Body Weight-Tumor Incidence Correlations in Long-Term Rodent Carcinogenicity Studies*

Transcription

1 Body Weight-Tumor Incidence Correlations in Long-Term Rodent Carcinogenicity Studies* JOSEPH K. HASEMAN, 1 ERIKA YOUNG, 1 SCOTTIE L. EUSTIS, 1 AND J. RICK HAILEY 2 1Statistics and Biomathematics Branch and 2Pathology Branch, National Institute of Environmental Health Sciences, P.O. Box 12233, Research Triangle Park, North Carolina ABSTRACT Associations between body weight and tumor incidence and among the incidences of selected site-specific tumors were examined for more than 4,000 male and female Fischer-344 rats and B6C3F1 mice in the National Toxicology Program historical control database. Incidences of certain site-specific tumors, most notably mammary gland and pituitary gland tumors in rats and liver tumors in mice, were shown to have a strong positive correlation with 52-wk body weight. Using individual animal data, logistic regression models were derived for predicting site-specific tumor incidence as a function of 52-wk body weight, age, and other factors. This association between body weight and tumor incidence can explain many of the decreased tumor incidences observed in National Toxicology Program carcinogenicity studies. Body weight differences between dosed and control groups can also mask carcinogenic effects for those sites sensitive to body weight changes. Thus, when designing long-term rodent carcinogenicity studies, measures should be taken to minimize potential body weight differences between dosed and control groups. There were a number of small but significant negative correlations among tumor incidences, reflecting primarily the lethality of the tumors in question. None of these correlations (nor the 2 small positive correlations found) are likely to have any impact on the interpretation of experimental results. However, the high negative correlation between pituitary gland and testis tumors in male Fischer-344 rats cannot be dismissed so easily, does not reflect tumor lethality, and is currently being studied further. Keywords. Tumor associations; historical control data; Fischer-344 rats; B6C3F1 mice; pituitary gland tumors; interstitial cell testis tumors; logistic regression models; National Toxicology Program INTRODUCTION In the evaluation of rodent carcinogenicity studies, significant correlations among the key variables of interest may have an impact on how the data are analyzed and interpreted. Perhaps the most important of these variables are site-specific tumor incidence and body weight. A number of investigators have reported an association between site-specific tumor incidence and body weight in Fischer-344 (F-344) rats and B6C3F1 mice, the animal models most often used by the National Toxicology Program (NTP) in its long-term carcinogenicity bioassays (7, 9, 13, 21, 24, 26). Seilkop (23) examined individual animal survival, body weight, and tumor incidence data from 55 NTP mouse studies and 53 NTP rat studies and tumor in- used a logistic model (1) to relate site-specific cidence to age at death and body weight at various points in time. In contrast, relatively little has been written on the correlation among tumor types. These correlations are important also because many of the statistical procedures used in the evaluation of rodent carcinogenicity data implicitly assume that tumors occur independently in distinct tissues (11, 25), whereas in practice this may not always be the case (3, 15). If 2 tumor types show a strong, naturally occurring positive correlation, this might provide clues as to possible mechanisms of action for a * Address correspondence to: Dr. Joseph K. Haseman, Statistics and Biomathematics Branch, National Institute of Environmental Health Sciences, PO. Box 12233, Research Triangle Park, North Carolina chemical found to be carcinogenic at one or both of these &dquo;correlated&dquo; tumor sites. Knowledge of such associations may also have an impact on how the data are evaluated statistically. It is more likely, however, that 2 tumor types will be negatively correlated (4, 17), reflecting high tumor lethality. That is, if both tumor types are rapidly lethal, then either tumor may cause the death of the animal before the other tumor has had a chance to develop, resulting in a negative association between tumor types. If appropriate survival-adjusted statistical methodology is carried out, then negative correlations resulting solely from high tumor lethality are unlikely to impact the evaluation of long-term rodent studies. In any case, the NTP database provides a unique opportunity