Estimation of Plasma Volume by Machine Learning to Improve the Interpretation of the Athlete Biological Passport.

The identification of confounding factors related to plasma volume (PV) fluctuations is crucial for appropriate qualitative interpretations of Athlete Biological Passport (ABP) profiles. As part of ongoing efforts to remove PV variance from the concentration-based biomarkers such as hemoglobin concentration ([Hb]), a new machine learning model for blood volume (BV) estimation using a single complete blood count analysis was applied within the ABP framework. Forty existing ABP profiles from elite athletes and healthy control subjects were used. PV was estimated using a machine learning model trained on a previous dataset. First, a visual display of the estimated PV shift was added in overlay of individual profiles. Alternatively, individual [Hb] thresholds were adjusted in a new graphical profile to account for PV variations. Finally, a set of ABP profiles with PV estimations was presented to ABP experts to assess the model's relevance in interpreting hematological data. A moderate correlation was found between measured and estimated PV in both men (r = 0.40, p < 0.0001) and women (r = 0.39, p < 0.0001), supporting the validity of the estimation model. In addition, ABP experts favorably assessed the available PV information, particularly the visual representation of PV. This novel estimation model offers distinct advantages (e.g., same biomarkers currently analyzed from routine ABP analyses) and could therefore be of particular interest. Further application of this model in the presence of specific and transient confounding factors may allow to confirm these results.