Enhancing the methodological framework for doping detection: from univariate to multivariate testing analysis on Athlete Biological Passport profiles

Code: 23D01BL

Alongside the evaluation of athletes’ samples for prohibited substances, the Athlete Biological Passport (ABP) has been established as a complementary pillar in the detection of doping. The fundamental principle of the ABP is to monitor over time athletes’ individual profiles with respect to selected biomarkers that may indirectly reveal anti-doping violations. These include, for example, markers from the Haematological Module–such as haemoglobin (Hgb) and haematocrites (Hct)–that may inform any use of substances for enhancing oxygen transport or delivery. Every time a new test is performed, doping violations are detected by noting significant deviations in the observed values from an athlete’s established levels for those biomarkers. The state-of-the-art of the ABP practical implementation is based on a Bayesian methodological framework called ADAPTIVE (Sottas et al., 2007). This approach combines population-based information with individual-based data for determining individual tolerance limits that discriminate between normal and atypical values in each of the biomarkers of interest. Such individual limits are continuously and adaptively updated as additional individual samples are observed. However, while allowing for personalised and adaptive ranges of tolerance, the ADAPTIVE approach is implemented on longitudinal profiles following a univariate testing approach. ABP profiles are analyzed by looking at every single biomarker independently and separately, without accounting for their dependency structure. To illustrate, for the hematological module, a “no start” decision can be determined by an atypical finding on at least one of the two primary biomarkers: Hgb and the OFF-score, a combination of Hgb and Hct. However, in general, biomarkers do not provide orthogonal information, either due to their intrinsic characteristics or because these are often combined quantities (e.g., the OFF score). Furthermore, little is known about their simultaneous alteration in the presence of prohibited substances. Clearly, if biomarkers present some correlation patterns (notably, this is the case of Hgb and Hct), univariate analyses may lead to questionable conclusions. The aim of this proposal is to generalize the current ADAPTIVE method–well established within the anti-doping community–allowing for simultaneous modelling and analysis of multiple biomarkers. The target is the joint evolution of multiple measurements over time. Inspired by the ubiquitous role of copula models in multivariate statistics (Nelsen, 2007), we propose to use a copula-based approach to flexibly represent the joint distribution of a set of biomarkers. This proposal will be developed under a methodological framework that leverages advanced tools from probability and statistics for better modelling complex characteristics of real-world data. We will focus on the haematological module and will evaluate the proposed framework on an original dataset of ABP profiles collected by NADO Italia.