National Center for Toxicological Research*

Size: px

Start display at page:

Download "National Center for Toxicological Research*"

Tamsin Anderson
6 years ago
Views:

1 Cause-of-Death Assignment at the National Center for Toxicological Research* RALPH L. KODELL, 1BOON-NAM BLACKWELL, 2 THOMAS J. BUCCI, 2AND DAVID L. GREENMAN 3 1Division of Biometry and Risk Assessment, 2Pathology Associates, Inc., and 3Office of Scientific Coordination, National Center for Toxicological Research, Jefferson, Arkansas The system for assigning cause of death in animal studies of carcinogenicity at the National Center for Toxicological Research (NCTR) is described. An empirical study of the NCTR s experience with its current cause-of-death assignment system based on selected representative experiments is reported. Issues investigated include the degree of confidence associated with histologic cause-of-death assignment, potential age-, dose-, and sex-related differences in assigned grades of certainty of cause of death, and frequencies of identification of various organ-specific and systemic diagnoses as the cause of death. Implications for ageadjusted statistical tests of carcinogenicity that require cause-of-death data are discussed. Keywords. Rodent bioassay; cancer; neoplasm; tumor; fatal; incidental; context of observation; ageadjusted INTRODUCTION Since its inception, the National Center for Toxicological Research (NCTR) has requested routine assignment of cause of death for all dead and moribund animals examined histologically in carcinogenicity studies. A determination as to whether a particular tumor under consideration was fatal to an animal or was simply an incidental finding at necropsy is important for determining the type of information that an animal contributes to age-adjusted statistical tests for carcinogenic effects (12). Biostatisticians and toxicologic pathologists have discussed the advisability and feasibility of assessing cause of death on a case-by-case basis in animal studies (4, 5, 8, 10, 11, 15, 16). At the EDo, Workshop it was concluded that when data on time-toonset of tumors are not available, a pathologist s assessment as to whether or not an animal died from a tumor, together with an expression of the degree of uncertainty, is useful in statistical analyses of carcinogenicity studies (5). However, the Interdisciplinary Discussion Group on Carcinogenicity Studies indicated that pathologists felt that designating tumors as either fatal or incidental was a difficult judgement that could achieve only partial accuracy * Address correspondence to: Dr. Ralph L. Kodell, Division of Biometry and Risk Assessment, HFT-20, National Center for Toxicological Research, 3900 NCTR Rd., Jefferson, AR (16). Statistical methods have been developed to aid in the interpretation of experiments for which causeof-death data are either lacking or perhaps unreliable ( 1, 7). A number of rodent studies in which cause of death has been determined on a case-by-case basis can be cited (e.g., 3, 4, 9, 13, 14). However, as was recently observed by Ettlin et al. (3), there is still considerable controversy among toxicologic pathologists as to whether or not the cause of death of an individual animal in toxicity and carcinogenicity studies can be established with accuracy. Although not all examining pathologists at the NCTR necessarily agree that cause of death can be assigned reliably, they are willing to provide their best assessment of cause of death in the interest of enhancing the statistical analyses of tumor data. Cause-of-death assignment in the histopathology system at the NCTR was at first very rigid, with no provision for a graded assessment or for equivocation. It has evolved over the years into a more flexible system, which includes 4 grades of certainty (6). The purpose of this paper is to report NCTR s experience with its current cause-of-death assignment system and to evaluate its usefulness for selected representative studies. Issues of interest include the proportions of animals assigned the various grades of certainty of cause of death, potential dose-, sex-, and age-related differences in assigned grades of certainty of cause of death, and frequencies

242 TABLE I. - Basic experimental information on 104-wk antihistamine studies. of identification of various organs and diagnoses as the cause of death.

