**Probabilistic Characterization of Missing Data in
Noncancer Risk Assessment.** *S. J. S. Baird, 87 Rockland
Place, Newton, MA 02164; A. I. Shlyakhter, Harvard University,
Department of Physics, Cambridge, MA 02138; and J. S. Evans,
Harvard School of Public Health, Department of Environmental
Health, 665 Huntington Ave., Building I-1308, Boston, MA 02115*

We have developed a new approach for characterizing the
uncertainty in the population threshold dose (Baird *et al.,
Human and Ecological Risk Assessment*, 1996, 2:79-102) where
uncertainty factors are represented as probability distributions
instead of point estimates. Here we describe approaches for the
probabilistic characterization of the Data Quantity/Quality
factor that accounts for missing bioassay data. Noncancer risk
assessment methods rely on the No-Observed-Adverse-Effect Level
(NOAEL) of the most sensitive or ‘critical’ effect.
Thus, all possible effects need to be studied to be certain that
the true ‘critical’ effect is identified. A complete
dataset commonly includes a rat chronic, mouse chronic, rat
developmental and rat reproductive bioassays. However, frequently
a complete dataset is not available for the compound of interest.
In this case, regulatory agencies have accounted for the chance
that the missing studies would have yielded a lower NOAEL by
using an point estimate uncertainty factor, often called a Data
Quantity/Quality factor. To fully account for the uncertainty in
the population threshold, the uncertainty in how much lower the
missing NOAEL(s) might be should be characterized
probabilistically. Using the complete datasets for 35 pesticides,
we developed 3 approaches for estimating the probability
distribution of the ratio of the estimate of the missing NOAEL to
the ‘true minimum’ NOAEL. Because the missing NOAEL is
only important if it is lower than the existing NOAELs, the
estimate from each approach only replaces the existing NOAEL if
it is lower. The three complimentary approaches include a simple
method, taking the ratios of the minimum of the available
bioassays and the minimum of the complete set of bioassays, the
‘true NOAEL’; a standard regression method, regressing
the minimum of the available studies on the ‘true
NOAEL’; and a constrained regression method, where a penalty
term is added to the regression equation to decrease the
likelihood that the estimated NOAEL is greater than the available
NOAEL. The advantages and limitations of each approach will be
discussed in relation to risk assessment and risk management
goals.

Work supported by Health Canada, under Contract No. 5366.