Prediction of plasma volume and total hemoglobin mass with machine learning

Abstract

Anemia is a widespread disease commonly diagnosed through hemoglobin concentration ([Hb]) thresholds set by the World Health Organization (WHO). However, [Hb] is subject to significant variations mainly due to shifts in plasma volume (PV) which impair the diagnosis of anemia and other medical conditions. The aim of this study was to develop a model able to accurately predict total hemoglobin mass (Hbmass) and PV based on anthropometric and complete blood count (CBC) analyses. 769 CBC coupled to measures of Hbmass and PV using the CO-rebreathing method were used with a machine learning tool in a numeric computing platform (MATLAB regression learner app) to calculate the model. For the predicted values, root mean square error (RMSE) was of 37.9 g and 50.0 g for Hbmass, and 194 ml and 268 ml for PV, in women and men, respectively. Measured and predicted data were significantly correlated (p<0.001) with the coefficient of determination (R2) ranging from 0.73 to 0.81 for Hbmass, and PV, in both women and men. The bland-altman bias between estimated and measured variables was in average of -0.69 for Hbmass and 0.73 for PV. This study proposes a valid model with a high prediction potential for Hbmass and PV, providing relevant complementary data in numerous contexts. This method can notably bring information applicable to the epidemiology of anemia, particularly in countries with high prevalence or in specific population such as high-altitude communities.