Within-subject biological variation estimated using real-world data strategies (RWD): a systematic review.

Abstract

Biological variation (BV) is defined as the variation in the concentration of a measurand around the homeostatic set point. This is a concept introduced by Fraser and Harris in the second part of the twentieth century. BV is divided into two different estimates: within-subject BV (CV_I) and between-subject BV (CV_G). Biological variation studies of biomarkers have been gaining importance in recent years due to the potential practical application of these estimates. The main applications of BV in the clinical laboratory include: the establishment of Analytical Performance Specifications (APS), estimation of the individual's homeostatic set point (HSP), calculation of Reference Change Value (RCV), estimation of individuality index calculation (II), and establishment of personalized reference intervals (prRI). The classic models for obtaining BV estimates have been the most used to date. In these studies, a target population ("normal" population), a sampling frequency and time, and a number of samples per individual, among other factors, are defined. The Biological Variation Data Critical Appraisal Checklist (BIVAC) established by the Task Group-Biological Variation Database (TG-BVD) of the European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) represents a guide for the evaluation and performance of these direct studies. These methods have limitations because they are laborious, expensive, invasive, and are based on an ideal population. In recent years, models have been proposed to obtain BV estimates based on the Real-World Data (RWD) strategy. In this case, we move from a model with a low number of individuals (direct methods) to a population model using the data stored in the Laboratory Information System (LIS). RWD methods are presented as an alternative to overcome the limitations of direct methods. Currently, there is little scientific evidence on the application of RWD models since only five papers have been published. In these papers, three different working algorithms are proposed (Loh et al., Jones et al., and Marques-Garcia et al.). These algorithms are divided into three fundamental stages for their development: patient data and study design, database(s) cleaning, and statistical strategies for obtaining BV estimates. When working with large amounts of data, RWD methods allow us to subdivide the population and thus obtain estimates into subgroups, what would be more difficult using direct methods. Of the three algorithms proposed, the algorithm developed in the Spanish multicenter project BiVaBiDa is the most complete, as it overcomes the limitations of the other two, including the possibility of calculating the confidence interval of the BV estimate. RWD methods also have limitations such as the anonymization of data and the standardization of electronic medical records, as well as the statistical complexity associated with data analysis. It is necessary to continue working on the development of RWD algorithms that allow us to obtain BV estimates that, which are as robust as possible.