Open Access

Research Article

Description of the Databases Top

An overview of the data is presented in Table 1. Experimental results described in detail by the U.S. EPA (2003b) were corrected in a few cases and supplemented as follows:

• We added esophageal tumors for diethylnitrosamine (DEN; Peto et al. 1984); liver but not esophageal tumors from this article were included in the U.S. EPA analysis (U.S. EPA 2003b). Additionally, we added control observations reported by Peto et al. (1991).

• The exposure time was corrected for some vinyl chloride groups; we also included additional control and comparison group information for 52-week exposures described by Maltoni et al. (1984).

• We consolidated 6,000 and 10,000 ppm exposure groups for vinyl chloride; both of these are far greater than saturating levels for the metabolic activation of this chemical. Results for control (zero-dose) groups were also consolidated in several cases.

• We added the results of a major single-dose study of N-nitrosomethylurea by Terracini et al. (1976) and data from several reports on carcinogenesis from ionizing radiation in rats and mice (Cahill et al. 1975; Castanera et al. 1971; Di Majo et al. 1990; Knowles 1985; Sasaki 1991).

• We deleted groups that did not show defined observations for controls (numbers of animals tested and numbers with tumors).

Data for two nonmutagenic chemicals (DDT and dieldrin) were eliminated from the analysis because of the complexity of the dosing protocol used. In these experiments, some groups were given gavage exposures, some direct dietary exposures, and some both in sequence. This rendered unambiguous calculations of comparable dosages for the different groups difficult.

The principal analyses maintain the subdivisions between continuous-dosing protocols (in which dosing was maintained at a given rate for a defined period) versus discrete-dosing experiments (in which only a single dose, or up to four single doses were given to the animals at defined ages).

The full databases as well as models used for the statistical analyses of continuous, discrete, and radiation dosing data are available on our website (Hattis 2004).

Modeling Methods Top

Dosimetric conversions. The assessment of comparable dosimetry for animals in different life stages has been a substantial issue in discussions of the analysis of these data. For various experiments in the original U.S. EPA listing (U.S. EPA 2003b), doses are quoted in terms of a concentration in an environmental medium (parts per million in diet or water or air to the individual for exposures after weaning, and to the mother in the case of fetal and birth-weaning exposures); in other cases, doses that were directly administered to animals via intraperitoneal or other injections were originally expressed in terms of micrograms per kilogram body weight or similar units. For entry into our analysis, we left the doses expressed in terms of environmental media concentrations unchanged, but we transformed the doses expressed as micrograms per kilogram body weight into micrograms/(kilogram body weight)^0.75 by multiplying by estimated individual body weights to the one-quarter power. [Body weights for this purpose were taken from Nomura (1976) for mice and from the NTP (1999) and Zhang et al. (2001) for rats.] The aim of this transformation was to use a dose metric that (to the extent possible with available information short of physiologically based toxicokinetic modeling) is expected to be approximately proportional to internal daily average systemic concentrations of the parent compounds or putative active metabolites for continuous dosing, or area under the concentration-time curve (AUC) for discrete dosing.

The basis for this approach is similar to the principal current basis for dosimetric conversions for interspecies projections of cancer risks: that risks are assumed to be similar across species if the internal time-integrated concentrations of active metabolites are similar across species. Similarity of internal time-integrated concentrations is assessed with the aid of observations that both bulk uptake and elimination processes tend to scale across species with metabolic rates--approximately in proportion to body weight to the three-quarters power (Boxenbaum 1982; Federal Council for Science, Engineering and Technology 1992; Travis and White 1988; Travis et al. 1990). We have recently found that a similar transformation reconciles clearance rates of drugs across age groups in humans--at least after a period of severely deficient clearance in the first few months of infancy. Table 2, documenting this result, is based on a new regression analysis of human data for pharmaceuticals and methods that have been previously described (Ginsberg et al. 2002; Hattis et al. 2003). We have not located a comparable set of in vivo clearance observations in rats or mice. The literature does contain several reports that indicate depressed liver-metabolizing activity in the neonatal period based on in vitro measurements of the activity of some liver enzymes (Basu et al. 1971; Macleod et al. 1972) and differences between the sexes in the maturation of metabolizing capabilities (with generally greater activity observed in males). To assess the possible influence of a neonatal deficit of either activating or detoxifying activity on our findings, in the "Results" we include comparative analyses of the single-dose data for apparent relative sensitivity at narrowly defined time windows--contrasting day 1 after birth with later periods before and after weaning. We performed these comparisons for the two carcinogens that are thought to be direct acting (not requiring metabolic activation) and for those that putatively need metabolic activation before directly DNA-reactive substances are generated. We also assessed differences in apparent life-stage-related sensitivity between the sexes.

