Accounting for Missing Bioassay Data in Uncertainty Analysis of Population Thresholds for Noncancer Health Effects. A. I. Shlyakhter and J. S. Evans, Dept. of Environmental Health, Harvard School of Public Health, 665 Huntington Ave., Boston, MA 02115
For assessment of noncancer health effects in humans, EPA requires a set of five animal bioassays for each chemical. Often some of the required bioassay data are not available. Here we describe how for a particular chemical one can account for missing bioassays using correlation analysis of the predictive power of different bioassays across the chemicals with available log(NOAEL) values. We assume that there are k bioassays already conducted for this chemical; each produced a log(NOAEL) value log(Nk). What is the chance that (k+1)th assay will produce a log(NOAEL) below a given value, log(N)? First we define a fictitious bioassay called "minimum" which for each chemical gives the lowest log(NOAEL). Correlation analysis shows that the rat reproductive bioassay is most closely correlated with the "minimum" and is the most predictive while the rat developmental bioassay is the least predictive. We use constrained minimization procedure which adds large penalty to the objective function whenever predicted minimum log(NOAEL) for a chemical falls above log(NOAEL) already observed in one of the bioassays. For some fraction of chemicals, our estimates of log(NOAEL) are not conservative. This involves certain health risks for people exposed. By increasing penalty we can minimize the fraction of such chemicals. On the other hand, if the penalty is too high, the predicted min log(NOAEL) values will be biased low. In this case our estimates of log(NOAEL) will be overconservative which involves economic costs of unnecessary protection. The task of balancing risks and benefits i.e. choosing the value of penalty should be left to decision makers.