Human Health Risk Assessment - Dealing with Contaminated Sites: From Theory Towards Practical Application - page 250

5.6.4 Probabilistic Human Health Risk Assessment
Since Human Health Risk Assessment can be characterized by large uncertainties,
an attractive alternative to deterministic Human Health Risk Assessment is proba-
bilistic Human Health Risk Assessment . The basic principle here is to replace point
estimates of input parameters with probability density distributions. Popular distri-
butions include uniform, log normal and triangular distributions. In performing the
calculations, called Monte Carlo calculations, values for the input parameters are
randomly extracted from these distributions. The result of the Human Health Risk
Assessment is a distribution of calculated exposures or Risk indices. Probabilistic
approaches are particularly popular for the calculation of exposure variability, for
example, for the Risk Characterisation for dibutylphthalate, as in Vermeire et al.
( 2001 ).
The big advantage of probabilistic modelling is that it shows the possible range
of exposures or Risk indices and thus provides insight into the variability and
uncertainty of the estimates.
One disadvantage of probabilistic modelling, however, is that a choice must be
made for the percentile of the protected population. Percentiles that are often chosen,
in most cases without a clear underpinning, are the 50th (the medium exposure), the
80th, 90th, or 95th percentile. Although it is often claimed that this represents a
political decision, scientists must play a role in explaining the exact meaning of and
consequences of different choices for a specific percentile.
Ane interesting criticism of probabilistic exposure modelling is that epidemiol-
ogists use probability to redistribute human anxiety; Lindley ( 2001 ) claimed that,
typically, they falsely reassure some and baselessly frighten the rest. This author
decided in favour of black and white decisions (good or bad). The only difference
between black and white decisions and probabilistic approaches is, according to
the author, that the former is decided on the antecedent conditions, while the prob-
abilistic approaches translate the unknown antecedent conditions into potentially
confusing information.
One alternative to Monte Carlo techniques is using fuzzy-stochastic modelling
approaches. In this approach, fuzzy membership functions are employed to quantify
the uncertainties and complexities. An example is given in Chen et al. ( 2003 ), who
applied this procedure to evaluate the human health risks resulting from subsurface
toluene contamination at a petroleum-contaminated groundwater system in western
Canada.
5.6.5 Reliability
In Section 1.5.4 a general exposition on reliability is given. In human exposure
modelling, the model user has to deal with many input parameters. It typically is not
efficient to determine each input parameter with the same degree of precision. In
good modelling practice, the model user pays specific attention to the identification
of the most important input parameters. The most important parameters are those