For ionizing radiation exposures, we have chosen to leave the doses in units of absorbed energy--rads or grays. If the oxidative products generated by radiation are the actual carcinogenic agents, and if these are predominantly destroyed by metabolism-dependent processes that operate at rates that scale with metabolic rates, it is possible that achieving comparable integrated dose ?? time levels of the active agents might require the same (body weight)^0.75 conversions as used for chemicals. Making such a transformation would tend to decrease the time-integrated dosage for the younger postnatal animals and therefore would tend to increase the assessed sensitivity per dose relative to adult exposures. As it happens, such a transformation would have brought the radiation results more closely into alignment with the results for mutagenic chemicals.

Equation fit and statistical optimization. One basic difference between our methodology and that used for these data by the U.S. EPA (2003b) is a transformation of the raw observations of tumor incidence in different groups into the estimated number of tumor transformations per animal. This corrects for the fact that researchers cannot usually distinguish between cases where one or more than one tumor was induced in a particular organ within a specific animal (or where more than one tumor would have been induced at the site studied had the animal lived to the end of the observation period). To accomplish this, we use the same Poisson transformation that has been traditionally used for the multistage and related statistical models of carcinogenesis.

The Poisson distribution is appropriate for processes that occur as the result of independent events where the number of possible events occurring in a particular unit of observation is unlimited. Our use of the Poisson distribution in this case derives from the basic fact that tumors start in individual cells (Fialkow 1997; Knudson, 1973, 1977). Each tumor is conceived to be an independent event arising as the result of the completion of the last stage mutation in one stem cell out of many other susceptible stem cells in a particular organ. It should be noted that this last-stage event will not generally have occurred during the preadult life stages that are the focus of our analysis, but the effects of these early life exposures will manifest as incremental tumors that occur during the life-long period of observation of the animals.

Fraction of animals with tumors = P_tumor = 1 - P_{no tumor} [1]

P_{no tumor} = e^-m = exp(-m), [2]

where m is the tumor transformations per animal at the studied site. Solving for m:

m = -ln(P_{no tumor}) = -ln(1 - fraction of animals with tumors).[3]

Because most of the experiments use only a single dose of carcinogen for each age group, no more sophisticated multistage treatment of tumor dose response is possible with these data. Given this, relative cancer transformation rates in different age groups in comparison with adult animals were estimated by fitting the continuous data to the following equation:

Fraction with tumors = 1 - e^{-[B + Aa + fF + cC + wW)]}, [4]

where B is the group background transformations per animal; A is the group adult transformations per animal at the highest adult dose rate; a is the fraction of the adult period with dosing at the maximum adult rate (this term reflects an adjustment where a group received less than the full adult dosing rate); f is the fraction of the fetal period with dosing at the maximum adult rate (also adjusted for dose rate as needed); F is the fetal:adult sensitivity ratio; c is the fraction of the birth-weaning period with dosing at the maximum adult rate (also adjusted for dose rate as needed); C is the birth-weaning:adult sensitivity ratio; w is the fraction of the weaning-60-day period with dosing at the maximum adult rate (also adjusted for dose rate as needed); and W is the weaning-60-day:adult sensitivity ratio.

In Equation 4, the terms designated with lowercase letters represent the input dosing and tumor response data for each group of dosed animals or controls. Where continuous daily dosing occurred over only part of a life stage, we entered the fraction of the life stage where dosing occurred. Similarly, where dosing for a particular group occurred at a fraction of the maximal rate given to adults, that fraction was entered as input data. This model form treats contributions to ultimate cancer transformation events from different life stages as additive.

