Policy-related decisions that are based, even in part, on animal experimental data depend strongly on the statistical modelling assumptions made in analyzing those data. This is true both for operating decisions about study design, and for substantive decisions about health effects. The data now available in the NTP (National Toxicology Program) databases are a rich source for study of the performance and interpretation of long-term carcinogen bioassays. The key point in this proposal is that single experiments may provide only weak data on some matters of great interest, but that correct statistical analysis can combine large numbers of such weak bits of data to produce usable conclusions. Our goal is to examine some common assumptions by using the data amassed in the course of the NTP carcinogen bioassays, and learn whether and how the assessment of carcinogenic risk may be improved. There has been no systematic study of these data as a whole from this point of view, and very few that examine even one or another specific point (e.g. background-rates; age-power laws), although the results of individual studies are scrutinized closely by NTP itself. Specific modelling assumptions to be examined are: l.The age dependence of primary tumor rates, and how this varies with dose and other parameters. 2.Non-tumor disease end-points as indicators of carcinogenicity or tumor lethality, or as end-points in their own right. 3.The adequacy or necessity of two year bioassays, and the information gained by using age at death. 4.The use of data from sacrificed animals. None of these, taken alone, is new. What is new is the systematic approach to their study.
