Statistical variation of selected clinical pathological and biochemical measurements in rodents.

Abstract

The variability in clinical pathological and biochemical measurements among replicates is often greater than the effects under study. To minimize the effects of this variability, it is recommended that replicated concurrent analyses (randomized blocks) of groups of animals be used. That is, some samples from each of the groups of animals to be compared are analyzed at the same time. This process is replicated until a sufficient number of animals are sampled. If replicated concurrent analyses are not conducted and clinical measurements are made at different times for different groups of animals, the biases may be large. Clinical data were examined from several experiments to illustrate that the problems of clinical measurements are not confined to a particular endpoint, species, or sex. In one case, four times as many animals would have been required using nonconcurrent analyses to achieve the same precision as for concurrent analyses. Permutation analyses of two of the data sets indicate that statistical conclusions concerning the comparison of average clinical levels in different groups of animals would have been incorrect, falsely indicating higher or lower levels in a group over one-half of the time with nonconcurrent analyses.