Harmonization initiatives in the generation, reporting and application of biological variation data

Study design and reliability

Numerous studies on BV components have been published during the last four decades. There are variations between these publications in study design, methodology and applied statistical approach that may affect the BV estimates they present. To generate high-quality BV data, adherence to strict experimental and statistical protocols is required. Furthermore, in order for estimates to be applicable in clinical practice today, the analytical methods used to derive the BV data must be comparable with contemporary methods in terms of analytical specificity, as obsolete methods using different principles may in fact deliver different BV estimates (e.g. switching from competitive immunoassay to immunometric assays for parathyroid hormone). Typically, when generating BV data, a group of reference individuals, in most cases cited as apparently healthy individuals, are studied. From each of these individuals, specimens are drawn at regular and appropriate time intervals while minimizing all sources of pre-analytical variation. All specimens must be drawn, processed, transported and stored under controlled and standardized conditions [2]. Ideally, analysis should be performed in duplicate, with all samples from the same individual being processed in the same analytical run if possible, while minimizing analytical sources of variation. Thereafter, appropriate statistical analysis including assessment of outliers and homogeneity of variances must be applied, followed by dissection of the CV_A, CV_I, and CV_G components by a nested analysis of variance (ANOVA), general/generalized linear model or a similar approach [2].

To produce robust, well-characterized BV data and to ensure that data are transferable across health care systems and different populations, BV studies must conform to an ideal standardized approach, be adequately powered and properly documented at the reporting stage. This is, unfortunately, not always the case. The number of replicates, number of samples and number of individuals included in the study influence the reliability and confidence intervals (CI) of the BV estimates, and the effect of these variables varies with the ratio of analytical standard deviation (SD_A) to within-subject SD (SD_I) [11]. Generally, the lower the SD_A/SD_I ratio, the narrower the CIs. If subgroups are to be explored, a balanced number of subjects in each planned sub-group is preferable. When published, BV estimates must be accompanied by their corresponding CIs to allow for the assessment of their reliability, comparison between subgroups and the comparability with other studies.

The European Biological Variation Study (EuBIVAS)

Responding to the need for reliable BV estimates delivered from big high-quality studies, the EFLM WG-BV has designed and implemented the European Biological Variation Study (EuBIVAS) [12]. In the EuBIVAS, six European clinical laboratories in five different countries followed a strict, detailed protocol for the recruitment of subjects and for the pre-analytical phase [13]. The study population included 91 healthy volunteers (38 males and 53 females, age 21–69 years). Fasting blood samples were drawn for 10 consecutive weeks (April–June 2015) for all individuals. All samples were processed and prepared in the same way as detailed [13] and stored at –80 °C prior to shipment on dry-ice to the coordinating center in Milan. All samples from each of the participants were analyzed in duplicate on platforms within the Milan laboratory under standardized conditions. The resultant data sets underwent rigorous scrutiny and appropriate statistical analysis to enable delivery of BV estimates accompanied by CIs. So far, estimates for liver enzymes [14] and creatinine using enzymatic and alkaline picrate methods [15] have been published. For these measurands, EuBIVAS estimates are lower than the corresponding estimates available in the online 2014 BV database (Table 1) [3]. Though the reasons for this may be multi-factorial, our data so far indicate that a large and strictly performed study such as the EuBIVAS delivers lower estimates than many other BV studies; these differences have implications for the setting of APS, diagnosis and monitoring and other applications of BV data. In addition, the EuBIVAS study has enabled comparison of measurements using different analytical approaches; in the case of serum creatinine the EuBIVAS results indicated that current alkaline picrate methods may not meet BV-based APS for imprecision [15]. The next phase of the EuBIVAS, providing BV estimates for electrolytes, lipids, hormones and specific proteins, is under way [12]. Thus, the EuBIVAS provides a valuable resource and infra-structure to enable delivery of high-quality and well characterized, and therefore transportable, BV data for a large number of measurands, using a protocol that is fully compliant with the newly developed Biological Variation Data Critical Appraisal Checklist (BIVAC) [16].

The Biological Variation Data Critical Appraisal Checklist (BIVAC)