The equation has two types of estimated parameters (designated with upper case letters). First, A and B are used only within specific experiments (a particular tumor type associated with exposure to a particular chemical in a particular animal group). By contrast, the three remaining "generic" parameters (F, C, and W) are estimated based on the results of all the dose groups for all chemicals and animals included in a particular run that contained some dosing within each life stage, compared with controls. Thus, for these generic parameters, the results represent summary central estimates [and upper (UCL) and lower confidence limits (LCL)] for all chemicals, tumor types, species (rats and mice), and other characteristics of the included experimental data. In light of this, in the "Results" we present alternative sets of estimates designed to explore the influence of sex, mutagenic character, tumor site, and other characteristics on the assessments of differences in susceptibility among life stages. Finally, because the doses used in the model fitting were expressed in terms of dose ÷ (body weight)^0.75, the units of the relative sensitivity parameters should similarly be understood to be

[5]

Estimates of the uppercase terms were derived by minimizing the "deviance" between observed and model predicted data points, as described by Haas (1994) and McCullagh and Nelder (1989): For nonzero numbers of tumors in a particular group the "deviance" is

[6]

where k is the number of dose groups; N_i is the number of animals with tumors in group i; T_i is the total number of animals in group i; _i is the model-predicted proportion of animals with tumors in group i; and ⁰_i is N_i/T_i.

This deviance-minimization optimization was accomplished in Microsoft Excel spreadsheets using the "solver" facility (Microsoft Corporation, Redmond, WA). Haas (1994) also provided procedures for deriving profile-likelihood-based confidence intervals (Venzon and Moolgavkar 1988) for these fitted parameters based on the chi-square statistic. For each confidence interval estimate, all parameters other than the one being assessed were allowed to vary. Thus, the upper and lower 95% confidence limits for the birth-weaning:adult sensitivity estimates reflect possible uncertainties in all the group background transformations per animal, group adult transformations per animal, and the sensitivities of fetal and weaning-60-day life stages relative to adults. A similar approach was used for the discrete dosing data and for the combined continuous and discrete data by dividing the doses by the estimated numbers of days in each dosing period (8 days for the fetal dosing period, 21 days for the birth-weaning life stage, 39 days for the weaning-60-day life stage, and 663 days for the adult period).

Results and Discussion: Relative Sensitivity of Different Life Stages in Animals Top

Before considering the age-related differential sensitivity results for continuous versus discrete dosing in detail, it is worth noting that they may be reflecting somewhat different factors. The continuous dosing results:

• Include enzyme induction effects, if any

• Inherently reflect a dilution of any fluctuations in short-term sensitivity caused by, for example, waves of cell proliferation in specific organs in narrow time windows

• Possibly present fewer complications from high-dose kinetic and dynamic nonlinearities

• Have somewhat more straightforward implications for adaptation of traditional chronic dosing assessments.

• On the other hand, the results from experiments where dosing was administered at discrete times:

• Almost always exclude direct enzyme induction effects

• Are capable of revealing short-term sensitivity fluctuations, to the extent that these occur

• Are likely to be done at somewhat higher dose rates, with some increase in potential complications from high-dose nonlinearities

• Have more straightforward implications for assessment of risks from acute exposure events.

Table 1.

Table 2.

Table 3.

Table 4.

Table 5.

Table 6.

Table 7.

Table 8.

Table 9.

Table 10.

Table 11.

Table 12.

Table 13.

Figure 1.

Lognormal plots of likelihood-based uncertainty distributions for cancer transformations per daily dose for various life stages for mutagenic chemicals (relative to comparable exposures of adults) for combined discrete and continuous dosing experiments in females. Log(birth-weaning/adult): y = 0.646 + 0.0785x; R² = 1.000. Log(fetal/adult): y = 0.246 + 0.134x; R² = 1.000. Log(weaning-60 days/adult): y = 0.0880 + 0.124x; R² = 0.999.

Figure 2.

