A Practical Approach, Second Edition=Ronald D. Ho.pdf

DEVELOPMENTAL AND REPRODUCTIVE TOXICITY STUDY FINDINGS 353Prediction of hazard in large human populations from small-scale animal studies is difficultbecause of the limited statistical power of the latter. Regulatory decisions derived from animalstudies may be based on the results from 20 to 25 pregnancies per dose level, whereas humanexposure may involve thousands or millions. Studying toxicity at the MTD presumes maximizationof response, which ameliorates the difference in power between the two situations. Thus, bioassaysutilizing conventional numbers of sampling units and doses near the therapeutic dose may notindicate what would be the important toxicologic events in larger human populations. Likewise,for chemicals, use of the MTD in the absence of kinetic data indirectly, though crudely, establishesexposure and maximizes the ability to detect sensitive responders. Advocates of the MTD conceptalso contend that studying toxicity approaching the mortality portion of the dose-response curveis essential in the routine evaluation of pharmaceuticals to provide clinicians the types of adverseevents that they may encounter at relevant therapeutic doses in clinical trial patients. Clearly, anegative data set in a preclinical study diminishes the ability of the clinical investigator to rapidlyidentify and manage dose-limiting toxicities. Data generated at the MTD may assist in identificationof the symptomatology that is critical for assessment of human toxicity or overdose and for selectingantidotes or appropriate courses of therapy.E. General Statistical ConsiderationsStatistical considerations play an especially important role in the evaluation of developmental andreproductive toxicity data. Both continuous (e.g., fetal body weight) and binary (e.g., postimplantationloss) measures are produced by these studies, and proper interpretation requires considerationof the statistical power of the study and the litter effect.1. Statistical PowerStatistical power is the probability that a true effect will be detected if it occurs. It is formallydefined as 1β, where β is the probability of committing a Type II error (false negative). 51 Power isdependent on the sample size, background incidence, and variability of the endpoint in question,and the significance (α) level of the analysis method. For example, the sample size (number oflitters) needed to detect a 5% or 10% change in an endpoint is dramatically lower for a continuousmeasure with low variability, such as fetal body weight, than for a binary response with highvariability, such as embryolethality (resorptions). 52 Consequently, significant changes in fetal weightare often detected at dose levels lower than those at which effects on embryo or fetal survival areobserved. However, despite an adequate sample size, the large number of endpoints evaluated in adevelopmental or reproductive toxicity study dictates that several spurious statistically significantdifferences will likely occur because given a significance level of 0.05, a Type I error (false positive)will occur 5% of the time. For example, because a standard developmental toxicity study withANOVA 53 /Dunnett’s 54 and Kruskal-Wallis/Mann-Whitney 55 statistical analyses performed on allparametric and nonparametric data, respectively, may involve as many as 100 to 300 individualstatistical hypothesis tests, the possibility exists for numerous spurious statistical findings.A replicated or unbalanced study design may also augment the statistical power of a study. Forexample, a large-scale, replicated dose-response study of the herbicide 2,4,5-trichlorophenoxyaceticacid demonstrated that such a study design may assist in resolution of problems of interspeciesvariability determination, high- to low-dose response extrapolation, and reproducibility of low-leveleffects. 522. The Litter EffectThe litter must be considered the experimental unit in developmental and reproductive toxicitystudies because the litter is the unit that is randomized, and individual fetuses or pups within litters© 2006 by Taylor & Francis Group, LLC