2 242 TABLE I. - Basic experimental information on 104-wk antihistamine studies. of identification of various organs and diagnoses as the cause of death. METHODS The studies selected for this analysis were 104- wk chronic feeding studies of antihistamines in Fischer 344 rats and B6C3F, mice conducted at the NCTR under the National Toxicology Program (NTP). The test chemicals studied were doxylamine (rats: experiment 406; mice: experiment 407), pyrilamine (rats: experiment 408; mice: experiment 409), and triprolidine (rats: experiment 414; mice: experiment 415). Details of the conduct of these studies are provided in NCTR technical reports for the given experiments. The primary interest here is in the assignment of cause of death to the dead and moribund animals from these studies. Table I presents for each experiment the basic design that gave rise to the dead and moribund animals that were examined. The total number of animals listed reflects the initial number assigned to each experiment, including those killed at both interim and terminal sacrifices. As shown in Table I, there was a much higher proportion of dead/moribund animals in the rat studies than in the mouse studies. For each experiment, all histologic examinations of dead and moribund animals were conducted by a single pathologist. There were three pathologists in all-one assigned to experiment 406; a second one assigned to experiments 407, 414, and 415; and a third one assigned to experiments 408 and 409. The NCTR s current system for grading the certainty of cause of death (cause of morbidity, in the case of moribund animals) is as follows. If a particular neoplasm (or other pathological condition) can be identified by the examining pathologist as the most likely reason for the animal s death or removal, then it is reported as the probable cause of death. At most 1 probable cause of death may be assigned. If a probable cause of death is not identified, then a neoplasm (or other lesion) that is believed to have contributed to an animal s death, but is not thought to be the singular cause, may be reported by the pathologist as a contributory cause. More than 1 contributory cause of death may be assigned to an animal. Lesions that are thought to have had a possible influence on an animal s death or removal may be listed as equivocal. If no neoplasm or other condition is reported as a probable, contributory, or equivocal cause of death for an animal, then that animal is reported as having died from an unknown cause. Although previous automated data collection systems for histopathology at the NCTR supported the routine recording of cause of death, the computerized system presently in use does not support this effort. Consequently, cause of death is recorded by the pathologist in prose in the &dquo;animal notes&dquo; section of the histopathology report. For the usual statistical analysis of tumor data, this information is merged with other histology data in individual animal records prior to analysis. Data preparation for the present analysis began with pathology animal notes reports that were generated for each experiment by a pathology data entry clerk. The clerk highlighted on the printed notes the histologic causes of death of all dead/moribund an-

243 TABLE 11. -frequency distribution for grade of certainty of histologic cause of death. - L-V1.J stanus ior cause oi ueam. b Number outside (inside) parentheses is number (percentage) of animals.

3 243 TABLE 11. -frequency distribution for grade of certainty of histologic cause of death. - L-V1.J stanus ior cause oi ueam. b Number outside (inside) parentheses is number (percentage) of animals. imals. A statistical assistant tabulated the data from the animal notes report, creating a separate record for each cause of death for each dead/moribund animal. Variables in each record included experiment number; dose level; sex; carcass identification number; pathologist s identification code; and indicators for cause-of-death grade of certainty, causeof-death organ, and cause-of-death diagnosis. The statistical assistant next keyed the tabulated data into a computer file. A third person then checked the accuracy of the computer file by comparing a printout of that file against the printed animal notes. RESULTS For each experiment, Table II gives a frequency distribution of animals according to the grade of certainty of the histologic cause(s) of death. In each of the experiments, a high proportion of causes of death was designated as probable. In all cases for which a probable cause was given, no contributory causes were listed. For the relatively small proportions of animals for which contributory causes were diagnosed, there were always at least 2 such causes given. In only 2 of those cases did an animal have 3 contributory causes assigned. The only animal for which a tumor was reported to be an equivocal cause of death had no other causes listed. In general, this indicates a tendency to assign a single cause of death per animal. The proportion of unknown histologic causes of death was relatively low in all 6 experiments (Table II), but was highest in 3 experiments that were assigned to the same pathologist (pathologist 2; experiments 407, 414, 415). Whereas the proportion of unknown causes of death was lowest for pathologist 1 (experiment 406), the proportion of contributory causes of death was highest for that pathologist. Where a species comparison can be made (for pathologists 2 and 3, Table II), there does not appear to be a difference between species regarding a pathologist s propensity to provide a specific diagnosis for histologic cause of death. In only 1 case was the histologic cause of death listed as equivocal. In these experiments, the examining pathologists did not show a need for this category of grading. Table III provides dose-specific frequency distributions for grade of certainty of histologic cause of death for each experiment. There is little indication that assignment of cause-of-death grade was related to dose of test agent, although for the mice there appears to be a slight tendency for lower proportions of unknown causes of death in the median and high TABLE IIL-Dose-specific frequency distributions for grade of certainty of histologic cause of death. COD stands for cause of death. b Number outside (inside) parentheses is number (percentage) of animals.

