Variability in anti-Müllerian hormone levels: a comment on Sowers et al., Anti-Müllerian hormone and inhibin B variability during normal menstrual cycles

Abstract

This paper (1)presents valuable and carefully collected data on the variability of serum Anti-Mullerian hormone (AMH) over the menstrual cycle, with 20 women each providing daily blood samples for one complete cycle. It is unfortunate that sampling were not prolonged giving data for all or part of a second cycle which would have allowed them to estimate repeatability and obtain rigorous statistical tests of cycle effects - this should be considered for future similar studies. Having obtained this data, the authors create 4 subgroups based on the AMH levels and then compare AMH between these groups (table 1). It is of course an elementary statistical error to perform hypothesis tests comparing groups on the variable that was used to create those groups - akin to testing whether men taller than 6ft are significantly taller than those below 6ft. This statistical howler serves to ring alarm bells for the statistically literate reader, but in itself does not undermine their conclusions. However they go on to graphically compare the variability of AMH between the low and high AMH groups (fig 1 A-D). As we would naively expect the variability in AMH to increase with the AMH level (and indeed the authors assuming this in performing their fitting on log-transformed data) it is not surprising that on the untransformed scale there is more variability in the higher AMH groups. Although it is difficult to read the graphs accurately, all 4 groups seem to be consistent with a peak-to-peak variation of a little less than two-fold. If the same data were plotted on a log scale the evidence for different behaviour between low and high AMH would, at least, be much weaker. Although their modelling framework would permit an (exploratory) hypothesis test, no such test is provided, but the graphical data would suggest that there areno significant differences between the groups.