to examine the correlations among a variety of tumor types in control animals. METHODS The data used in our evaluation consisted of all NTP 2-yr rodent carcinogenicity studies on the Toxicology Database Management System (TDMS) as of January 1, There were 84 such studies on F-344 rats and 90 studies on B6C3F1 mice. Control data from 4,295 male rats, 4,271 female rats, 4,607 male mice, and 4,581 female mice were evaluated. The primary data consisted of individual animal body weights at 3 time points (6 wk, 6 mo, and 12 mo), age at death, and tumor incidence data for 8 of the most frequently occurring tumor types in each sex-species group (10). The tumor incidences at a given site reflect the combined rate of benign and malignant neoplasms (e.g., liver adenoma or carcinoma; subcuta- 256

2 257 TABLE I.-Fifty-two-wk body weight categories used in moving average calculations summarized in Figs The number of animals per weight range category typically varied from 200 to 600; the total numbers of weight range categories were 82, 82, 83, and 113 for male rats, female rats, male mice, and female mice, respectively. neous tissue fibroma, fibrosarcoma, or sarcoma; mammary gland fibroadenoma, adenoma, or adenocarcinoma). To evaluate the association between tumor incidence and body weight, a moving average method was used to calculate tumor incidence for animals falling into selected body weight range categories. The weight ranges used are summarized in Table I. Using a moving average approach (with each weight range generally containing animals) allowed for a &dquo;smoothing&dquo; of the data and a simple graphical representation of observed relationship between tumor incidence and body weight. Tumor incidence, however, may be influenced by variables other than body weight. For example, age of the animal, type of control group (e.g., untreated, corn oil gavage, skin paint, inhalation chamber controls), test laboratory, and housing protocol may also be important factors for some tumors. Extending the work of Seilkop (23) (and drawing upon a much larger database), logistic regression models (1) were derived using individual animal data to predict site-specific tumor incidence as a function of body weight and other factors. To evaluate the correlation between 2 site-specific tumors, data were divided into 4 time categories: 0-12 mo on study, mo, mo, and 24 mo (the last category being essentially the animals sacrificed at the end of the study). The presence or absence of each of the 2 tumor types being compared was then determined for each individual animal, and Mantel-Haenszel tests (16) were used to pool the 2-X-2 tables over the 4 time categories. Tumor correlations were evaluated for the 4 most frequently occurring tumors in each sex-species group as well as any tumor having a background rate of at least 10%. ---,....,~., FIG. Association between mammary gland tumor incidence and 52-wk body weight in female F-344 rats. tumor types, there was some evidence of an association, but the association was much weaker than for those tumors noted above. For mammary gland tumors in female rats and liver tumors in mice, Figures 1, 4, and 5 show a steady increase in tumor incidence as the 12-mo body weight increases. The range of tumor incidences for differing 12-mo body weights, especially for mice, is truly remarkable. For pituitary tumors, the association between tumor incidence and body weight is still evident (Figs. 2 and 3) but is not as striking as the correlation seen for liver tumors in mice or for mammary gland tumors in female rats. A slight &dquo;peak&dquo; in pituitary gland tumor incidence was observed in lower body weight-range males and females (Figs. 2 and 3), but the reason for this marginal increase is unknown. can affect tu- at 12 To illustrate how strongly body weight mor incidence, the 50 heaviest (averaging 62.6 g RESULTS Correlations Between Tumor Incidence and Body Weight Our evaluation confirmed Seilkop s (23) finding of a strong positive correlation between tumor incidence and 12-mo body weight for mammary gland tumors in female F-344 rats (Fig. 1), pituitary gland tumors in F-344 rats (Figs. 2 and 3), and liver tumors in B6C3F1 mice (Figs. 4 and 5). There was also a weaker but still significant (p < 0.001) correlation between 12-mo body weights and lung tumor incidence in control male B6C3F1 mice (data not shown). Using 6-wk or 6-mo body weights reduced, but did not eliminate, these associations. For certain other ---I..-~ ,-..-~- ~I FIG. 2.-Association between pituitary gland tumor incidence and 52-wk body weight in male F-344 rats.

