Probabilistic Machine Learning with Low-Cost Sensor Networks for Occupational Exposure Assessment and Industrial Hygiene Decision Making.

Occupational exposure assessments are dominated by small sample sizes and low spatial and temporal resolution with a focus on conducting Occupational Safety and Health Administration regulatory compliance sampling. However, this style of exposure assessment is likely to underestimate true exposures and their variability in sampled areas, and entirely fail to characterize exposures in unsampled areas. The American Industrial Hygiene Association (AIHA) has developed a more realistic system of exposure ratings based on estimating the 95th percentiles of the exposures that can be used to better represent exposure uncertainty and exposure variability for decision-making; however, the ratings can still fail to capture realistic exposure with small sample sizes. Therefore, low-cost sensor networks consisting of numerous lower-quality sensors have been used to measure occupational exposures at a high spatiotemporal scale. However, the sensors must be calibrated in the laboratory or field to a reference standard. Using data from carbon monoxide (CO) sensors deployed in a heavy equipment manufacturing facility for eight months from August 2017 to March 2018, we demonstrate that machine learning with probabilistic gradient boosted decision trees (GBDT) can model raw sensor readings to reference data highly accurately, entirely removing the need for laboratory calibration. Further, we indicate how the machine learning models can produce probabilistic hazard maps of the manufacturing floor, creating a visual tool for assessing facility-wide exposures. Additionally, the ability to have a fully modeled prediction distribution for each measurement enables the use of the AIHA exposure ratings, which provide an enhanced industrial decision-making framework as opposed to simply determining if a small number of measurements were above or below a pertinent occupational exposure limit. Lastly, we show how a probabilistic modeling exposure assessment with high spatiotemporal resolution data can prevent exposure misclassifications associated with traditional models that rely exclusively on mean or point predictions.

Continuous in-home PM2.5 concentrations of smokers with and without a history of respiratory exacerbations in Iowa, during and after an air purifier intervention.

BACKGROUND: Americans spend most of their time indoors. Indoor particulate matter (PM) 2.5 µm and smaller (PM2.5) concentrations often exceed ambient concentrations. Therefore, we tested whether the use of an air purifying device (electrostatic precipitator, ESP) could reduce PM2.5 in homes of smokers with and without respiratory exacerbations, compared with baseline. METHODS: We assessed PM2.5 concentrations in homes of subjects with and without a recent (≤3 years) history of respiratory exacerbation. We compared PM2.5 concentrations during 1 month of ESP use with those during 1 month without ESP use. RESULTS: Our study included 19 subjects (53-80 years old), nine with a history of respiratory exacerbation. Geometric mean (GM) PM2.5 and median GM daily peak PM2.5 were significantly lower during ESP deployment compared with the equivalent time-period without the ESP (GSD = 0.50 and 0.37 µg/m3, respectively, p < 0.001). PM2.5 in homes of respiratory exacerbators tended (p < 0.14) to be higher than PM2.5 in homes of those without a history of respiratory exacerbation. CONCLUSIONS: Subjects with a history of respiratory exacerbation tended to have higher mean, median, and mean peak PM2.5 concentrations compared with homes of subjects without a history of exacerbations. The ESP intervention reduced in-home PM2.5 concentrations, demonstrating its utility in reducing indoor exposures. NOVELTY OF STUDY: Our work characterizes PM air pollution concentrations in homes of study subjects with and without respiratory exacerbations. We demonstrate that PM concentrations tend to be higher in homes of participants with respiratory exacerbations, and that the use of an inexpensive air purifier resulted in significantly lower daily average PM concentrations than when the purifier was not present. Our results provide a helpful intervention strategy for purifying indoor air and may be useful for susceptible populations.

Porous Polyurethane Foam for Use as a Particle Collection Substrate in a Nanoparticle Respiratory Deposition Sampler.

