the HyADS reduced-complexity model for the analysis of power plant pollution, we have continued to deploy it to characterize long-term trends in power plant specific pollution in two signature projects very near completion. In particular, Henneman et al (2022a) evaluates 22 years of coal power plant exposures to document how declining coal emissions have impacted exposures and exposure inequities in the United States. Henneman et al (2022b) uses the same long-term database of coal source impacts to evaluate Medicare deaths attributable to coal power plant pollution. Both of these works represent major culminating works of the HyADS reduced complexity model for environmental justice and epidemiological applications, and should be submitted by mid 2022.

Year 6 has also seen an expansion of scope to include other sources of pollution. Preliminary work has used the HyADS model to characterize spatially- and temporally- varying exposures to flaring associated with oil extraction in Texas. Early progress mirrors that from the power plant domain, with indications of new insights relative to existing proximity-based flaring exposures and more sophisticated modeling efforts that cannot easily scale to large-scale epidemiologic studies.

Finally, we have extended the core philosophy of reduced complexity modeling to exposures associated with mobile sources. In Henneman et al (2021), we find that simple and widely available proxies of on-road emissions can reproduce observed near-road concentration gradients and changes over time, as compared with ground-based observations, CMAQ simulations, and satellite products. This work points towards the possibility of using such metrics for large scale epidemiological or accountability studies of polices to reduce emissions of traffic-related air pollution.

• Henneman, Rasel, Choirat, Zigler (2022a). 22 Years of Declining PM2.5 Coal Source Exposure and Inequities in the US. In progress.
• Henneman, Choirat, Dominici, Zigler (2022b). Coal’s Toll: 20 years of Medicare Deaths Attributable to Electricity Generation. In progress.
• Henneman L , Shen H, Hogrefe C, Russell AG, Zigler CM (2021). Four decades of United States mobile source pollutants: spatial-temporal trends assessed by ground-based monitors, air quality models, and satellites. Environmental Science & Technology, 55(2), 882-892.

Domain 2: Statistical Methods for Causal Inference.  We have continued to develop novel methodologies for causal inference, with particular focus on network interference, causal mediation analysis, and machine learning. Work continues on methods for causal inference with bipartite interference, most notably with generalized network propensity score estimation in Zigler et al (2022), which is currently being revised for journal publication. Extensions to the case of time-varying treatments are in process to directly accommodate the temporal nature of pollution transport. While most of the project’s work on bipartite interference has used HyADS outputs to characterize the structure of interference, work in Wikle et al. (2022) offers a mechanistic statistical model for annual sulfate concentrations related to power plant SO2 emissions, representing a statistical contribution to spatiotempral modeling in its own right as well as a feasible approach to estimating interference structures in analyses of causal inference with interference. Extensions to the explicit causal inference case are ongoing. A commentary on issues related to causal inference with multivariate environmental exposures, done in the context of a study of power plant closings, appears in Zigler (2021), and holds relevance to many of the aims of the project. Finally, we have made key progress in machine learning methods for the ubiquitous problem of adjusting for meteorology in analyses of air pollution in Tec et al (2022+), which is nearly complete and ready for submission. This works holds serious potential to advance methodology for spatial confounding adjustment and conduct meteorological detrending in studies of long-term trends in air pollution, and follows a similar spirit to work in Qiu et al. (2022+), being completed with Project 5 investigators.

• Wikle N , Hanks E, Henneman L, and Zigler CM (2022). A mechanistic model of annual sulfate concentrations in the United States. Journal of the American Statistical Association In press.
• Zigler, Forastiere, and Mealli (2022). Bipartite Interference and Air Pollution Transport: Estimating Health Effects of Power Plant Interventions, https://arxiv.org/abs/2012.04831
• Zigler CM. Invited Commentary: The promise and pitfalls of causal inference with multivariate environmental exposures. American Journal of Epidemiology, 2021, 190(12), 2658-2661.
• Qiu, Zigler, and Selin (2022+). Statistical and Machine Learning Methods for Evaluating Trends in PM2.5 and O3 under Changing Meteorological Conditions. In Progress.
• Tec, Scott, and Zigler (2022+). Weather2vec for Causal Infernece: Latent Representation of Non-Local Confounding with Applications in Air Pollution Studies. In Progress.

Domain 3: Health Disparities and Environmental Justice.

We have continued to extend original project objectives to encompass studies of health disparities and environmental justice. Work in Jbaily et al. (2022) develop a data platform to ink demographic data from the UCS Census Bureau and ambient PM2.5 exposures across the United States and study exposure disparities across racial/ethnic and income distributions of the US population. Dey et al (2020) offer a similar perspective at the intersection of air pollution, COVID-19, and racial inequity. Douda et al (2021) specifically addresses racial disparities in preterm birth associated with exposure to coal-fired power plant pollution.

Ongoing work, in coordination with investigators from the CACES center, takes a more local view of air pollution exposure disparities. Ongoing research is investigating discrepancies between a high-quality mobile monitoring campaign with fine-scale intraurban characterization of pollution and a national land use regression model. In particular, we are learning whether certain types of areas urban areas have pollution that is over or under predicted by land use regression (relative to mobile monitoring) and relating whether easily obtainable features of the urban environment relate to these discrepancies, as well as whether these discrepancies tend to accrue to certain population groups.

• Jbaily, Zhouy, Liu, Lee, Kamereddine, Verguet, and Dominici (2022). Air Pollution Exposure Disparities Across US Population and Income Groups. Nature, 601, 228-233.

Project 1. Regional Air Pollution: Mixtures Characterization, Emission Inventories, Pollutant Trends, and Climate Impacts

We will continue our current research in Objectives 3 and 4, related to the smoke impacts on air quality in the United States. In particular, we began validation of the NOAA Hazard Mapping System (HMS), which provides near-real time maps of smoke plume outlines and density across