Multivariate approaches to improve the interpretation of laboratory data

More than two decades ago, Callum Fraser provided evidence that the “normal” values of many measurands lie within a much narrower range than the population-based reference interval (RI) [1]. Furthermore, quantitative studies on within- and between-subject biological variation have shown that most laboratory measurands exhibit marked individuality. This individuality can be quantified using the index of individuality (II), first introduced by Harris in 1981 [2]. Despite this, individual test results are still commonly interpreted against broad population-based reference intervals. Such an approach neglects a crucial fact: many measurands are highly individualized and regulated around unique, stable values – so-called setpoints – which differ significantly between individuals. Recently, Foy et al., writing in Nature, presented compelling evidence underscoring the clinical importance of these personalized set points. By analyzing decades of blood-test data across nine key parameters, they demonstrated that incorporating personalized reference intervals into clinical diagnostics could advance precision medicine to a new level [3]. In an accompanying editorial, Steven J. R. Meex and Kristin Moberg Aakre highlighted that “personalized reference intervals offer an exciting path forward, potentially reshaping how preventive medicine is viewed” [4]. At the same time, they cautioned that “these models are rarely implemented in routine practice” owing to high model uncertainty and the difficulty of establishing whether observed changes truly reflect clinically relevant variation – factors that currently limit their widespread adoption. We have recently emphasized that a fundamental step in advancing value-based laboratory medicine (VBLM) is enhancing the interpretation of laboratory results through the adoption of objective criteria [5]. In fact, the interpretation of laboratory results is a comparative process that necessitates the availability of reliable additional information beyond the numerical data itself, including accurate terminology, harmonized measurement units, reference values such as reference intervals (RI) and decision limits (DL), and interpretative comments. These parameters, usually defined as “comparators” aim to make the data “actionable” [6]. In addition to the lack of harmonization in the adoption of the recommended measurement units, current references used for decision-making practice still rely on data derived from population studies. Recently, Coskun and Coll. developed an algorithm to estimate personalized reference intervals (pRI) based on analytes homeostatic set points (HSPs) and within-subject biological variations (CV_I) [7], 8], thus providing new evidence and findings for replacing population-based reference data. In this issue of the Journal, Coskun and coll. move to another fundamental step in improving the interpretation of laboratory results. The authors, in fact, underline that current interpretation practices apply reference intervals and reference change values in a univariate manner – that is, each analyte in the panel is interpreted independently and no reference data are available to interpret the panel as a whole [9]. However, clinicians use test panels containing multiple analytes to enhance clinical significance and improve the accuracy of decision-making. As correctly highlighted by Coskun and coll. “metabolism is a network of biomolecules, each of which is related to others” [9]. Thus, a paradigm shift is necessary in interpreting patients’ laboratory data. Measurands should be evaluated collectively to capture the metabolic and functional state of the organism. This necessitates a multivariate approach, applying multivariate RIs (MRIs) to panels of related analytes. According to the Authors “multivariate approaches – such as MRI and multivariate reference change value (MRCV) – which are based on the correlations among measurands, offer a more comprehensive and metabolically oriented framework to interpret laboratory data. This enables a more realistic and clinically relevant assessment of patient results. The model also provides a pragmatic solution for the interpretation of omics data. In particular, proteomics and metabolomics have yielded valuable insights into alterations in protein profiles and biomolecules associated with disease [10]. Nevertheless, despite extensive research, the clinical translation of these multiple analytes has remained limited, highlighting a persistent gap between discovery and routine medical practice. As a clinically meaningful and pragmatic strategy, grouping biomarkers into relevant panels and interpreting them collectively – with the support of MRI for the entire panel – rather than individually, may accelerate their clinical implementation and enhance the accuracy of diagnosis and decision-making. Although the models described by the authors seem quite complex, the availability of advanced information technologies and further recently published information [11] should allow clinical laboratories to apply these concepts in clinical practice. The laboratory report, through the quality of the information it conveys, serves as the true calling card for both clinicians and patients, with personalized reference intervals providing the essential key to the correct interpretation of laboratory data. Enhancing the laboratory report, therefore, represents a fundamental opportunity to advance Value-Based Laboratory Medicine (VBLM).