3 ~ FIG. 3.-Association between pituitary gland 52-wk body weight in female F-344 rats. tumor incidence and FIG. 5.-Association between liver tumor incidence and 52-wk body weight in female B6C3F1 mice. mo) female B6C3F1 mice in the NTP database had a liver tumor incidence of 78% (39/50). In contrast, the 50 lightest female mice (averaging 25.6 g at 12 mo) had a liver tumor incidence of only 4% (2/50). Mean survival was slightly higher (686 vs 655 days) in the heavier animals, but this difference could not explain the striking difference in liver tumor incidence observed in these 2 groups. An analogous comparison for male B6C3Fl mice showed a similar difference in liver tumor incidence (72 vs 22%), with mean survival time being similar in the heaviest and lightest groups (687 and 699 days, respectively). In addition to body weight, several other factors were found to influence tumor incidence. For example, Table II summarizes the incidences of liver tumors in control male and female mice by route of administration. Clearly, control mice from skin paint and drinking water studies have a much higher incidence of liver tumors than do other mouse controls. While body weight could explain some of this difference (see Table II), both factors (body weight and route of administration) were found to be significant (p < 0.001) in the logistic regression model for mouse liver tumors. As noted by Seilkop (23), mouse liver tumor incidence is also influenced by the method of housing used in the study (group vs individual). For example, in the NTP database, group-housed control female mice averaged 9.3% liver tumors, while the corresponding mean liver tumor rate in individually housed control female mice was 29.1 %. A similar difference was observed for male was also found to be mice. Thus, the method of housing a significant (p < 0.001) term in the logistic regression model. Similarly, for male rats, the incidence of pituitary gland neoplasms was much higher in individually housed (55.4%) than in group-housed animals (30.8%). This strong association was not seen in female rats. Table III gives the coefficients of the best fitting logistic regression models for those tumors showing the strongest correlation with body weight. These parameter estimates derived from the NTP database enable an investigator to predict the tumor incidence in control animals of a particular age and body weight. To illustrate the procedure, suppose we have an individually housed, untreated control female mouse from a feeding study who weighed 35 g at 12 mo and who died after 680 days on study. Using the parameter estimates in Table III, we TABLE II.-Incidence of liver tumors (%) and mean 12-mo body weights in control B6C3F1 mice by route of administration. FIG. 4.-Association between liver tumor incidence and 52-wk body weight in male B6C3F1 mice.

4 - 259 TABLE III.-Logistic regression model parameter estimates.a Abbreviations: MM = male mice; FM = female mice; MR = male rats; FR = female rats. a Model: Prob(tumor) 1/[l = + exp(1)], where T = a. + a, (weight) + az (weight)2 + a3 (age) + a4 (house) + a5 (route). Weight weight (g) = = at 12 mo; age length of time on study (days) (generally, age of animal wk); house 0 if group-housed = or 1 if individually housed; route 0 if chemical = was administered by skin or water or 1 otherwise. would estimate the probability tumor as follows: that this animal has a liver T = (35) (680) (1) (1) = , Prob (tumor) = 1/[1 + exp(1.8146)] = 14.0%. This value also agrees closely with the empirical data summarized in Fig. 5. Using the logistic regression models summarized in Table III, predicted tumor incidence probabilities were calculated for each animal in the NTP control database. Observed and predicted tumor rates were then summed to obtain observed and over animals within each study predicted tumor rates for each control group. Figures 6 and 7 compare observed and predicted mouse liver tumor incidences. These scatterplots confirm that the models summarized in Table III can explain much of the studyto-study variability observed in tumor rates for these particular neoplasms. Factors