Patient-specific measures of a biomarker for the generation of individual reference intervals: hemoglobin as example

Abstract

Although hemoglobin concentration measurement is among the most commonly performed blood tests, the description of global population parameters, heterogeneous factors, and within-subject variations in patients with disease remains incomplete. As absolute action values are being published in the medical literature and by government healthcare agencies, these measures are important to define patient-specific ranges of biomarkers. Here, a global clinical trial data set composed of 1,537,932 hemoglobin values from 416,374 patients and 372 clinical indications was generated over 2 years by automated analyzers in a global network of 5 laboratories. Within- and between-subject components of variance and the effect of factors age, gender, nationality, and clinical indication were determined using unbalanced multiway analysis of variance. Average within-subject variances differed significantly depending on the clinical indication (0.15-1.3 g(2)/dL(2)) but, nevertheless, remained significantly lower than between-subject variances. The main sources of between-subject variation were clinical indication and gender, followed by age and nationality. An adaptive Bayesian approach was then used to generate patient-specific ranges of hemoglobin for drug safety and efficacy assessment in clinical trials. The same methodology can be applied to the evaluation of any biomarker signal in translational medicine.