The BIVAC is designed to assess the quality of BV publications by verifying whether all essential elements that may impact upon veracity and utility of the associated BV estimates are present [16]. Its main focuses are the effect of study design, the measurement procedure and statistical handling of data on CV_I estimates. The BIVAC consists of 14 quality items (QI), which can be awarded scores A, B, C or D (Table 2) [16]. Based on the individual scores for each of the QI, an overall grade is set for the publication under review. The grade A is achieved if the publication shows full compliance with all BIVAC QI. If the lowest score for any QI is a B, then the overall grade is a B and similarly C or D if the lowest QI score is a C or D, respectively. In the BIVAC scoring system, the QI related to the overall grade are shown as a subscript, as exemplified by C_4,8,13 and B_{6,7,8,9,11,13}. BV estimates derived from studies that receive one or more D scores are considered unsuitable for use in clinical practice. The BIVAC QI relating to subjects (QI 2), samples (QI 3) and the measurement procedure (QI 4) are considered critical for the reliability and applicability of the associated measures of BV and thus can be scored as D (Table 2). It is an absolute requirement that populations from which BV data have been derived are adequately characterized. Furthermore, details on the study population, samples, sample material and timing of samples are necessary to compare with other studies, to deliver CIs and to generate global BV estimates. Additionally, it is essential that historical publications assess the same measurand as contemporary methods for the reported BV estimates to be relevant. QI 5–7 refer to pre-analytical procedures, estimates of CV_A and demonstration of steady state, all of which are required to obtain high-quality BV estimates. Statistical elements QI 8–12 relate to analysis of outliers, normality, homogeneity of variances, the applied statistical model and CI which are central for the quality of the associated BV estimates and their applicability. QI 1 (scale), 13 (number of included results) and 14 (concentrations) do not reflect the reliability of the BV estimates in themselves, but are necessary elements for interpretation and application of the data. Applying the BIVAC to 128 BV publications for 28 different measurands revealed that the QI for outlier analysis and variance homogeneity testing were most often scored as C, meaning that appropriate analysis of these elements were missing in more than 60% of publications [16]. Although many of the historical papers on BV were performed according to the standards existing at the time of publication, it is, based on our review, evident that many BV publications omit, or fail to address, essential elements, in particular for statistical analysis, which may affect the reliability and applicability of the reported BV estimates.

Meta-analysis of BV studies

For commonly requested measurands, many studies on BV have been published, as exemplified for total cholesterol, where 46 publications are included in the online 2014 BV database [3]. The common estimate that is presented in this database represents the median of estimates from studies fulfilling its inclusion criteria [6]. We have recently developed a meta-analysis approach for the delivery of global estimates [16]. With this approach, BV publications for the measurand in question are identified by systematic searches and relevant publications are thereafter appraised by the BIVAC. Only studies receiving an overall BIVAC grade A, B or C are considered fit to be considered for the meta-analysis. Thereafter, the study design is reviewed so that only studies with similar characteristics are included in the meta-analysis; as a first step studies performed in healthy adults where sampling is weekly. For the meta-analysis, the associated BIVAC grade and the width of the CI are used as weight, with the global estimate being delivered by a weighted median approach [16]. If a sufficient number of higher quality studies, i.e. BIVAC grades A and B, are available, the global estimate may be based on these alone.

The EFLM Biological Variation Database

As part of our initial BIVAC review of BV publications for enzymes, lipids, kidney and diabetes related measurands [16], work is in progress in providing updated global estimates for all these 28 measurands. These, and other measurands that will be critically appraised in the near future, will be published in a new Biological Variation Database, available on the EFLM website. Here global BV estimates with the evidence behind them and related APS will be accessible, with separate estimates being delivered for different populations and sampling intervals when relevant. This will be an important initiative in providing updated evidence-based global estimates easily accessible to users. Once this database covers the majority of relevant measurands and gains momentum in reaching potential users, it may contribute to a harmonized approach so that where appropriate, the same global BV estimates and APS will be applied in laboratories worldwide.

The Biological Variation Data Reporting Checklist

There have historically been no internationally recognized standards for production, reporting and transmission of BV data. The WG-BV published in 2015 the Biological Variation Data Reporting Checklist [18], which identifies key elements required in studies to enable accurate and effective use of BV data across population groups and healthcare systems. It is based on the same structure as the STARD [19] and identifies six main minimum data set domains including (1) title/abstract/keywords, (2) introduction, (3) methods, (4) data analysis, (5) results and (6) discussion. Compliance with this checklist indicates that studies are fit for purpose, that essential statistical analyses such as outlier and variance homogeneity testing have been reported, that common terminology has been used and that BV estimates are accompanied by key metadata. Additionally, the BIVAC, although designed for critical appraisal of already published studies (16), provides a detailed guide for those designing and delivering future studies and an aide for reviewers of manuscripts reporting on those to assure publication of the required high-quality BV publications.

Terminology for components of BV

There is a wide range of terms and symbols used for the definition of components of BV in the published literature. A recent study reviewed the 13 highest ranked clinical chemistry journals and identified 68 terms and 25 symbols for components related to individuals and 47 terms and 18 symbols for components related to groups of individuals [20]. There is a clear need for a harmonized use of terminology, and the EFLM WG-BV supports the terms and symbols for components of BV suggested by Simundic et al. [20], as outlined in Table 3. This terminology has also been supported by the Clinical Chemistry and Laboratory Medicine [21]. Additionally, up until now, there has not been a Medical Subject Heading (MeSH) heading for BV, which both adds to the varied use of BV terms as well as complicating literature-searching strategies [16]. The WG-BV suggested in 2016 the establishment of a new MeSH term for “Biological Variation” to the National Library of Medicine. This was accepted and the MeSH term has been available from December 2017, thus facilitating systematic searches for BV publications. However, the National Library of Medicine usually does not re-index previously indexed articles when new headings are added.