.EXPOSURE ASSESSMENT MODELS
By Kelvin Vine March 2006
CONTENTS
INTRODUCTION
What is modelling?
The tiered approach to exposure modelling
Tier 1 - Specialised Priority Setting Tools
Tier 2 - Screening Level Tools
Tier 3 - Higher Tier Tools
Purpose of this essay
BENEFITS OF MODELLING
Estimation of exposure
Generalisation of results
Limitations of modelling
Inaccuracies
Validation
CONCLUSION
REFERENCES
Exposure modelling has been carried out in some sense since the very beginnings of the Health and Hygiene field. Regulators have always had to make routine estimates of how much of a chemical hazardous to health workers could be exposed to under certain conditions. As the Health and Hygiene field advanced, this estimation has taken on a more precise and empirical nature in order to take as much of the guess-work out as possible.
Modelling is defined as (International Labour Organisation, 2000):
“… a logical or empirical construct which allows estimation of individual or population exposure parameters from available input data”.
The goal of modelling therefore is to effectively predict the exposure to toxic substances quickly, safely and accurately for a wide range of possible scenarios.
The tiered approach to exposure modelling
Exposure models are categorised under 3 tiers, the criteria for classification being the accuracy and thoroughness of predictions. In most cases, models from all three tiers will be used as to effectively model any given scenario (Environmental Protection Agency, 2007). The three tiers are:
Tier 1 - Specialised Priority Setting Tools
Usually utilised as the first step in exposure modelling, tools in this tier are not very accurate or tailored to any particular environment but can be used to quickly give a rough estimate of the risk from large numbers of substances. In this way toxic substances may be prioritised for more detailed modelling using tier 2 and tier 3 tools (Environmental Protection Agency, 2007). Examples of tier 1 tools are Source Ranking Database (SRD) and Use Cluster Scoring System (UCSS).
Tier 2 - Screening Level Tools
Tier 2 tools offer greater accuracy than tier one tools while remaining fast and easy to use. They have the capability to be much more accurate than tier 1 tools as they cover many more parameters which can be modified by the user to more accurately model a scenario. Simplicity and usability is maintained however by the provision of carefully chosen default values, which can be used as is in the absence of more accurate data from the user. These tools are designed to be conservative in that they tend to overestimate the risk. This is done purposefully to ensure risk is not underestimated and to promote confidence in ‘safe’ risk estimations (Environmental Protection Agency, 2007). Examples of tier 2 tools are Chem-Steer and E-Fast.
When great levels of accuracy are required, such as for a detailed risk assessment, higher tier tools are used. These tools are designed to be tailored by the user to model a very specific exposure scenario based on numerous measured values. As such, they are meant to be used only by persons with sound technical knowledge but have the advantage in providing comprehensive exposure estimates (Environmental Protection Agency, 2007). An example of a tier 3 tool is Wall Paint Exposure Assessment Model (WPEM)
To discuss the use of exposure assessment models, such as EASE, is assessing health risk from exposure to hazardous substances in the workplace.
The key benefit of exposure modelling is that exposure can be assessed without actually exposing persons. This is helpful where entirely new processes are being established with little historical data to work with. It is also helpful when substances are so toxic that even small exposures would result in irreversible damage to workers.
An evaluation of the COSHH model of exposure assessment was conducted by (Tischer, Bredendiek-Kamper, & Poppek, 2003). It was duly noted that while most of the information available on model validity in existence is based on environmental modelling, general concepts can be transferred to workplace exposure models such as COSHH and EASE. Different scenarios were modelled and also measured to determine the validity of the model.
Modelling of exposure to medium-high volatility liquids handled in millilitre quantities, an optical workshop, resulted in accurate predictions with all measured exposures falling within the predicted band. However when quantities of these volatile liquids were increased, in a woodwork shop, a considerable number of measured data points exceeded the predicted range.
When exposure to Volatile Organic Compounds (VOCs) was modelled in a printery, the study determined that most of the measured values were well below the upper limit of the predicted range and some were even below the lower limit. This indicates that the model was somewhat suitable for this application, although some uncertainly arises because of the measured values that fell below the predicted range.
Another benefit of modelling is that results generated by the model can be applied to the general population assuming certain rules of data collection has been followed. If for example the data collected is from persons who are statistically representative of general population, then exposure assessments can be applied generally.