4 244 TABLE IV.-Sex-specific frequency distributions for grade of certainty of histologic cause of death. COD stands for cause of death. b Number outside (inside) parentheses is number (percentage) of animals. dose groups compared to the control and low dose groups, while for the rats the opposite is true. Table IV provides similar information on grade for each sex. For these experiments, the grade of certainty of cause of death appears not to have been strongly related to the sex of the animals, although there was slightly greater certainty among male rats than female rats and among female mice than male mice. Table V provides time-specific frequency distributions for grade of certainty of cause of death. The selected time intervals correspond to those used by the NTP for nonparametric analysis of incidental tumor data (2). These data give no consistent indication that grade of certainty of cause of death was related to an animal s time on study. A frequency distribution for both organ-specific and systemic diagnoses that occurred at least twice in the combined 3 rat experiments appears in Table VI. The category of highest frequency in all 3 experiments was mononuclear cell leukemia, a systemic disease involving multiple organs. Approximately 45% of all deaths in the rat experiments were attributed to that disease, with little variation among experiments. Among organ-specific lesions, pituitary gland adenoma was consistently listed with highest frequency. Table VII gives a similar frequency distribution for the 3 mouse experiments. No particular organ-specific diagnosis occurred consistently with high frequency in the mouse experiments, although hepatocellular carcinoma of the liver occurred often in experiment 415. Categories of highest frequency included various types of lymphoma, again a systemic disease involving multiple organs. From 30% to 50% of all deaths in the mouse experiments were ascribed to lymphomas. DISCUSSION In general, these experiments show that NCTR s pathologists were willing to assign a single histologic cause of death per animal. In a high proportion of cases, a probable cause of death was identified. In only 1 case was the histologic cause of death listed as equivocal. Clearly, in these experiments, the examining pathologists did not see a need to have a category between contributory and unknown for as- TABLE V. - Time-specific frequency distributions for grade of certainty of histologic cause of death. COD stands for cause of death. Number outside (inside) parentheses is number (percentage) of animals. &dquo;

5 I ABLE V 1.- l hree rat experiments: trequency distribution tor organ-specl1ic and systemic ftistologic diagnoses occumng at least twice as probable causes of death. 245 signing causes of death. These data indicate that the equivocal category might be superfluous and could be deleted from the cause-of-death assignment system. With respect to the issue of whether or not &dquo;old age&dquo; lesions complicate the interpretation of tumorigenicity assays, in these studies the degree of certainty in assigning cause of death did not appear to be associated in any consistent way with an animal s age. In addition, there was little indication that the assignment of cause-of-death grade was related to either the dose level of the test agent or the TABLE VIL-Three mouse experiments: frequency distribution for organ-specific and systemic histologic diagnoses occurring at least twice as probable causes of death.