Porous polyurethane foam was evaluated to replace the eight nylon meshes used as a substrate to collect nanoparticles in the Nanoparticle Respiratory Deposition (NRD) sampler. Cylindrical (25-mm diameter by 40-mm deep) foam with 110 pores per inch was housed in a 25-mm-diameter conductive polypropylene cassette cowl compatible with the NRD sampler. Pristine foam and nylon meshes were evaluated for metals content via elemental analysis. The size-selective collection efficiency of the foam was evaluated using salt (NaCl) and metal fume aerosols in independent tests. Collection efficiencies were compared to the nanoparticulate matter (NPM) criterion and a semi-empirical model for foam. Changes in collection efficiency and pressure drop of the foam and nylon meshes were measured after loading with metal fume particles as measures of substrate performance. Substantially less titanium was found in the foam (0.173 µg sampler-1) compared to the nylon mesh (125 µg sampler-1), improving the detection capabilities of the NRD sampler for titanium dioxide particles. The foam collection efficiency was similar to that of the nylon meshes and the NPM criterion (R2 = 0.98, for NaCl), although the semi-empirical model underestimated the experimental efficiency (R2 = 0.38). The pressure drop across the foam was 8% that of the nylon meshes when pristine and changed minimally with metal fume loading (~ 19 mg). In contrast, the pores of the nylon meshes clogged after loading with ~ 1 mg metal fume. These results indicate that foam is a suitable substrate to collect metal (except for cadmium) nanoparticles in the NRD sampler.

Toxicity of copper oxide nanoparticles in lung epithelial cells exposed at the air-liquid interface compared with in vivo assessment.

The toxicity of spark-generated copper oxide nanoparticles (CuONPs) was evaluated in human bronchial epithelial cells (HBEC) and lung adenocarcinoma cells (A549 cells) using an in vitro air-liquid interface (ALI) exposure system. Dose-response results were compared to in vivo inhalation and instillation studies of CuONPs. Cells were exposed to filtered, particle-free clean air (controls) or spark-generated CuONPs. The number median diameter, geometric standard deviation and total number concentration of CuONPs were 9.2 nm, 1.48 and 2.27×10(7)particles/cm(3), respectively. Outcome measures included cell viability, cytotoxicity, oxidative stress and proinflammatory chemokine production. Exposure to clean air (2 or 4h) did not induce toxicity in HBEC or A549 cells. Compared with controls, CuONP exposures significantly reduced cell viability, increased lactate dehydrogenase (LDH) release and elevated levels of reactive oxygen species (ROS) and IL-8 in a dose-dependent manner. A549 cells were significantly more susceptible to CuONP effects than HBEC. Antioxidant treatment reduced CuONP-induced cytotoxicity. When dose was expressed per area of exposed epithelium there was good agreement of toxicity measures with murine in vivo studies. This demonstrates that in vitro ALI studies can provide meaningful data on nanotoxicity of metal oxides.

Toxicity assessment of air-delivered particle-bound polybrominated diphenyl ethers.

Human exposure to polybrominated diphenyl ethers (PBDEs) can occur via ingestion of indoor dust, inhalation of PBDE-contaminated air and dust-bound PBDEs. However, few studies have examined the pulmonary toxicity of particle-bound PBDEs, mainly due to the lack of an appropriate particle-cell exposure system. In this study we developed an in vitro exposure system capable of generating particle-bound PBDEs mimicking dusts containing PBDE congeners (BDEs 35, 47 and 99) and delivering them directly onto lung cells grown at an air-liquid interface (ALI). The silica particles and particles-coated with PBDEs ranged in diameter from 4.3 to 4.5 µm and were delivered to cells with no apparent aggregation. This experimental set up demonstrated high reproducibility and sensitivity for dosing control and distribution of particles. ALI exposure of cells to PBDE-bound particles significantly decreased cell viability and induced reactive oxygen species generation in A549 and NCI-H358 cells. In male Sprague-Dawley rats exposed via intratracheal insufflation (0.6 mg/rat), particle-bound PBDE exposures induced inflammatory responses with increased recruitment of neutrophils to the lungs compared to sham-exposed rats. The present study clearly indicates the potential of our exposure system for studying the toxicity of particle-bound compounds.

Validation of an in vitro exposure system for toxicity assessment of air-delivered nanomaterials.