One example of this generalisation was a study conducted to model the relationship between road traffic pollution and asthma (Furguson, Maheswaran, & Daly, 2004). This study looked at several model types and existing studies on road traffic pollution and asthma. In this case, a model that is shown to accurately predict exposure in one area of the country can be generalised to predict exposure in another part, bearing in mind that data input to the model would be different for different locales. In order to generalise however, additional inputs which would be likely to change will have to e incorporated in the model, for example wind speed and direction, distance from roadway and number of cars passing. Thus while the generalisations of models may not give the most accurate assessment in all locations, it is preferable to conducting monitoring in every single part of a country which is wholly impractical.
Inaccuracy in modelling stems from two factors: variability and uncertainty (Cullen & Fray, 1999). Variability occurs because no two individuals are alike. Individuals differ in physical and biological characteristics such as body mass, breathing rates, behavioural and dietary patterns and even variability of these characteristics in the same individual over time. Variability limits the usability of the results of a model by limiting the percentage of persons in a population to which the model can be applied with confidence. The effect of variability can never be completely removed but may be reduced by studying the distribution of characteristics important to the functioning of the model within the population in which it is to be used. Fortunately, characteristics such as those mentioned previously usually follow a normal distribution within any population. Thus while being beyond our control, the distribution is predictable and can be quantified for use in a model within a determinable confidence interval.
Uncertainty occurs because of the incompleteness of empirical data used as inputs in the model. It can be thought of as the result of imperfections in measurement instruments, random and systematic measurement errors or any other factor that may contribute to measured empirical values not being correct. Some uncertainties may be the amount of chemical that the worker is actually exposed to, estimates of emission rates, absorption rates, fate and transport efficiencies, dose-response relationships all lead to uncertainty in the results produced by modelling. Like variability, uncertainty cannot be removed completely. However, it can be reduced to a minimum by using instruments with the accuracy and sensitivity required, and using them in the recommended way.
Because of these sources of error, it is important to ensure that the results of modelling are accurate (International Labour Organisation, 2000). It is helpful if results of the model can be compared against the results of monitoring, at least on a small scale. Representative samples can be taken from a population to ensure that the level of exposure predicted by the model is in fact what workers are being exposed to. If such biological monitoring is not possible, at least two different models should be used to predict the same exposure. Thus accuracy of the models can be buttressed by correlation between the results of more than one model. If the results of the different models do not correlate then the accuracy of exposure assessment immediately can be brought into question.
In the study mentioned previously where the COSHH Essentials model was validated (Tischer, Bredendiek-Kamper, & Poppek, 2003), it was observed that while exposure prediction was accurate for the optical workshop, predictions were very inaccurate for woodworking shops and slightly biased to under predicting exposure in the printery. For this reason it is very important that models be validated for each use before predictions can be relied upon. Care needs to be exercised when applying modelling to a totally new scenario and systems of exposure monitoring put in place.
Despite the existence of a large number of models designed to predict exposure under various scenarios, the amount of variance and uncertainty between scenarios itself remains largely unpredictable. Because of this direct measurement remains the only method to ascertain without doubt the degree of exposure to hazardous substances in the workplace. Where direct measurement is impractical, at least two different models should be used to predict the same exposure and their results compared.
Cullen, A. C., & Fray, C. H. (1999). Probabalistic Techniques in Exposure Assessment: A Handbook for dealing with Variability and Uncertainty in Models and Inputs. Springer.
Environmental Protection Agency. (2007, February 8th). Exposure Assessment Tools and Models. Retrieved March 5th, 2007, from Environmental Protection Agency (USA) website: http://www.epa.gov/opptintr/exposure/
Furguson, E. C., Maheswaran, R., & Daly, M. (2004). Road-traffic pollution and asthma – using modelled exposure assessment for routine public health surveillance. International Journal of Health Geographics , 3 (24).
International Labour Organisation. (2000). Environmental Health Criteria 214. Genevea: ILO.
Tischer, M., Bredendiek-Kamper, S., & Poppek, U. (2003). Evaluation of the COSHH Essentials Exposure Predictive Model on the Basis of BAuA Field Studies and Existing Substance Exposure Data. Allals of Occupational Hygeine , 47, pp. 557-569.
