Statistical issues in implementing the marker method

The detection of growth hormone (GH) abuse by athletes raises statistical problems as well as biochemical ones. We outline the statistical approaches to the various issues which have arisen during the work of the GH-2000 and GH-2004 teams; in particular, it considers the need to develop a test which detects GH abuse in any elite athlete ‘beyond reasonable doubt’. The test needs to be robust enough to withstand legal challenge, while minimising the risk of false accusation. The paper identifies various issues which arise in the development of such a test, and describes how these were resolved.Since GH is a naturally occurring hormone whose concentration varies substantially, its abuse cannot be detected by direct measurement. The methodology considered here made use of markers whose levels are more stable but are influenced by GH. The statistical methods employed aimed to make the best use of these markers, taking account of all factors contributing to errors in measurement.There were two key steps in the statistical investigation undertaken to develop the GH detection algorithm. The first was the requirement to identify GH-dependent biomarkers which would identify GH doping reliably and robustly for a significant length of time. The second was to calibrate the GH detection method in the elite athlete population, so that the method would be applicable to all athletes, regardless of age, sex and ethnicity, and regardless of whether they had recently sustained an injury.In practice, further work was needed to ensure that the methodology met the WADA testing protocol rules, but also that the proposed method can be used by any WADA accredited lab without placing any athlete at an unfair disadvantage and ensuring a high level of confidence in any result produced.