Lognormal plots of likelihood-based uncertainty distributions for cancer transformations per daily dose for various life stages for mutagenic chemicals (relative to comparable exposures of adults) for combined discrete and continuous dosing experiments in females. Log(birth-weaning/adult): y = 0.646 + 0.0785x; R² = 1.000. Log(fetal/adult): y = 0.246 + 0.134x; R² = 1.000. Log(weaning-60 days/adult): y = 0.0880 + 0.124x; R² = 0.999.

Results for overall continuous chemical, discrete chemical, and radiation dosing data sets. Table 3 shows the results of fitting the continuous and discrete dosing data as a whole, together with similar results for radiation exposures. In all three sets of data, the birth-weaning period is suggested to be the most sensitive per day of dosing, followed by the fetal period and the weaning-60-day period. Each independent data set yields a central estimate of the birth-weaning sensitivity that is about 5- to 10-fold greater than the sensitivity per day of dosing in adulthood, with doses expressed per body weight^0.75.

Mutagenic versus nonmutagenic chemicals. In the case of the continuous dosing data, some of the chemicals were classified by the U.S. EPA (2003b) as mutagenic, and some not. (All of the chemicals with discrete dosing data, and ionizing radiation, are mutagenic.) Table 4 shows the continuous dosing results broken out for mutagens versus nonmutagens. In contrast with the mutagens, for nonmutagenic carcinogens none of the age groups manifest significantly greater sensitivity than is seen for adults (defined as 1 in these tables). It should also be noted that separating out the nonmutagens leaves the mutagenic compounds showing significantly more birth-weaning period sensitivity than is seen for either the discrete-dosing chemical data or the radiation observations.

Male versus female animals. Tables 5^-7 show the contrast between results in male versus female animals for continuously dosed mutagens, mutagenic chemicals delivered in discrete doses, and radiation experiments, respectively. The differences appear most prominent for the continuous dosing data (Table 5), where males seem to have much larger increases in sensitivity relative to adults for the fetal and birth-weaning life stages, and by contrast, females show a large increase in sensitivity for the weaning-60-day period. Considerable reserve is in order in interpreting the latter result, however, in the light of the slender database available for the continuous dosing analysis (only 3 chemicals and 16 dose groups for each sex) and the fact that neither the larger set of discrete-dosing data (Table 6) nor the radiation-dosing data (Table 7, based on fetal and weaning-60-day stages only) exhibits a similar enhanced female relative sensitivity for the weaning-60-day period, compared with males.

One way of weighing the different observations from continuous versus discrete chemical dosing experiments is to combine the two sets of results into a single model for analysis. The results of such a combination for male and female life-stage relative sensitivity ratios are shown in Table 8. The combined data tend to reinforce the suggestion that there are male-female differences in age-related sensitivity patterns but fail to sustain the initial suggestion from the continuous dosing data of an increase in the sensitivity for females in the weaning-60-day period relative to adults. On the other hand, the combined data do indicate an increased sensitivity for this period in males. The combined data for the fetal and birth-weaning periods indicate much more prominent excess sensitivity relative to adults in males than in females.

Distributional form for the statistical uncertainties in estimated life stage/adult sensitivities. Figures 1 and 2 show lognormal probability plots (Hattis and Burmaster 1994) of the statistical uncertainty distributions for the life stage:adult sensitivity ratios for the male and female combined discrete and continuous dosing data for mutagenic carcinogens. In this type of plot, correspondence of the points to the fitted line is an indicator of the fit of a lognormal distribution to the statistical uncertainties in central estimate life stage:adult sensitivity ratios. (The Z-score that makes up the x-axis is the number of standard errors above or below the median of the normal distribution log₁₀ transformed values.) It can be seen that the uncertainty distributions are well described by the lognormal fits. We stress that these plots are of confidence limits on the aggregate central tendency results for all chemicals in the covered groups. The uncertainties in estimates for individual chemicals are being analyzed separately (Hattis et al., unpublished data), together with implications for human risk for a particular mutagenic chemical.

