Biological variation: a still evolving facet of laboratory medicine

Numerical data on the components of random biological variation (BV) have been generated for over 40 years. These have been used for a variety of purposes in laboratory medicine, including setting analytical quality specifications, assessing the significance of the difference seen in serial results from an individual and investigating the utility of conventional population-based reference values. ¹ The generation of such data is not easy. ² However, applications are facilitated through the availability of databases which give single figures for the components of BV, namely, within-subject (CV_I) and between-subject (CV_G).

Generation of a comprehensive database of components of BV was initiated in 1997 by the Analytical Quality Commission (AQC) of the Spanish Society of Clinical Chemistry (SEQC). ³ A scoring system was designed to evaluate the robustness of data, and this has five essential criteria for inclusion: papers must be designed to examine BV, the performance index (PI = CV_A/0.5CV_I) must be <2.0, where CV_A is the analytical component of variation, the components must be derived by ANOVA or another recommended calculation method, ² the data must be from apparently healthy individuals and the median estimates of CV_I and CV_G must be documented. The structure of the on-line database and the criteria used for generation and update have been recently described. ⁴ The database provides analytical quality specifications for imprecision, bias and total allowable error, at minimum, desirable and optimum levels of quality. ⁵ It contains data on 358 measurands from 247 papers as single numbers for CV_I and CV_G. Where multiple studies exist for a single measurand, the incorporated numbers represent median values. The database has been updated every two years and is currently in its eighth edition. ⁶ Recently, a number of demerits have become apparent. For example, for 202 measurands, the data appear in a single publication, 129 have data in two to nine publications and 27 in ≥10 publications. Interestingly, 55% of the measurands in the first group had the maximum scores for both the PI and the statistical methods used, which suggests that these estimates of BV are reliable: similar data exist for the other two groups. ⁴

It is dogma that, in general, estimates of CV_I are similar over time span of study, number of samples, health and disease and other factors. However, considerable heterogeneity of the CV_I estimates across studies is a major weakness of the database. Carobene et al. ⁷ showed that published BV data for alanine aminotransferase, aspartate aminotransferase and gamma glutamyl transferase demonstrated a wide range of values derived from inconsistent protocols. The quality of the presentations of the data was variable. These findings raise further concerns around the utility of the data currently available and highlight the need for critical appraisal of publications on BV. The Working Group on BV (WG-BV) of the European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) has presented these concerns at the 1st EFLM Strategic Conference on Defining Analytical Performance Goals. The many insightful presentations are available on the Internet. ⁸

The WG-BV has made recommendations regarding publication of the results of studies on BV. The most important is the Standard for Reporting of Studies of Biological Variation, which incorporates a checklist: this will soon be made available on the EFLM website. The checklist identifies key elements to be reported to enable safe, accurate and effective transport of BV data-sets across health-care systems and provides a comprehensive list of items that should be reported in any publication on BV in the usual title/abstract/keywords, introduction, methods, data analysis, results and discussion sections. The checklist is mapped to the domains of a minimum data-set required to enable transportability of the published data. Given the importance of BV data, it would seem timely to reassess the information held in the current BV database to ensure that a revision, due in 2016, delivers the most robust of information to enable this process.

Recently, Simundic et al. ⁹ published a proposal for the standardised use of symbols and terms to define the components of BV: recommended were CV_I for within-subject BV, CV_G for between-subject BV and CV_A for analytical variation. For analytical variation, it was recommended that mode of derivation and type (such as reproducibility, reliability or total) as well as the number of analyses, runs and time period should always be provided. In addition, terms and calculation formulae were recommended for the reference change value (RCV) and the index of individuality (II).

We hope that authors will adhere to the proposed EFLM checklist and symbols and terms when undertaking studies on the generation and application of BV data, particularly on as yet unstudied measurands. Reviewers and editors have major roles in ensuring these desirable developments. In addition, we advocate use of the work of Roraas et al., ¹⁰ which will prove of significant value in the design of the experimental work and the calculation of the now regarded as essential confidence intervals for CV_I estimates.