that may be contributing to the unexplained variability include lab-to-lab variability and time-related trends. The parameter estimates given in Table III can also be used to evaluate tumor trends in dosed animals. For example, Haseman and Johnson (10) showed that many of the observed dose-related decreases in tumor incidence observed in NTP carcinogenicity studies can be ex- plained by the lower body weights in the dosed animals. Consider the 3 decreases in mammary gland tumor incidence summarized in Table IV. These observed tumor rates are compared with the rates that would be predicted in control animals of equivalent size and mean survival. Two approaches are used to determine predicted tumor rates: (a) the logistic regression model summarized in Table III and (b) the observed tumor response of animals of equivalent mean body weight, based on the weight gain categories of Table I (summarized in Fig. 1). Note that these 2 methods produce very similar predicted values. The decreasing trends in mammary gland tumor incidence for chlorendic acid and monuron appear to be due solely to the effects of reduced body weight in the dosed animals; that is, the tumor rates observed in these dosed groups are similar to what would be expected in control animals of equivalent size. For codeine, however, body weight differences cannot totally explain the reduction in mammary gland tumor incidence observed (Table IV). For a further discussion of this decrease in tumor incidence, see Dunnick et al (2). Correlations Betweerc Site-Specific Tumors Although most correlations between site-specific tumor types were not statistically significant, there were 8 significant (p < 0.01) negative correlations and 2 significant &dquo;i., PIG. o.&horbar;comparison ot observed and predicted liver tumor rates torr 70 studies in female B6C3Fl mice. FIG. 7.--Comparison of observed and predicted 70 studies in male B6C3Fl mice. liver tumor rates for

5 260 TABLE IV.-Observed and predicted mammary gland tumor incidence in female F-344 rats. TABLE V.-Summary of significant (p < the NTP control database. 0.01) tumor associations in Excluding all animals dying prior to 52 wk. b Based on the logistic regression model (Table III). I Based on the observed tumor rate in animals of equivalent mean body weight (using the body weight range categories of Table I). Abbreviations: FR = female rats; MR = male rats; MM = male mice; FM = female mice. a Measure of association defined as WZlXY, where W = probability of Tumor A given the presence of Tumor B, X = probability of no Tumor A given the presence of Tumor B, Y = probability of Tumor A given the absence of Tumor B, and Z = probability of no Tumor A given the absence of Tumor B. positive correlations (Table V). The strongest associations were the negative correlations between subcutaneous tissue tumors and liver tumors and between subcutaneous tissue tumors and malignant lymphoma in male mice. These correlations are examined in more detail in Table VI. Several features of the data suggest that these correlations are due primarily to the relatively high lethality of the 2 tumor types, which would lessen the chance of in the same animal. the 2 tumor types occurring together First, the tumor rates are higher in those animals dying during the last 6 mo of the study when compared to those animals surviving until the end of the study (Table VI). If the tumors were nonlethal, we would have expected to see a steadily increasing tumor prevalence as the animals aged. Second, the significant correlation is due almost entirely to those animals dying early in the study. For example, among control male mice that died with subcutaneous tissue tumors during the second year of the 2-yr study, the proportion of animals having malignant lymphoma was only 2.5% (4/159; see Table VI). However, TABLE VI.-Strongest negative correlations found among tumor types in the NTP control database. Abbreviatons: sub subcutaneous tissue = tumors; liver liver = = tumors; lymph a From 20 recent NTP dermal/inhalation studies. malignant lymphoma; pituitary pituitary gland = tumors; testis testis = tumors.

