Multivariate Bayesian Adaptive Model of Urinary Steroid Biomarker Data for Improved Detection of Doping in Athletes
Working with scientists from the anti-doping lab at the Sports University of Cologne, we developed a method that allows for simultaneous modelling of multiple biomarkers instead of focusing on a single biomarker at a time. By generating personalised limits for each athlete using the multivariate biomarker profile, we achieve more accurate detection of atypical observations which require further investigation as potential doping cases. Preliminary empirical results from applying the proposed method to known typical and atypical cases show promise for reducing time delays and costs associated with follow-up testing. We have also developed a software implementation for use by anti-doping laboratories.