Models to estimate biological variation components and interpretation of serial results: strengths and limitations

Abstract

Biological variation (BV) has multiple applications in a variety of fields of clinical laboratory. The use of BV in statistical modeling is twofold. On the one hand, some models are used for the generation of BV estimates (within- and between-subject variability). Other models are built based on BV in combination with other factors to establish ranges of normality that will help the clinician interpret serial results for the same subject. There are two types of statistical models for the calculation of BV estimates: A. Direct methods, prospective studies designed to calculate BV estimates; i. Classic model: developed by Harris and Fraser, revised by the Working Group on Biological Variation of the European Federation of Laboratory Medicine. ii. Mixed-effect models. iii. Bayesian model. B. Indirect methods, retrospective studies to derive BV estimates from large databases of results. Big data. Understanding the characteristics of these models is crucial as they determine their applicability in different settings and populations. Models for defining ranges that help in the interpretation of individual serial results include: A. Reference change value and B. Bayesian data network. In summary, this review provides an overview of the models used to define BV components and others for the follow-up of patients. These models should be exploited in the future to personalize and improve the information provided by the clinical laboratory and get the best of the resources available.