6 261 among the male mice that died without subcutaneous tissue tumors during this same time period, the proportion of animals having malignant lymphoma was much greater : 13.3% (140/1,053; see Table VI). A corresponding difference was not seen in animals surviving until the end of the study [6.3% (9/142) vs 8.3%, (254/3,053); see Table VI). The most likely explanation for this result is that both subcutaneous tissue tumors and malignant lymphoma are moderately lethal neoplasms in male B6C3F1 mice; thus, either tumor, when present, tends to kill the animal before the other tumor can develop. Malignant lymphoma is well recognized to be a relatively lethal neoplasm in B6C3F1 mice, whereas subcutaneous tissue tumors frequently arise as a direct result of fighting among grouphoused male mice. This fighting often results in repeated skin wounding with its associated dermal inflammation and fibrosis and may lead to the early sacrifice of the affected animal (9). An alternative explanation for the negative correlation is that 1 of the 2 tumor types is lethal, the other is not, and the lethal neoplasm selectively occurs in animals resistant to developing the nonlethal tumor. This nonrandom pattern of tumor occurrence is unlikely but possible. For a further discussion of the estimated lethality of commonly occurring tumors in F-344 rats and B6C3F1 mice, see Portier et al (18). The only statistically significant positive correlations observed were between adrenal gland pheochromocytoma and mononuclear cell leukemia in male rats and between mammary gland and pituitary gland neoplasms in female rats (see Table V). These associations were found primarily in those animals surviving until the final sacrifice. Although interstitial cell tumors of the testis in male F-344 rats were not part of our primary statistical analysis because of the high spontaneous incidence of this neoplasm [approximately 90%; see Haseman (7) and Haseman et al (9) and Table VII], many recent NTP dermal and inhalation studies have shown a surprisingly low incidence of this tumor. Further investigation revealed that these decreases were invariably associated with elevated rates of pituitary gland neoplasms. This negative correlation cannot be due to tumor lethality, because interstitial cell tumors of the testis are nonlethal neoplasms in male F-344 rats, and the incidence of this tumor steadily increases as the animals age (Table VI). Tables VI and VII summarize this association for 20 recent NTP dermal and inhalation studies. The most striking example of this correlation was the diethylphthalate study, in which the incidence of interstitial cell tumors of the testis was only 8%, whereas the incidence of pituitary gland tumors was 89% (Table VII). Evaluation of individual animal tumor data confirmed the high negative association indicated by the overall study tumor rates. The odds ratio associated with the negative correlation for the 20 NTP dermal and inhalation studies is only 0.244, much lower than the odds ratio for any other tumor correlation in the database (see Table V). One factor that may have contributed to this correlation is the difference in housing protocols: both inhalation and TABLE VII.-Incidence of pituitary gland tumors and testis interstitial cell tumors in control male F-344 rats in 20 recent NTP dermal and inhalation studies. a Control range (45 studies): pituitary tumors 14-60%; testis = tumors 74- = 98%. dermal studies use individual housing, whereas most other types of studies use group housing. As noted previously, the housing protocol used in the NTP studies had a significant impact on pituitary gland tumor incidence in control male F-344 rats (see Table III). DISCUSSION With the exception of the negative correlation between pituitary gland neoplasms and interstitial cell tumors of the testis in male F-344 rats (an association that is currently being studied further), the magnitude of the correlations among tumor types is too small to have an appreciable effect on the statistical procedures used to evaluate rodent carcinogenicity studies. The slight negative correlations between certain site-specific tumors appear to be reflecting primarily the lethality of the tumors in question, which can be taken into account by appropriate survival-adjusted statistical analysis. Two small but significant positive correlations were also found (Table V). The first was a correlation between mammary gland and pituitary gland neoplasms in female rats, an association that was significant only in the animals sacrificed at the end of the study. Because both of these neoplasms are correlated with body weight (10, 21, 23), the data were further stratified by 12-mo body weight, and the tumor correlations reexamined. This additional stratification weakened the association between these 2 tumors slightly, but it remained highly significant (p < 0.001). Another possible explanation for the positive correlation between pituitary gland and mammary gland tumors

