Exposure Assessment Strategies: Modeling

PO119
Exposure Assessment Strategies: Modeling

Tuesday, June 2, 2015, 2:00 PM - 4:00 PM

SR-119-01 A Method for Constructing Informative Priors for Bayesian Modeling of Occupational Hygiene Data

H. Quick, CDC, Atlanta, GA; T. Huynh, University of Minnesota, Minneapolis, MN; G. Ramachandran, University of Minnesota, Minneapolis, MN

Objective: In man​​y occupational hygiene settings, the demand for more accurate, more precise results is at odds with limited resources. To combat this, practitioners have begun using Bayesian methods to incorporate prior information into their statistical models, hoping to gain more refined inference from their data. This is not without risk. Incorporating prior information that disagrees with the information contained in data can lead to spurious conclusions, particularly if the prior is too informative. The objective of this paper is to propose a method for constructing informative prior distributions for normal and lognormal data that are intuitive to specify and robust to bias.

Methods: We conducted computer simulation studies to compare 5 different priors and used root mean square error as our evaluation metric.

Results: In complete uncensored data case, our prior performed similarly as the non-informative prior. For censored data cases, our prior showed a marked improvement over the non-informative priors.

Conclusions: Our method for specifying informative priors is viable tool for analyzing industrial hygiene data.


SR-119-02 Predicting Construction Related Silica Exposures Using Input from Chamber and Field Studies

S. Arnold, G. Ramachandran, University of Minnesota, Minneapolis, MN

Objective: Predicting silica exposures is an important first step in effectively managing and preventing overexposure. Mathematical models are useful for predicting exposures in construction environments but their use has been limited because of the inherent challenges of quantitatively defining model inputs. This study was conducted using chamber and field studies to better characterize these inputs.

Methods: Dry wall sanding was simulated by a professional drywall finisher in an exposure (11.9 m3) chamber under highly controlled conditions. Respirable dust and crystalline silica (quartz) generation rates were quantified by back-calculating the generation rate from the airborne concentration. Other model parameters, such as the ventilation rate, room dimensions and work practice information were collected in the field, providing high quality model inputs for three models; the well-mixed room, near field far field and turbulent eddy diffusion models. Personal exposure measurements were collected using real time and time weighted (integrated) sampling. Measured and modeled exposure estimates were compared.

Results: A generation rate was derived under highly controlled conditions for pole and hand sanding, respectively. The use of the chamber to better characterize model inputs that are difficult to evaluate in the field was very useful and facilitated the use of several models to predict exposures to silica.

Conclusions: Using chamber to quantify those parameters inherently challenging to characterize in the field in combination with field data enhances model parameter estimates and ultimately, model performance.


SR-119-03 Characterization of Chrysotile Asbestos Fiber Removal Rates from Air

J. Sahmel, H. Avens, P. Scott, A. Burns, C. Barlow, K. Unice, A. Madl, J. Henshaw, D. Paustenbach, Cardno ChemRisk, Boulder, CO

Objective: Some theoretical estimates of fiber settling rates have raised concern that asbestos fibers may continue to pose an inhalation hazard long after a primary fiber release event has ended. However, only a small number of empirical data sets exist regarding rates of fiber removal from air. A study was conducted to gain further empirical data on this topic.

Methods: Air samples were collected while clothing contaminated with chrysotile asbestos was handled for 15 min. To evaluate the decrease in chrysotile concentrations after clothes handling ended, air sampling was performed during the first 5 min, first 15 min, and second 15 min after the activity ceased. This process was repeated for six total events. The study chamber was about 58 cubic meters with 3.5 air changes per hour. Samples were analyzed by NIOSH Method 7402. The PCME asbestos concentrations during the fiber settling/removal period were fit to an exponential decay curve from which fiber half-life and time to removal of 99% of fibers were estimated. In addition, the results were compared to theoretical estimates of the fiber removal rate that would be expected from ventilation alone and from gravitational fiber settling alone.

Results: PCME chrysotile concentrations during the 15-30 min after clothes handling ended decreased 80%-92% compared to concentrations during active clothes handling. Across all six events, the fiber half-life was estimated to be 4.8 min (95% CI: 3.4-8.6) and the time to 99% fiber removal was estimated to be 32.2 min (95% CI: 22.4-56.9). Theoretical calculations based on ventilation alone yielded an estimated half-life of 11.9 min and 99% fiber removal by 78.9 min. Theoretical calculations based only on gravitational settling yielded half-lives of 0.6 to 24 hr. depending on the PCME fiber dimensions evaluated, while estimated times to 99% removal ranged from 4 to 160 hr.

Conclusions: Nearly all fibers were removed from the air within approximately 30 min. This study is consistent with other studies in finding that theoretical estimates of fiber removal rates based solely on gravitational settling over-estimate the time that fibers remain airborne. The measured fiber removal rates were also greater than estimates based on ventilation and gravitational effects combined, suggesting that other mechanisms such as adhesion to surfaces, diffusion, agglomeration of particles to one another, and electrification also play an important role in fiber removal processes.


SR-119-04 Turbulent Eddy Diffusion Models in Exposure Assessment: Determination of the Eddy Diffusion Coefficient

Y. Shao, G. Ramachandran, S. Arnold, S. Arce, University of Minnesota, Minneapolis, MN

Objective: The turbulent eddy diffusion models could provide details of a continuous concentration gradient with distance from the source. But the use of these models in exposure assessment is limited due to the lack of knowledge on the isotropic eddy diffusion coefficient, D. Recent studies have hypothesized that this parameter is related to the air changes per hour (ACH) and the room dimensions. The goal of this work is to determine this relationship. 