Rats versus mice. Table 9 shows comparative results for life-stage-specific relative tumor sensitivities in rats versus mice for the combined discrete and continuous dosing experiments. There is a suggestion that the rat data may indicate somewhat larger effects relative to adults for the fetal and weaning-60-day life stages; however, the 95% confidence limits overlap. In the light of the very limited numbers of chemicals with relevant observations for rats, there should be no strong inference that the suggested rat/mouse differences are real.

Direct-acting carcinogens versus agents requiring metabolic activation. All but two of the mutagenic carcinogens covered in the database are thought to require metabolic activation to produce DNA-reactive agents (U.S. EPA 2003). The two exceptions are the nitrosoureas--methyl- and ethylnitrosourea. Comparing life stage:adult sensitivity results for the metabolically activated versus direct-acting compounds can shed light on whether the previous results, including the relevant dosimetry, are likely to have been appreciably distorted by immaturity of metabolic activating systems in the neonatal period.

Table 10 shows the relevant comparison using our standard breakdown of life stages, based on the single-dose data. The results indicate a clear difference in fetal sensitivity for direct-acting versus metabolically activated compounds. As might have been expected, there is, if anything, less carcinogenic susceptibility in the fetal period for metabolically activated compounds, whereas the fetal life stage shows 5- to 25-fold greater sensitivity than adults for the direct-acting nitrosoureas.

Table 11 shows the results of using a finer breakdown of time periods, made possible by the focus on data resulting from direct dosing at discrete times. Beyond the fetal period, there is no apparent difference in the pattern of relative sensitivity with age between the nitrosoureas and the metabolically activated carcinogens. In both cases, relative sensitivity peaks near birth and declines progressively thereafter until it reaches about double the adult sensitivity at day 21. Beyond the fetal period, there is thus no indication of a perinatal deficit in metabolic activating activity for this set of carcinogens.

Direct dosing in the birth-weaning period versus dosing via lactation. Another important dosimetric issue is whether the lactational exposures resulting from primary dietary exposure to maternal animals are in fact equivalent to doses directly administered to pups during the birth-weaning period. Table 12 shows the results of separate estimations of the relative tumor susceptibility for direct versus lactational exposure for the combined set of continuous and discrete dosing experiments. The data show that no diminution in birth-weaning sensitivity is indicated for lactational exposures compared with direct administration of known doses. If anything, the lactational exposures appear somewhat more potent than direct administration per unit of estimated external exposure, although the 95% confidence limits overlap. One possible interpretation of this result, if repeated, is that some of the bolus doses given in the direct administration experiments may have partially saturated metabolic activation pathways, leading to less effective dose of DNA-reactive metabolites per unit exposure than when similar materials are administered more slowly via milk.

Radiation results for different times during the "adult" period. The "adult" comparison groups for the discrete chemical dosing experiments generally were exposed in early adulthood--within 4-6 months of age. By contrast, the radiation experiments include groups extending to much older ages--up to 16-18 months. As shown in Table 13, these data indicate a considerable reduction in sensitivity for radiogenic cancer induction with advancing age.

Liver tumors versus tumors in other organs. As indicated in Table 1, many of the tumors studied in these rodent experiments come from the liver, particularly for the continuous dosing studies. We have found that, in general, life-stage-specific enhancements of sensitivity seem to be greater for the liver than for the lung, but life-stage-specific excesses in sensitivity are still apparent for the aggregate of nonliver, nonlung organs (Hattis 2004).

Toward quantitative applications in human health risk assessment. On a qualitative level, this analysis provides more detailed understanding and confidence in the fact that there is an increased early-life sensitivity for mutagenic carcinogens--reinforcing the conclusions drawn by the U.S. EPA (2003b). The next step toward applying these data for quantitative human risk assessment is to develop time/age mapping between rodents and people. What ages in people approximately correspond to the rodent fetal, birth-weaning, and weaning-60-day periods studied in this analysis? We are developing a preliminary mapping based on the times at which rodents and people attain various fractions of the average body weights they have at sexual maturity (Hattis et al., unpublished data). In this second article we also use a Monte-Carlo model-based distributional analysis of the combined uncertainties in a) the central estimates of life-stage-related differences in carcinogenesis susceptibility, as derived in this article; b) the chemical-to-chemical variation in the life-stage-related susceptibility estimates; and c) the rodent/human time mapping uncertainty. Quantitative assessment of these three uncertainties together is needed for full distributional analyses of cancer risks for exposures in early life stages.

