Rationale for using data on biological variation

Data included in the databases

The BV components have been estimated from healthy individuals or patients suffering diverse pathologies and, therefore are separated in two different BV databases.

These databases have been updated every 2 years since 1999, by searching at Pubmed website using biological variation as the keyword.

The number of papers rejected has evolved over time, at the beginning the main reasons for rejection were not using the mathematical ANOVA test and showing a CV_A excessively high compared with CV_I. In recent years, because an increasing number of papers dealing with non-healthy situations were published, the main reason for rejection was a non-specific design of the study to determine the components of BV.

As was already seen by Fraser [6], CV_I from young people, adults and the elderly are quite similar. Children constitute a specific group; in a recent study it has been seen that CV_I values of the pediatric population seems to be different compared to those from adults, being higher for ceruloplasmin (11.3% vs. 5.8%) and glucose (11.4% vs. 5.6%) and lower for C-reactive protein (CRP) (19% vs. 42%) and γ-glutamyltransferase (GGT) (2.7% vs. 13%) [7].

Table 1 shows an extract of information contained in the database for serum glucose. The highest CV_I does not correspond to the larger number of subjects studied (n=1105), neither to the longer study (365 days) nor to the larger number of samples taken per subject (S_s=12). Moreover, the wide variation in CV_I shown in this table cannot be related to the year of publication of the study, thus results seem to be independent of the evolution of technology for glucose testing, but might be due to different methods (e.g., outlier exclusion, homogeneity etc.) used to treat the data and calculate the CV_I.

It should be emphasized that studies obtaining samples within a day, show the lowest CV_I values. For example, serum cholesterol the lowest CV_I=1.5% corresponds to a paper that takes samples every 8 h within a day, whereas the median CV_I is 6.1%. Furthermore, serum cardiac troponin I shows the lowest CV_I=3.4% for a frequency of sampling of 4 h, whereas the median CV_I is 12.9%. For this reason values from 1-day studies have been excluded from the final estimations of CV_I and CV_G in the database from healthy subjects.

Our second database deals with patients diagnosed with different pathologies. The first publication in 2007 compiled 66 analytes and included 34 disease states [13], whereas the 2014 update contains 97 analytes and 41 diseases, covering many types of pathologies. For the majority of analytes, CV_I values from patients seem to be similar to those from healthy individuals, as was already seen by Fraser [14]. An example extracted from our database is CV_I for albumin in patients with diabetes mellitus type I, chronic renal impairment and chronic hepatopathy are 2.8%, 2.9% and 3.3%, respectively, whereas in healthy people CV_I is 3.1%.

However, in some pathologies, the CV_I of the analytes seems to be higher than the CV_I of healthy subjects, as shown in Table 2. This should be taken into account when applying the reference change value (RCV) to consecutive results of an analyte, either for automatic verification of results [26] or for reporting significant changes in a patient’s health status [27].

Use of databases in laboratory medicine

The use of BV data from healthy people covers many purposes, the most widely accepted being to derive quality specifications for analytical imprecision, bias (B_A) and total error [28, 29]. For monitoring purposes analytical imprecision has to be maintained below 0.5*CV_I. In this situation the contribution of laboratory error to the total variation is calculated as 12% [30]. For diagnosis, case finding and screening purposes, B_A has to be maintained below 0.25*(CV_I+CV_G) in order to share population-based reference intervals [31]. These are the most common specifications which are known as “desirable” specifications. Other levels of quality namely “minimum” (more permissive) and “optimum” (more restrictive) have also been suggested [32].

The 2014 updated database shows these three levels of quality specifications for 358 analytes [4].

BV data from healthy subjects are also important to design internal quality control procedures for the analytical phase, which should include four steps:

• To define total error allowable (TE_A);

• To measure analytical imprecision (CV_A) and B_A;

• To calculate σ metrics: σ=(TE_A–B_A)/CV_A;

• To search for an appropriate control rule.

Selection of TE_A is a key point, because it will result in the use of a permissive or restrictive rule for accepting analytical runs. When applying the Stockholm [28] and Milan [33] hierarchical criteria for defining quality specifications, if a “top criterion” is applied, σ could be small and this could lead to a restrictive operative rule being used. On the contrary, if criterion at the bottom of the hierarchy is used, σ will usually be larger and a more relaxed operative rule would be appropriate. For example, creatinine measured in one of our laboratories using Jaffe method has CV_A=2.2% and B_A=4.3%. If the quality specification for total error is based on BV, TE_A=8.9%, σ=2.1 and a multi-rule with several controls per run should be used. If the specification is based on the state of the art, TE_A=20, σ=7.1 and a simple rule with two controls per run would be used. TE_A=20% is the minimum level of quality defined by consensus of four Spanish scientific EQA organizers, based on the results obtained by the participants in the four programs [34]; this value is similar to other organizations, such as Clinical Laboratory Improvement Amendments (CLIA) [35] and RiliBÄK [36]. It seems to be clear that defining analytical quality specifications could have an important impact on patient safety.