6 246 sex of the animal. It is possible that different results would be obtained for strongly carcinogenic test chemicals. The fact that the frequency of unknown histologic causes of death was fairly high in some experiments is not necessarily a problem with respect to statistical tests for carcinogenicity. The important point is whether a particular tumor was the cause of death of an animal. It seems reasonable to assume that any tumor in an animal for which the cause of death is unknown is itself not the cause of death. With respect to contributory and equivocal causes, the issue is not so clear-cut. In the statistical analysis of a particular type of tumor, some analysts might include in the fatal category tumors listed as contributory or equivocal causes of death, while others might treat such tumors as incidental findings. It is not possible to determine with the present data whether the experiments with higher proportions of unknown or contributory histologic causes of death are more or less accurate than those with lower proportions. There is no way to assess the degree of accuracy of the presently reported histologic cause-of-death data, with the available information. However, the propensity for pathologists in these NCTR experiments to assign singular conditions as probable causes of death does support the notion that they could identify with a reasonable degree of confidence a lesion most likely to be responsible for an animal s leaving the experiment. It also supports the feeling expressed by Ettlin et al (3) that agreement among pathologists from different laboratories might be reached on criteria to classify age-related lesions as fatal or incidental. For statistical analyses of tumor data, having cause-of-death data on a case-bycase basis is preferable to a statistician s making a blanket assumption that all tumors of a given type are either fatal or nonfatal. Even when such a blanket assumption by a statistician is justifiable, it can only be made based on the case-by-case experience of toxicologic pathologists. ACKNOWLEDGMENTS The authors thank Becky Rogers and Lisa Wiley for compilation and explanation of the cause-ofdeath data, Richard Allen for assistance with interpreting organ and diagnosis codes, Susan Taylor for tabulation of data and preparation of computer files, and Paula Card for data verification. Also thanks go to Dr. Darnell Jackson for making technical reports available; to Dr. Winslow Sheldon for manuscript review, clarification of terms and procedures, and many helpful recommendations; and to Drs. David Gaylor and James Chen for encouragement. REFERENCES 1. Archer LE and Ryan LM (1989). On the role of causeof-death data in the analysis of rodent tumorigenicity experiments. Appl. Stat. 38: Bailer AJ and Portier CJ (1988). Effects of treatmentinduced mortality and tumor-induced mortality on tests for carcinogenicity in small samples. Biometrics 44: Ettlin RA, Stirnimann P, and Prentice DE (1994). Causes of death in rodent toxicity and carcinogenicity studies. Toxicol. Pathol. 22: Haseman JK, Eustis SL, and Ward JM (1994). Contributing causes of death and the use of this information in the statistical evaluation of tumor data. In: Pathology of the Aging Rat, U Mohr, DL Dungworth, and CC Capen (eds). ILSI Press, Washington, DC, pp Hughes DH, Bruce RD, Hart RW, Fishbein L, Gaylor DW, Smith JM, and Carlton WW (1983). A report on the Workshop on Biological and Statistical Implications of the ED 01 Study and Related Data Bases. Fund. Appl. Toxicol. 3: Kodell RL, Farmer JH, Gaylor DW, and Cameron AM (1982). Influence of cause-of-death assignment on time-to-tumor analyses in animal carcinogenesis studies. J. Natl. Cancer Inst. 69: Lagakos SW, and Louis TA (1988). Use of tumour lethality to interpret tumorigenicity experiments lacking cause-of-death data. Appl. Stat. 37: Lee P (1988). Assumptions in analyses of the bioassay: A statistician s view. In: Carcinogenicity: The Design, Analysis, and Interpretation of Long-Term Animal Studies, HC Grice and JL Ciminera (eds). Springer-Verlag, New York, pp Maita K, Hirano M, Harada T, Mitsumori K, Yoshida A, Takahashi K, Nakashima N, Kitazawa T, Enomoto A, Inui K, and Shirasu Y (1988). Mortality, major cause of moribundity, and spontaneous tumors in CD-1 mice. Toxicol. Pathol. 16: Morgan KT, and Eustis SL (1988). Criteria for classification of neoplasms for pathologists and statisticians. In: Carcinogenicity: The Design, Analysis, and Interpretation of Long-Term Animal Studies, HC Grice and JL Ciminera (eds). Springer-Verlag, New York, pp Park C (1988). Biological assumptions in the bioassay. In: Carcinogenicity: The Design, Analysis, and Interpretation of Long-Term Animal Studies, HC Grice and JL Ciminera (eds). Springer-Verlag, New York, pp Peto R, Pike MC, Day NE, Gray RG, Lee PN, Parish S, Peto J, Richards S, and Wahrendorf J (1980). Guidelines for simple, sensitive significance tests for carcinogenic effects in long-term animal experiments. In: Long-Term and Short-Term Screening Assays for Carcinogens: A Critical Appraisal (IARC Monographs, Supplement 2), R Montesano, H Bertsch, and L Tomatis (eds). International Agency for Research on Cancer, Lyon, pp Shimokawa I, Higami Y, Hubbard GB, McMahan

7 247 CA, Masoro EJ, and Yu BP (1993). Diet and the suitability of the male Fischer 344 rat as a model for aging research. J. Gerontol. 48: B27-B Shimokawa I, Yu BP, Higami Y, Ikeda T, and Masoro EJ (1993). Dietary restriction retards onset but not progression of leukemia in male F344 rats. J. Gerontol. 48: B68-B Shubik P (1983). Objective of carcinogenicity testing. Fund. Appl. Toxicol. 3: Squire RA (1988). Criteria for classifying neoplasms and use of data on nonneoplastic lesions. In: Carcinogenicity: The Design, Analysis, and Interpretation of Long-Term Animal Studies, HC Grice and JL Ciminera (eds). Springer-Verlag, New York, p. 107.

Journal of Statistical Software

JSS Journal of Statistical Software August 2006, Volume 16, Issue 7. http://www.jstatsoft.org/ A Computational Tool for Testing Dose-related Trend Using an Age-adjusted Bootstrap-based Poly-k Test Hojin