Statistical Methods in Cancer Research, Volume 3: The Design and Analysis of Long-Term Animal Experiments (IARC Scientific Publication 79)
Wolfe, Robert A., Journal of the American Statistical Association
J. J. Gart, D. Krewski, P. N. Lee, R. E. Tarone, and J. Wahrendorf (eds.). Lyon, France: International Agency for Research on Cancer, 1986. x + 219 pp. $45 (available from Oxford University Press).
Although books such as Statistical Methods for Cancer Studies (Cornell 1984) supply overviews of statistical methods appropriate to the field of cancer studies, for many years statisticians have had to unearth sundry articles from the statistical literature to gain an overview of methods appropriate to the design and analysis of animal-carcinogenesis experiments. Peto et al. (1980) has been among the most important of such articles. The methods described by Peto et al. are motivated, expanded, and updated in this new IARC publication.
The first three chapters summarize design issues for evaluating carcinogenic effects by animal experiments. They present aspects of the scientific objectives and the practical limitations of animal-carcinogenesis experiments that affect choices of statistical methodology. These chapters do not (and are not intended to) constitute a handbook for designers of animal experiments. Rather, they provide the basis for the statistical methods used subsequently in the book. Two major topics are identified and then developed all the way from motivation by practical issues to resolution by suitable, but fairly straightforward, statistical methods. These topics (a) account for how heterogeneity of subjects affects cancer outcomes, and (b) account for observations of the time to tumor that are incomplete because of either death from other causes or the inability to detect when cancer first appears in a living organism. Other topics pertinent to the design of animal experiments and interpretation of animal data are not as fully developed in this book, but the bibliography provides more-than-adequate guidance for interested readers.
Chapter 5, on hypothesis testing, is by far the strongest in the book. It reviews stratification as a method to control for heterogeneity, as well as approaches commonly used to analyze both lethal and nonlethal tumors. Worked examples clarify the actual computations described in this chapter.
Chapter 6, on dose-response modeling, reviews dose-response methods for both proportions and the time to tumor, although for several reasons the coverage of dose-response models for proportions is less than satisfying. The link between models for proportions and models for the time to tumor is presented, but few derivations are given. Even though this chapter does discuss several of the most important models used for carcinogenesis dose-response relationships, it gives few worked examples for modeling methods and gives little direction for how to move from specific knowledge of cancer mechanisms toward appropriate dose-response models. Instead, it presents the different models mostly as a series of mathematical equations that could be applied to data. Even its discussion of interpreting kinetic models does not afford enough insight into why or how the models are appropriate. Readers interested in this important topic should note the good literature citations.
The last two chapters discuss several topics that are of substantial importance but for which statistical methods are less well developed, because either the problems are difficult to resolve on the basis of animal-experiment data or the methods have not yet been fully explored. Among the special topics in Chapter 7 are multiple comparisons, ordinal responses, litter effects, multiple tumor types, and historical controls, but few derivations and even fewer examples are given. …