Δ(13)C and δ(13)C values of endogenous urinary steroids represent physiological random variables. Measurement uncertainty and biological scatter likewise contribute to the variances. The statistical distributions of negative controls are well investigated, but there is little knowledge about the corresponding distributions of steroid-users. For these reasons valid discrimination of steroid users from non-users by (13)C/(12)C analysis of endogenous steroids requires elaborate statistical treatment. Corresponding Bayesian approaches are presented following an introduction to the rationale. The use of mixture models appears appropriate. The distribution of routine data has been deconvolved and characterized accordingly. The mixture components, which presumably represent steroid users and non-users, exhibit considerable overlap. The validity of a given result depends on both the analytical uncertainty and the prior probability of doping offenses. Low analytical uncertainties but high prior probabilities facilitate valid detection of doping offenses. Two recommendations can be deduced. First, before starting an (13)C/(12)C analysis, any initial suspicion should be well-substantiated. This precludes use of permissive criteria derived from the steroid profile. Secondly, knowledge of relevant (13)C/(12)C distributions is required. This must cover representative numbers of authentic steroid users. Finally, it is desirable that the conditional probability for steroid administration rather than the measurement uncertainty is calculated and reported. This quantity possesses superior validity and it is largely independent of laboratory bias. The findings suggest and facilitate flexible handling of decision limits. Proposals for the evaluation of stable isotope data are presented.
Copyright © 2012 John Wiley & Sons, Ltd.