Another application of BV data from healthy people is to interpret the EQA reports. Figure 1 illustrates an example for serum creatinine in a report of the SEQC-EQA program. When a laboratory has a percentage deviation versus the peer-group mean higher than the BV-derived acceptance limit, this could initiate an alert for a corrective action.

Figure 1:

SEQC-EQA monthly report for serum creatinine for a laboratory.

Left side figure: Frequency histogram of results for a control sample. Left side table: total and accepted number of results from all laboratories, method-group and peer-group (same method and instrument) of laboratories, as well as mean and standard deviation for these three groups. Right side figure: last 12 results of the individual laboratory compared with the standard deviation index related to the peer-group mean (horizontal dot-lines) and the limits derived from BV for total error (gray shadows). Right side table: laboratory result in SI and conventional units, its deviation related to the peer-group mean, expressed in standard deviation index and in percentage. In bold: desirable deviation for total error derived from BV for creatinine. SEQC-EQA uses stabilized (non-commutable) control sera, targeted by the peer-group mean.

Although some colleagues believe that reaching the BV-derived specifications may be extremely difficult, Figure 2 illustrates that 80%–100% of the majority of analyte results included in the SEQC-EQA serum biochemistry scheme of 2013 achieved the goal; even when considering the most “difficult” analyte (sodium), 55% of the results were also within the BV limits.

Figure 2:

Percentage of results satisfying BV limits for TE_A.

SEQC-EQA report. This figure corresponds to the 2013 serum basic biochemistry program, which had 888 participating laboratories. X-axis: analytes included in the serum biochemistry program with quality specifications for total error allowable (TE_A) derived from BV are published [4]. Alb, albumin; ALP, alkaline phosphatase; ALT, alanin-aminotransferase; Amy, α-amylase; AST, aspartate-aminotransferase; bil, bilirubin; Cal, calcium; Cho, cholesterol; Cl, chloride; Cre, creatinine; Fe, iron; GGT, γ-glutamyltransferase; Glu, glucose; HDLc, HDL-cholesterol; K, potassium; LDH, lactate dehydrogenase; Li, lithium; Mg, magnesium; Na, sodium; Osm, osmolality; P, phosphate; Pro, total protein; Tri, triglycerides; Ura, urate; Ure, urea; Y-axis: percentage of results reaching the quality specifications for TE_A.

From another perspective, BV data are also useful to assess whether there is any current available method providing laboratory information that could potentially compromise patient safety. Figure 3 shows serum creatinine results obtained by 23 Spanish laboratories participating in a SKML pilot study using commutable controls with values assigned by reference methods. The samples are targeted with reference methods, undertaken in either the Joint Committee for Traceability in Laboratory Medicine (JCTLM) listed reference laboratories or in International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) network laboratories [37]. All laboratories using the alkaline picrate kinetic (Alk picr k) (method gave unacceptably high results at the clinical decision level, while various laboratories using the compensated method produced dispersed results. Only the enzymatic method gave all results within the limits derived from BV for total error [38]. It has to be emphasized that this can be firmly stated because this EQA uses commutable samples that are targeted with reference methods.

Figure 3:

Serum creatinine.

Percentage deviation to reference-method value. X-axis: serum creatinine concentration, expressed in mmol/L, of 11 control samples. Y-axis: percentage deviation of results with respect to the reference-method value. Methods used by participating laboratories: Alk picr k, alkaline picrate kinetic (Jaffé); Alk pick co, alkaline picrate kinetic compensated; Enz, enzymatic method.

BV can also be applied to see if a change in serial results can be explained by within-subject BV and analytical variation only. In this case, the RCV has to be used for a change between two [2, 39, 40] or more [41] consecutive results. The formula for a change between two results is: RCV=z*2^1/2(CV_A²+CV_I²)^1/2 where z is the probability that a change between the two results can be explained by biological and analytical variation only. The probability for detecting changes increases when considering two analytes together. An example can be seen in Figure 4 where RCV of various renal post-transplanted patients for creatinine and urate combined reveals significant changes in patients suffering complications, whereas no changes are seen for patients with no complications [42].

Figure 4:

Renal post-transplanted patients.

RCV for creatinine and urate combined. Results of a study of 75 renal recipients made in a hospital laboratory of one of the authors. Patients were monitored for 3 years and consecutive results using of creatinine and urate were used to calculate RCV [42]. X- axis: percentage differences between consecutive results of creatinine in patients post-transplanted. Y-axis: percentage differences between two? consecutive results of urate in patients post-transplanted. Vertical lines: reference change values for creatinine calculated together with urate, at both sides from zero. Horizontal lines: reference change values for urate calculated together with creatinine, at both sides from zero.