7 262 is that prolactin-secreting cells are reported to be the dominant cell type in most spontaneous pituitary neoplasms of F-344 rats (22). It is therefore plausible that some pituitary tumors may produce increased levels of prolactin, which stimulates growth of mammary tissue. The factors responsible for the small positive correlation between adrenal pheochromocytoma and leukemia in male F-344 rats are unknown. Neither of these correlations appear to have much impact on NTP rodent carcinogenicity studies. For example, in 379 NTP studies, no chemical has been declared carcinogenic in female rats based on both mammary gland and pituitary gland tumors, and only a single chemical (2-mercaptobenzothiazole) has produced both leukemia and adrenal pheochromocytoma in male F-344 rats (11). If the 2 positive tumor correlations noted above had impacted study outcomes, we would have expected to see more chemicals carcinogenic at both of these &dquo;correlated&dquo; tumor sites. In fact, even in the 2-mercaptobenzothiazole study, there was no observed association between the occurrence of leukemia and adrenal pheochromocytoma in individual dosed male rats. Of possibly more concern is the high correlation found between body weight and the incidence of certain sitespecific tumors. These correlations are important because reduced body weights in dosed animals may mask the detection of carcinogenic effects at those sites sensitive to body weight changes. For example, in the recent NTP dietary restriction study, ad libitum-fed mice receiving salicylazosulfapyridine by gavage showed a striking increase in liver tumors, an effect that was not present in lighter, food-restricted animals receiving the same dose of the chemical (7, 13). Thus, when designing long-term rodent carcinogenicity studies, measures should be taken to minimize potential body weight differences among groups. Dietary restriction, achieved by limiting the amount and/or availability of food, is one possible solution to this problem (5, 6, 14). A related strategy, sometimes called dietary control, is to establish a targeted idealized growth curve and adjust feed consumption periodically as necessary to achieve animals within this acceptable weight range (5). There are other approaches, however, to deal with this problem. For example, the diets generally used in longterm studies are designed to optimize the growth of animals for breeding purposes and are not ideal for longterm use (19). A less rich maintenance diet might be formulated that would help control body weights (20). Furthermore, because heavier animals are more prolific breeders, it is likely that selection for heavier animals over a number of generations has resulted in heavier animals being supplied for long-term studies. Measures could be taken to reverse this trend (19). Despite our best efforts, however, certain long-term rodent carcinogenicity studies may result in body weight differences between dosed and control groups. When on the in- such differences are large, they could impact terpretation of study results for those tumor types showing a strong correlation with body weight. For further discussion of this matter, see Hart et al [5], Hart and Turturro [6], Haseman and Johnson [10], Kari and Abdo [13], and Seilkop [23]. ACKNOWLEDGMENTS The authors are grateful to Drs. Clarice Weinberg, Gregg Dinse, and Joel Mahler for their helpful comments during the preparation of this paper. REFERENCES 1. Dinse, GE and Lagakos SW (1983). Regression analysis of tumor prevalence data. Appl. Stat. 32: (corrected data tables in Appl. Stat. 33: 79-80). 