References Top

1. Basu TK, Dickerson JWT, Parke DVW. 1971. Effect of development on the activity of microsomal drug-metabolizing enzymes in rat liver. Biochem J 124:19–24. Find this article online
2. Boxenbaum H.. 1982. Interspecies scaling, allometry, physiological time, and the ground plan of pharmacokinetics. J Pharmacokinet Biopharm 10(2):201–227. Find this article online
3. Cahill DF, Wright JF, Godbold JH, Ward JM, Laskey JW, Tompkins EA. 1975. Neoplastic and life-span effects of chronic exposure to tritium. II. Rats exposed in utero. J Natl Cancer Inst 55:1165–1169. Find this article online
4. Castanera TJ, Jones DC, Kimeldorf DJ, Rosen VJ. 1971. The effect of age at exposure to a sublethal dose of fast neutrons on tumorigenesis in the male rat. Cancer Res 31:1543–1549. Find this article online
5. Di Majo V, Coppola M, Rebessi S, Covelli V.. 1990. Age-related susceptibility of mouse liver to induction of tumors by neutrons. Radiat Res 124:227–234. Find this article online
6. Federal Coordinating Council for Science, Engineering and Technology. 1992. Draft report: a cross-species scaling factor for carcinogen risk assessment based on equivalence of mg/kg^3/4/day. Fed Reg 57:24152–24173. Find this article online
7. Fialkow PJ. 1977. Clonal origin and stem cell evolution of human tumors. In: Genetics of Human Cancer (Mulvihill JJ, Miller RW, Fraumeni JF, eds). New York:Raven Press, 439-452.
8. Ginsberg G, Hattis D, Sonawane B, Russ A, Banati P, Kozlak M, et al. 2002. Evaluation of child/adult pharmacokinetic differences from a database derived from the therapeutic drug literature. Toxicol Sci 66:185–200. Find this article online
9. Haas CN. 1994. Dose response analysis using spreadsheets. Risk Anal 14:1097–1100. Find this article online
10. Hattis D. 2004. Human Interindividual Variability Homepage. Available: http://www2.clarku.edu/faculty/dhattis [accessed 15 June 2004].
11. Hattis D, Anderson E.. 1999. What should be the implications of uncertainty, variability, and inherent "biases"/"conservatism" for risk management decision making? Risk Anal 19:95–107. Find this article online
12. Hattis D, Burmaster DE. 1994. Assessment of variability and uncertainty distributions for practical risk analyses. Risk Anal 14:713–730. Find this article online
13. Hattis D, Ginsberg G, Sonawane B, Smolenski S, Russ A, Kozlak M, et al. 2003. Differences in pharmacokinetics between children and adults--II. Children's variability in drug elimination half-lives and in some parameters needed for physiologically-based pharmacokinetic modeling. Risk Anal 23:117–142. Find this article online
14. Knowles JF. 1985. Changing sensitivity of neonatal rats to tumorigenic effects of N-nitroso-N-ethylurea and X-radiation, given singly or combined. J Natl Cancer Inst 74:853–857. Find this article online
15. Knudson AG. 1973. Mutation and human cancer. Adv Cancer Res 17:317–352. Find this article online
16. Knudson AG. 1977. Genetics and etiology of human cancer. Adv Hum Genet 8:1–66. Find this article online
17. MacLeod SM, Renton KW, Eade NR. 1972. Development of hepatic microsomal drug-oxidizing enzymes in immature male and female rats. J Pharmacol Exp Ther 183:489–498. Find this article online
18. Maltoni C, Lefemine G, Ciliberti A, Cotti G, Carretti D. 1984. Experimental research on vinyl chloride carcinogenesis. In: Archives of Research on Industrial Carcinogenesis, Vol 2 (Maltoni C, Mehlman MA, eds). Princeton, NJ:Princeton Scientific Publishers, 1-533.