To overcome the limitations of in vitro exposure of submerged lung cells to nanoparticles (NPs), we validated an integrated low flow system capable of generating and depositing airborne NPs directly onto cells at an air-liquid interface (ALI). The in vitro exposure system was shown to provide uniform and controlled dosing of particles with 70.3% efficiency to epithelial cells grown on transwells. This system delivered a continuous airborne exposure of NPs to lung cells without loss of cell viability in repeated 4h exposure periods. We sequentially exposed cells to air-delivered copper (Cu) NPs in vitro to compare toxicity results to our prior in vivo inhalation studies. The evaluation of cellular dosimetry indicated that a large amount of Cu was taken up, dissolved and released into the basolateral medium (62% of total mass). Exposure to Cu NPs decreased cell viability to 73% (p<0.01) and significantly (p<0.05) elevated levels of lactate dehydrogenase, intracellular reactive oxygen species and interleukin-8 that mirrored our findings from subacute in vivo inhalation studies in mice. Our results show that this exposure system is useful for screening of NP toxicity in a manner that represents cellular responses of the pulmonary epithelium in vivo.

Distribution of particle and gas concentrations in Swine gestation confined animal feeding operations.

OBJECTIVES: Dust mass concentrations, temperatures, and carbon dioxide concentrations were mapped in a modern, 1048-pen swine gestation barn in winter, spring, and summer. METHODS: In each season, two technicians measured respirable mass concentrations with an aerosol photometer and temperatures and carbon dioxide concentrations with an indoor air quality monitor at 60 positions in the barn. Stationary photometers were also deployed to measure mass concentrations during mapping at five fixed locations. RESULTS: In winter when building ventilation rates were low (center-barn mean air velocity=0.34 m s(-1), 68 fpm) to conserve heat within the barn, mass and carbon dioxide concentrations were highest (mass geometric mean, GM=0.50 mg m(-3); CO2 GM=2060 ppm) and fairly uniform over space (mass geometric standard deviation, GSD=1.48; CO2 GSD=1.24). Concentrations were lowest in summer (mass GM=0.13 mg m(-3); CO2 GM=610 ppm) when ventilation rates were high (center-barn mean air velocity=0.99 m s(-1), 196 fpm) to provide cooling. Spatial gradients were greatest in spring (mass GSD=2.11; CO2 GSD=1.50) with low concentrations observed near the building intake, increasing to higher concentrations at the building exhaust. CONCLUSIONS: Mass concentrations obtained in mapping were generally consistent with those obtained from stationary monitors. A moderately strong linear relationship (R2=0.60) was observed between the log of photometer-measured mass concentration and the log of carbon dioxide concentration, suggesting that carbon dioxide may be an inexpensive alternative to assessing air quality in a swine barn. These results indicate that ventilation can effectively reduce contaminant levels in addition to controlling temperature.

Single-particle SEM-EDX analysis of iron-containing coarse particulate matter in an urban environment: sources and distribution of iron within Cleveland, Ohio.

The physicochemical properties of coarse-mode, iron-containing particles and their temporal and spatial distributions are poorly understood. Single-particle analysis combining X-ray elemental mapping and computer-controlled scanning electron microscopy (CCSEM-EDX) of passively collected particles was used to investigate the physicochemical properties of iron-containing particles in Cleveland, OH, in summer 2008 (Aug-Sept), summer 2009 (July-Aug), and winter 2010 (Feb-March). The most abundant classes of iron-containing particles were iron oxide fly ash, mineral dust, NaCl-containing agglomerates (likely from road salt), and Ca-S containing agglomerates (likely from slag, a byproduct of steel production, or gypsum in road salt). The mass concentrations of anthropogenic fly ash particles were highest in the Flats region (downtown) and decreased with distance away from this region. The concentrations of fly ash in the Flats region were consistent with interannual changes in steel production. These particles were observed to be highly spherical in the Flats region, but less so after transport away from downtown. This change in morphology may be attributed to atmospheric processing. Overall, this work demonstrates that the method of passive collection with single-particle analysis by electron microscopy is a powerful tool to study spatial and temporal gradients in components of coarse particles. These gradients may correlate with human health effects associated with exposure to coarse-mode particulate matter.

Innovative application of fluoro tagging to trace airborne particulate and gas-phase polybrominated diphenyl ether exposures.