2. Dunnick JK, Elwell MR, and Haseman JK (1996). Decreased incidence of spontaneous mammary gland neoplasms in female F344 rats treated with amphetamine, methylphenidate, or codeine. Cancer Lett. 102: Fears TR, Tarone RE, and Chu KC (1977). False-positive and falsenegative rates for carcinogenicity screens. Cancer Res. 37: Gart JJ, Krewski D, Lee PN, Tarone RE, and Wahrendorf J (1986). Statistical Methods in Cancer Research, Vol. III: The Design and Analysis of Long-Term Animal Experiments. International Agency for Research on Cancer, Lyon, pp Hart RW, Keenan K, Turturro A, Abdo KM, Leakey J, and Lyn- Cook B (1995). Caloric restriction and toxicity. Fund. Appl. Toxicol. 25: Hart R and Turturro A (1995). Dietary restriction: An update. In: Dietary Restriction: Implications for the Design and Interpretation of Toxicity and Carcinogenicity Studies, R Hart, D Neumann, and R Robertson (eds). ILSI Press, Washington, D.C., pp Haseman JK (1995). Statistical considerations in long term dietary restriction studies. In: Dietary Restriction: Implications for the Design and Interpretation of Toxicity and Carcinogenicity Studies, R Hart, D Neumann, and R Robertson (eds). ILSI Press, Washington D.C., pp Haseman JK, Arnold J, and Eustis SL (1990). Tumor incidences in Fischer 344 rats: NTP historical data. In: Pathology of the F344 Rat, GA Boorman, SL Eustis, C Montgomery, and M Elwell (eds). Academic Press, San Diego, pp Haseman JK, Bourbina J, and Eustis S (1994). The effect of individual housing and other experimental design factors on tumor incidence in B6C3F1 mice. Fund. Appl. Toxicol. 23: Haseman JK and Johnson FM (1995). Analysis of rodent NTP bioassay data for anticarcinogenic effects. Mut. Res. 350: Heyse JF and Rom D (1988). Adjusting for multiplicity of statistical tests in the analysis of carcinogenicity studies. Biom. J. 8: Huff J, Cirvello J, Haseman J, and Bucher J (1991). Chemicals associated with site-specific neoplasia in 1394 long-term carcinogenesis experiments in laboratory rodents. Environ. Health Perspec. 93: Kari F and Abdo K (1995). The sensitivity of the NTP bioassay for carcinogen hazard evaluation can be modulated by dietary restriction. In: Dietary Restriction: Implications for the Design and Interpretation of Toxicity and Carcinogenicity Studies, R Hart, D Neumann, and R Robertson (eds). ILSI Press, Washington, D.C., pp Keenan KP, Smith PF, Hertzog P, Soper K, Ballam GC, and Clark L (1994). The effects of overfeeding and dietary restriction on Sprague-Dawley rat survival and early pathology biomarkers of aging. Toxicol. Pathol. 22: Lu Y and Malani HM (1995). Analysis of animal carcinogenicity experiments with multiple tumor types. Biometrics 51: Mantel N and Haenszel W (1959). Statistical aspects of the analysis of data from retrospective studies of disease. J. Natl. Cancer Inst. 22: NTP Ad Hoc Panel (1984). Report of the NTP Ad Hoc Panel on Chemical Carcinogenesis Testing and Evaluation. NTP Board of Scientific Counselors, pp

8 Portier CJ, Hedges JC, and Hoel DG (1986). Age-specific models of mortality and tumor onset for historical control animals in the National Toxicology Program s carcinogenicity experiments. Cancer Res. 46: Rao GN (1995). Husbandry procedures other than dietary restriction for lowering body weight and tumor/disease rates in Fischer 344 rats. In: Dietary Restriction: Implications for the Design and Interpretation of Toxicity and Carcinogenicity Studies, R Hart, D Neumann, and R Robertson (eds). ILSI Press, Washington, D.C., pp Rao GN (1996). New diet (NTP-2000) for rats in the National Toxicology Program toxicity and carcinogenicity studies. Fund. Appl. Toxicol. 32: Rao GN, Piegorsch WW, and Haseman JK (1987). Influence of body weight on the incidence of spontaneous tumors in rats and mice of long-term studies. Am. J. Clin. Nutr. 45: Sandusky GE, Van Pelt CS, Todd GC, and Wightman K. (1988). An immunocytochemical study of pituitary adenomas and focal hyperplasia in old Sprague-Dawley and Fischer 344 rats. Toxicol. Pathol. 16: Seilkop S (1995). The effect of body weight on tumor incidence and carcinogenicity testing in B6C3F1 mice and F344 rats. Fund. Appl. Toxicol. 24: Thurman JD, Bucci TJ, Hart RW, and Turturro A (1994). Survival, body weight, and spontaneous neoplasms in ad libitum-fed and food-restricted Fischer-344 rats. Toxicol. Pathol. 22: Turnbull BW and Mitchell TJ (1978). Exploratory analysis of disease prevalence data from survival/sacrifice experiments. Biometrics 34: Witt WM, Sheldon WG, and Thurman JD (1991). Pathological endpoints in dietary restricted rodents: Fischer 344 rats and B6C3F1 mice. In: Biological Effects of Dietary Restriction, L Fishbein (ed). Springer-Verlag, New York, pp