How should data on murine spontaneous abortion rates be expressed and analyzed?

Spontaneous abortions in the CBA x DBA/2 model are normally reported as number of resorptions/total number of implantations (R/T), pooling data from individual mice. The significance of differences between groups has been determined using non-parametric statistics (e.g. chi-square or Fisher's Exact test) based on a priori predictions. Recently, it has been argued that medians with box plots should replace the accepted standard, but this deprives readers of data needed to verify P-values, and leads to inferences incompatible with biological and statistical reality.

Raw data on 173 individual CBA x DBA/2 matings were analyzed by median and mean, along with R/T data from 18 independent experiments containing 5-10 mice per group. Raw data from 19 CBA x BALB/c matings were similarly analyzed.

Individual CBA x DBA/2 mouse resorption rates showed a non-Gaussian distribution, but the mean and median differed by <0.5%. Resorption data from 6 and 12 independent pools of mice were normally distributed. Only the mean enabled a between-group P-value calculation. CBA x BALB/c matings gave a median of 0 and mean of 5.1%; the data were not normally distributed, but that was because of a bimodal distribution. One group of mice had 0 abortions, and the second a mean of 13.9% abortions, and the data from the latter group were normally distributed.

Although it is possible to compare individual mice, and even individual implantation sites, in resorption (abortion) studies, as the relevant question is the significance of differences between treatment groups of mice, and reproducibility, the established classical method of reporting R/T should continue to be provided. In CBA x BALB/c matings, where abortion rates are low, using the median is misleading and may obscure the existence of two distinct populations.