Polybrominated diphenyl ethers (PBDEs) are flame retardants applied as coatings to many consumer products, including household items. PBDEs are released and produce airborne vapors and dusts. Inhalation of particle-phase and/or gas-phase PBDEs is therefore a major route of exposure. In an attempt to mimic realistic airborne exposures, actual uptake, and deposition of particles and vapors, we prepared and characterized particles for future animal exposure studies. To trace the particles in environmental and biological systems, we employed fluoro tagging. We synthesized, characterized, and employed three PBDE congeners, 35, 47, and 99, and five fluoro-substituted PBDEs (F-PBDEs), 17-F5' 25-F5', 28-F3', 35-F5', 47-F3, and 99-F3', for this study. The PBDE congeners were selected because they are commonly found in house dust. For that reason, we coated spherical silica particles of 3 microm and C18 endcapped silica as representative and inert support materials, with 20, 30, and 40% PBDEs. We determined the particle size distributions by aerodynamic particle size spectrometry and the morphology by scanning electron microscopy. The suitability of the fluoro-tagged tracers to mimic their corresponding parent PBDEs was investigated by extraction studies from spiked blood serum. Our study is of fundamental importance to the development of xenobiotic tracers for monitoring routes of human exposure to PBDEs and understanding uptake of PBDEs from particles and vapors.

Ultrafine and respirable particles in an automotive grey iron foundry.

Ultrafine particle number and respirable particle mass concentrations were measured throughout an automotive grey iron foundry during winter, spring and summer using a particle concentration mapping procedure. Substantial temporal and spatial variability was observed in all seasons and attributed, in part, to the batch nature of operations, process emission variability and frequent work interruptions. The need for fine mapping grids was demonstrated, where elevations in particle concentrations were highly localized. Ultrafine particle concentrations were generally greatest during winter when incoming make-up air was heated with direct fire, natural gas burners. Make-up air drawn from roof level had elevated respirable mass and ultrafine number concentrations above ambient outdoor levels, suggesting inadvertent recirculation of foundry process emissions. Elevated respirable mass concentrations were highly localized on occasions (e.g. abrasive blasting and grinding), depended on the area within the facility where measurements were obtained, but were largely unaffected by season. Particle sources were further characterized by measuring their respective number and mass concentrations by particle size. Sources that contributed to ultrafine particles included process-specific sources (e.g. melting and pouring operations), and non-process sources (e.g. direct fire natural gas heating units, a liquid propane-fuelled sweeper and cigarette smoking) were additionally identified.

Laboratory and field evaluation of crystallized DOW 704 oil on the performance of the Well Impactor Ninety-Six fFine particulate matter fractionator.

Subsequent to the 1997 promulgation of the Federal Reference Method (FRM) for monitoring fine particulate matter (PM2.5) in ambient air, U.S. Environmental Protection Agency (EPA) received reports that the DOW 704 diffusion oil used in the method's Well Impactor Ninety-Six (WINS) fractionator would occasionally crystallize during field use, particularly under wintertime conditions. Although the frequency of occurrence on a nationwide basis was low, uncertainties existed as to whether crystallization of the DOW 704 oil may adversely affect a sampling event's data quality. In response to these concerns, EPA and the State of Connecticut Department of Environmental Protection jointly conducted a series of specialized tests to determine whether crystallized oil adversely affected the performance of the WINS fractionator. In the laboratory, an experimental setup used dry ice to artificially induce crystallization of the diffusion oil under controlled conditions. Using primary polystyrene latex calibration aerosols, standard size-selective performance tests of the WINS fractionator showed that neither the position nor the shape of the WINS particle size fractionation curve was substantially influenced by the crystallization of the DOW 704 oil. No large particle bounce from the crystallized impaction surface was observed. During wintertime field tests, crystallization of the DOW 704 oil did not adversely affect measured PM2.5 concentrations. Regression of measurements with crystallized DOW 704 versus liquid dioctyl sebacate (DOS) oil produced slope, intercept, and R2 values of 0.98, 0.1, and 0.997 microg/m3, respectively. Additional field tests validated the use of DOS as an effective impaction substrate. As a result of these laboratory and field tests, DOS oil has been approved by EPA as a substitute for DOW 704 oil. Since the field deployment of DOS oil in 2001, users of this alternative oil have not reported any operational problems associated with its use in the PM2.5 FRM. Limited field evaluation of the BGI very sharp cut cyclone indicates that it provides a viable alternative to the WINS fractionator.