A Statistical Comparison of Test Protocols Used in the Assessment of Combustion Product Toxicity

Abstract

Previously we reported a criterion for choosing an appropriate statistical model to describe the tolerance distribution for estimation of the median effec tive dose [1]. In the present paper we have shown how this statistical approach can be used to assess the precision of toxicity test data and the predictive abili ties of test methods. Specifically, we conducted statistical analyses of data generated by using the University of Pittsburgh (UP) and the National Bureau of Standards (NBS) combustion toxicity protocols. These evaluations revealed that 30% of the LD50 values from the UP test were not valid while with the NBS test 26% and 5% of the LD50 values under flaming and non-flaming conditions, respectively, were not valid. Comparisons of the two test methods also showed that there was not agreement on ranking across protocols except for the very few with extreme toxicities. These results indicated that caution should be exercised in the review and interpretation of combustion toxicity data and that these tests are probably only useful in identifying consistently super-toxic chemicals.