Z-scores-based methods and their application to biological monitoring: an example in professional soccer players

• Published in: Biostatistics (Oxford, England) England), 2019-01, Vol.20 (1), p.48-64
• Main Authors: Saulière, Guillaume, Dedecker, Jérôme, Marquet, Laurie-Anne, Rochcongar, Pierre, Toussaint, Jean-Francois, Berthelot, Geoffroy
• Format: Article
• Language: English
• Subjects: Humanities and Social Sciences; Methods and statistics
• ISSN: 1465-4644; 1468-4357
• DOI: 10.1093/biostatistics/kxx044

Summary The clinical and biological follow-up of individuals, such as the biological passport for athletes, is typically based on the individual and longitudinal monitoring of hematological or urine markers. These follow-ups aim to identify abnormal behavior by comparing the individual's biological samples to an established baseline. These comparisons may be done via different ways, but each of them requires an appropriate extra population to compute the significance levels, which is a non-trivial issue. Moreover, it is not necessarily relevant to compare the measures of a biomarker of a professional athlete to that of a reference population (even restricted to other athletes), and a reasonable alternative is to detect the abnormal values by considering only the other measurements of the same athlete. Here we propose a simple adaptive statistic based on maxima of Z-scores that does not rely on the use of an extra population. We show that, in the Gaussian framework, it is a practical and relevant method for detecting abnormal values in a series of observations from the same individual. The distribution of this statistic does not depend on the individual parameters under the null hypothesis, and its quantiles can be computed using Monte Carlo simulations. The proposed method is tested on the 3-year follow-up of ferritin, serum iron, erythrocytes, hemoglobin, and hematocrit markers in 2577 elite male soccer players. For instance, if we consider the abnormal values for the hematocrit at a 5% level, we found that 5.57% of the selected cohort had at least one abnormal value (which is not significantly different from the expected false-discovery rate). The approach is a starting point for more elaborate models that would produce a refined individual baseline. The method can be extended to the Gaussian linear model, in order to include additional variables such as the age or exposure to altitude. The method could also be applied to other domains, such as the clinical patient follow-up in monitoring abnormal values of biological markers.