Are Criteria for Use of Mechanistic Data As Input to the Linearized Multistage (LMS) Model Sufficiently Well Defined? R. B. Conolly, J. S. Kimbell, and F. J. Miller, CIIT, Research Triangle Park, North Carolina 27709
The LMS model provides what is generally thought to be a conservative description of the quantitative relationship between carcinogen exposure and tumor development. Input to the LMS typically consists of exposure data (e.g. ppm in air or mg/kg/day by gavage) and the associated tumor response. When the relationship between external exposure (e.g., ppm formaldehyde in air) and internal dose (e.g., formaldehyde-derived DNA-protein cross-links [DPXI) is sublinear, use of the internal dose as input to the LMS in place of the exposure metric leads to predictions of higher risk when expressed as risk/ dose surrogate. Moreover, the more sublinear the data, the higher the risk so expressed will be. This happens because the risk based on the dose surrogate is conditional (e.g., tumor risk conditional on DPX). However, when a pharmacokinetic adjustment is applied so that the conditional risk is re-expressed in terms of an exposure metric (e.g., risk/ppm formaldehyde in air), the adjusted risk estimate may be lower than that obtained when the original exposure metric was used as input to the LMS. Starr (Risk Anal. 10:85-91, 1990) showed that the LMS predicts less risk using DPX as input in place of ppm formaldehyde. In principle, modifying the input to the LMS is also applicable to measures of early tissue response which are part of the mechanism leading ultimately to tumor development. We assumed that both mutagenicity and cytolethality-induced cell replication are parts of the mechanism by which formaldehyde is carcinogenic and that the exposure-response curve for DPX is proportional to that for formaldehyde-induced mutation. DPX data from F344 rats exposed to formaldehyde for 3 months were taken from Casanova et al. (Fundam. Appi. Toxicol., in press) and cell replication data from a 2-year formaldehyde pathogenesis study (Monticello, Ph.D. dissertation, Duke University, 1990). To analyze the tumor data described by Kerns et al. (Cancer Res. 43:4382-4392, 1983), the following inputs to the LMS were used: formaldehyde exposure (0, 2.0, 5.6, 14.3 ppm), tumors (O/ 156,0/159,2/153, 94/140), DPX (0, 1.86,6.92,35.2 pmol/mg DNA), cell replication (1, 1, 1, 4.64 [exposed/control]), DPX x cell replication (0, 1.86, 6.92, 163). Note that the latter data set is more sublinear with respect to ppm formaldehyde in air than is the DPX-only data set. For rats exposed to 0.1, 0.5 or 1.0 ppm formaldehyde, the 95% upper bound on risk was predicted by the LMS to be 1.0 x 10-4, 6.3 x 10-4 and 1.6 x 10-3, respectively, for DPX-only and 2.4 x 10-4, 1.3 x 10-3 and 2.8 x 10-3, respectively, for DPX x cell replication. This result shows that use of mechanistic data as input to the LMS does not necessarily lead to predictions of lower risk. It also points out the need for better definition of criteria for the use of mechanistic data as input to the LMS. It is not clear if the result obtained here with the DPX x cell replication data set is an artifact of the structure of the LMS model, which would differ if, for example, an alternative model were used which could more naturally incorporate data on mutation and cell replication.