Methods: An exposure chamber (11.9 m3, L/W/H: 2.8×2.15×2m) was constructed with the ability to vary flow rate from 0.5 to 5 ACHs. A continuous Acetone source was located in the chamber center and was controlled using a Harvard syringe with a generation rate from 0.05 to 0.5 ml/min. Two Dräger X-AM 7000 vapor monitors were used with a multiplexer to measure the chemical concentrations at six symmetric locations on both sides of the source in the chamber. A spherical turbulent eddy diffusion model accounting for wall effects and/or advection was used to predict concentrations at the same locations. The difference between the measured and predicted concentrations was minimized by a MATLAB least-squares function to best estimate D. Experiments were conducted at five different values of ACH and three generation rates. The effect of data analysis averaging times on the value of D was also studied.

Results: The model can predict average exposures in close proximity to emitting sources, as long as the turbulent diffusion coefficient (D) can be estimated. The advection and wall effects were not negligible in our chamber where some monitoring locations were 0.4 m away from the walls or/and the intake filters. The monitored value was about 2 times higher than the prediction without wall-reflections correction. The turbulent diffusion model was sensitive to the value of D, where a 0.0001 m2/s scale change of D could result in a 1~5 ppm scale change of predicted concentration. At 0.7~0.8 ACH, the estimated D was 0.0005 m2/s. There is a significant positive linear correlation between ACH and the turbulent diffusion coefficient. 

Conclusions: The consideration of wall effects and/or advection in the turbulent diffusion model is very necessary to a relatively small test chamber. The ability to estimate the turbulent diffusion coefficient; air change rate and room dimensions can significantly increase the ease of use of this model for occupational exposure assessment. 


SR-119-05 Exposure Models: How Accurate Are They? Evaluating Exposure Model Performance under Highly Controlled Conditions

S. Arnold, G. Ramachandran, University of Minnesota, Minneapolis, MN

Objective: Exposure modeling, which has been shown to improve decision making across a broad range of domains provides a systematic and transparent approach for making exposure judgments. But they may be undervalued as tools for making accurate exposure judgments as occupational hygienists (OHs) lack training opportunities that would provide immediate feedback on their judgment accuracy, and therefore they maybe underutilizing models. Guidance directing OHs on which model would produce the most accurate exposure estimate under a defined set of conditions is needed. The models need to be systematically evaluated in both chamber and field environments. In this work, studies were conducted to systematically evaluate a range of models and provide guidance on their appropriate use under highly controlled (chamber) conditions.

Methods: A full size exposure chamber (11.9 m3) was constructed, providing a highly controlled environment in which simulations could be conducted and model parameters could be measured with a high degree of certainty. 162 studies (3 agents x 3 generation rates x 3 ventilation rates x 3 repetitions x 2 chamber configurations) were conducted for three chemical agents (acetone, toluene, methylene chloride) under varying air exchange rates and generation rates in which airborne concentrations were measured in real-time, providing time varying exposure data. Chamber conditions were varied to simulate well mixed room and 2-zone environments. A range of exposure models (Well Mixed Room, 2 Zone, and Turbulent Eddy Diffusion) were evaluated. Measured and modeled exposures were compared and performance assessed according to ASTM 5157.

Results: A rich database of exposure model parameter values was developed, which will be useful for applying these models. Concordance between modeled and measured exposures was generally very good, but varied with chamber conditions. For example, in the acetone studies when the ventilation was set to low ~ 0.5 ACH, the modeled and measured time varying concentration were essentially the same, but as steady state conditions were approached, the modelled concentration was 10–20% less than the measured concentration. At higher ACH (~2–5), the modeled time varying and steady state concentrations was within 5% of the measured time varying and steady state concentrations.

Conclusions: These studies will be pivotal in developing model selection guidance on selecting and applying these models.


SR-119-06 Exposure Models: How Accurate Are They? Evaluating Exposure Model Performance under Real World Conditions

S. Arnold, G. Ramachandran, University of Minnesota, Minneapolis, MN

Objective: Exposure modeling is an approach found to be useful for guiding accurate exposure judgments across many professional domains, but may be undervalued and underutilized in OH. This may be because OHs lack training opportunities that would provide immediate feedback on their judgment accuracy and because little guidance exists for directing OHs on which model would produce the most accurate exposure estimate under a defined set of conditions. To meet this need the models need to be systematically evaluated in both chamber and field environments. We will present the results from studies conducted in several work places to systematically evaluate a range of models and provide guidance on their appropriate use under real world conditions.

Methods: Five studies were conducted at four different work places, providing real world environments in which measurements were collected while specific tasks were performed (weighing powder, finishing drywall, cleaning industrial equipment, finger nail manicuring and floor cleaning). Model parameters were directly measured whenever possible, or estimated when they could not be measured, providing parameter values with a greater degree of uncertainty compared to parameter values generated under highly controlled conditions. Real time and integrated personal samples were collected, along with time activity budgets. A range of exposure models presented in IH Mod (Well Mixed Room, Two Zone, and Turbulent Eddy Diffusion) were evaluated.

Results: A rich database of real world exposure model parameter values was developed, which will be useful for applying these models. The highly variable conditions typical of construction environments was reflected in more variable model parameter values, compared to the highly controlled conditions found in the manufacturing setting. Differences in working environments and tasks were important determinants in model performance. For example, while the 2 Zone model showed concordance with measured exposures in the highly controlled manufacturing setting where equipment was cleaned, there was less concordance between modeled and measured exposures in the construction setting during drywall finishing.

Conclusions: A rich database of real world exposure parameter values was developed that will be useful in applying the Well Mixed Room, Two Zone, Turbulent Eddy Diffusion models. These studies will be pivotal in developing model selection guidance.​