The efficient use of testing resources is crucial in the fight against doping in sports. The athlete biological passport relies on the need to identify the right athletes to test, and the right time to test them. Here we present an approach to longitudinal tracking of athlete performance to provide an additional, more intelligence-led approach to improve targeted antidoping testing. The performance results of athletes (male shot putters, male 100 m sprinters, and female 800 m runners) were obtained from a performance results database. Standardized performances, which adjust for average career performance, were calculated to determine the volatility in performance over an athlete's career. We then used a Bayesian spline model to statistically analyse changes within an athlete's standardized performance over the course of a career both for athletes who were presumed "clean" (not doped), and those previously convicted of doping offences. We used the model to investigate changes in the slope of each athlete's career performance trajectory and whether these changes can be linked to doping status. The model was able to identify differences in the standardized performance of clean and doped athletes, with the sign of the change able to provide some discrimination. Consistent patterns of standardized performance profile are seen across shot put, 100 m and 800 m for both the clean and doped athletes we investigated. This study demonstrates the potential for modeling athlete performance data to distinguish between the career trajectories of clean and doped athletes, and to enable the risk stratification of athletes on their risk of doping.
Keywords: Bayesian; competition results; data analytics; monitoring; target testing.
© 2019 John Wiley & Sons, Ltd.