19. McCullagh P, Nelder JA. 1989. General Linear Models. 2nd ed. London:Chapman and Hall.
20. National Research Council. 2002. Estimating the Public Health Benefits of Proposed Air Pollution Regulations. Washington, DC:National Academy Press.
21. Nomura T.. 1976. Comparison of tumour susceptibility among various organs of foetal, young and adult ICR/Jcl mice. Br J Cancer 33:521–533. Find this article online
22. NTP. 1993. Toxicology and Carcinogenesis Studies of 1,3-Butadiene (CAS No. 77-47-4) in B6C3F₁ Mice (Inhalation Studies). TR-434. Research Triangle Park, NC:National Toxicology Program.
23. NTP. 1999. Toxicology and Carcinogenesis Studies of Pentachlorophenol (CAS No. 87-86-5) in F344/N Rats (Feed Studies). TR-483. Research Triangle Park, NC:National Toxicology Program.
24. OMB. 2000. Guidelines to Standardize Measures of Costs and Benefits and the Format of Accounting Statements. Memorandum from Jacob J. Lew, 22 March 2000. Washington, DC:Office of Management and Budget. Available: http://www.whitehouse.gov/omb/memoranda/​m00-08.pdf [accessed 11 May 2004].
25. Peto R, Gray R, Brantom P, Grasso P.. 1984. Nitrosamine carcinogenesis in 5120 rodents: chronic administration of sixteen different concentrations of NDEA, NDMA, NPYR and NPIP in the water of 4440 inbred rats, with parallel studies on NDEA alone of the effect of age of starting (3, 6 or 20 weeks) and of species (rats, mice, or hamsters). IARC Sci Publ 57:627–665. Find this article online
26. Peto R, Gray R, Brantom P, Grasso P.. 1991. Effects on 4080 rats of chronic ingestion of N-nitrosodiethylamine or N-nitrosodimethylamine: a detailed dose-response study. Cancer Res 51:6415–6451. Find this article online
27. Sasaki S.. 1991. Influence of the age of mice at exposure to radiation on life-shortening and carcinogenesis. J Radiat Res 32: suppl 273–85. Find this article online
28. Terracini B, Testa MC, Cabral JB, Rossi L. 1976. The roles of age at treatment and dose in carcinogenesis in C3HfDp mice with a single administration of N-nitroso-N-methylurea. Br J Cancer 33:427–439. Find this article online
29. Travis CC, White RK. 1988. Interspecific scaling of toxicity data. Risk Anal 8:119–125. Find this article online
30. Travis CC, White RK, Ward RC. 1990. Interspecies extrapolation of pharmacokinetics. J Theor Biol 142:285–304. Find this article online
31. U.S. EPA. 2001. Risk Assessment Guidance for Superfund. Vol III, Part A. Process for Conducting Probabilistic Risk Assessment. EPA/540-R-02-002. Washington, DC:U.S. Environmental Protection Agency.
32. U.S. EPA. 2003a. Draft Final Guidelines for Carcinogen Risk Assessment (External Review Draft). Washington, DC:U.S. Environmental Protection Agency, Risk Assessment Forum. Available: http://cfpub.epa.gov/ncea/cfm/recordispl​ay.cfm?deid=55445 [accessed 15 June 2004].
33. U.S. EPA. 2003b. Supplemental Guidance for Assessing Cancer Susceptibility from Early-Life Exposure to Carcinogens (External Review Draft). Washington, DC:U.S. Environmental Protection Agency, Risk Assessment Forum. Available: http://cfpub.epa.gov/ncea/cfm/ recordisplay.cfm?deid=55446 [accessed 11 May 2004].
34. Venzon DJ, Moolgavkar SH. 1988. A method for computing profile-likelihood-based confidence intervals. Appl Stat 37:87–94. Find this article online
35. Zhang S, Leske DA, Lanier WL, Holmes JM. 2001. Postnatal growth retardation exacerbates acidosis-induced retinopathy in the neonatal rat. Curr Eye Res 22(2):133–139. Find this article online