A multivariate Bayesian learning approach for improved detection of doping in athletes using urinary steroid profiles

Area of Science:

• Sports Science
• Biomarker Analysis
• Forensic Toxicology

Background:

• Urinary steroid profile analysis is key for anti-doping in sports.
• Current methods use univariate models for individual biomarker limits.
• Simultaneous analysis of multiple biomarkers could improve abnormality detection.

Purpose of the Study:

• To propose a multivariate Bayesian adaptive model for longitudinal data analysis in anti-doping.
• To extend existing single-biomarker models used in forensic toxicology.
• To enhance the detection of abnormal values in athletes' urinary steroid profiles.

Main Methods:

• Developed a multivariate Bayesian adaptive model for longitudinal biomarker data.
• Utilized Markov chain Monte Carlo (MCMC) sampling methods.
• Implemented a one-class classification algorithm to handle scarce abnormal data and adapt decision boundaries.

Main Results:

• The model was tested on a database of 229 athletes with diverse sample classifications (normal, atypical, abnormal).
• Demonstrated improved detection performance compared to traditional univariate methods.
• Showcased the effectiveness of a multivariate approach in identifying doping.

Conclusions:

• The proposed multivariate Bayesian model offers a more robust and personalized approach to anti-doping.
• This method has significant potential to enhance the accuracy and efficiency of doping detection in sports.
• Further adoption of multivariate strategies is recommended for advanced anti-doping strategies.