Within-subject and between-subject biological variation estimates of 21 hematological parameters in 30 healthy subjects

Introduction

A complete (or full) blood count (CBC) is frequently requested in clinical practice and is particularly important for the diagnosis and monitoring of various clinical situations such as anemia, infection, malignancies, trauma, bleeding disorders, etc. A recent study [1] found that counting of leukocytes and measurement of hemoglobin (Hgb) were amongst the most informative of laboratory data, particularly in emergency departments. Widespread application of the CBC is associated with a requirement for objective analytical performance specifications (APSs) and an understanding of biological variation (BV), to enable safe and valid clinical interpretation of CBC components.

The first European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) Strategic Conference updated the hierarchy of models used to establish APSs as [2]:

• Model 1. The effect of analytical performance on clinical outcomes;

• Model 2. The BV of the measurand;

• Model 3. State-of-the-art analytical performance of the measurement.

It has been proposed that the APSs for Hb, platelets and neutrophil measurands of the CBC should be based on clinical outcomes (Model 1) [3]. However, no reliable clinical outcomes are yet available. Model 2 offers an alternative approach for the establishment of APSs and also provides the opportunity to use BV data for many other clinical applications.

Concerns have been raised around the quality of earlier BV studies [4, 5] and, consequently, the robustness of BV estimates collated and made available online in the online 2014 BV database [6]. Additionally, data is lacking for some CBC parameters and the data are not stratified according to sex. Some hematological measurands differ between females and males and the lack of sex-stratification may conceal requirements for more stringent APSs and other indices derivable from BV data (e.g. reference change values [RCVs]).

To address concerns about quality issues associated with acquisition of BV data, the EFLM Biological Variation Working Group initiated the European Biological Variation Study (EuBIVAS); this is a multicenter study based in six European countries that collects samples employing a common stringent protocol [7]. The study has delivered updated BV estimates for several measurands [8, 9], and those for others are in preparation. Additionally, the EFLM BV Task and Finish group developed the Biological Variation Critical Appraisal Checklist (BIVAC) for critical appraisal and meta-analysis of published BV data [10].

The analytical approach used by EuBIVAS is valid for measurands that are stable in samples stored for batched analysis. This is not possible for the measurands of CBC, given the requirement for fresh whole blood. The aim of our study was to deliver BV estimates for 21 CBC parameters based on the collection of fresh blood samples. While the analytical approach necessarily varies from that employed within EuBIVAS, all other elements of that study apply. This indicates that our study has followed rigid pre-analytical and analytical protocols and applied a rigorous statistical approach in line with the EFLM checklist [11] and the EuBIVAS protocol [7] to assure the derivation of the highest-quality BV estimates for the calculation of APSs, and other BV-related applications.

Materials and methods

Results

After initial evaluation in terms of the inclusion/exclusion criteria for the results from the first sample set, five females with Hgb levels <120 g/L and one male with a BMI >30 kg/m² were excluded. Thus, 30 of the 36 recruited subjects (17 females and 13 males) were included in the study, of whom 29 completed all 10 scheduled collections and one subject six collections. The mean ages of the females and males were 28 (range, 19–46) and 25 (range, 20–36) years, respectively.

One subject was excluded from the basophil dataset because the measurements were lower than the limit of quantification, as also suggested by homogeneity variance testing. In total, 3.9% of obtained data were excluded from final analysis. All subjects were considered to be in steady-state during the study.

With the exception of the mean corpuscular hemoglobin concentration and mean corpuscular volume (MCV), the mean values of all erythrocyte and reticulocyte parameters in females and males differed significantly, rendering partitioning of the CV_G data between the sexes necessary to derive sex-stratified estimates. However, there were no significant differences in CV_I estimates between females and males (Table 1).

For leukocyte- and platelet-related parameters, the mean values of all parameters except those of leukocytes, eosinophils, basophils and platelets differed significantly between females and males (Table 2). Significant differences in the CV_I estimates for leukocytes, monocytes, neutrophils and eosinophils were also evident (Table 2), as indicated for leukocytes in Figure 1.

Figure 1:

Median values with range (minimum–maximum) of erythrocyte (A), leukocyte (B) and platelet (C) counts for individuals based on weekly samplings for 10 weeks.

Dashed lines indicate 5th and 95th percentiles, the continuous line the median value with 95% CIs.

Generally, estimates of the components of BV of the analytes from our study appear to be lower than those listed in the online 2014 BV database (Tables 1 and 2), providing more stringent APSs and RCVs (Table 3). The NHSP (within 5% of the actual value) was 2 in the erythrocyte group, but much higher in the leukocyte group (exceeding 60 for the basophil count). In the leukocyte group, particularly for basophils and eosinophils, widening the target range to 20% still required the measurement of three to four samples to derive the estimates of homeostatic set points.

II of all hematological parameters are shown in Table 3. The IIs of 12 of 21 parameters were lower than 0.60 and for the remaining parameters II ranged from 0.60 to 1.25.

Discussion

Integrated analysis of hematological parameters employing the new generation of hematologic analyzers affords rapid, relatively inexpensive and non-invasive differential diagnosis and disease monitoring in clinical practice. Various clinical situations such as anemia, malignancies, infections bleeding disorders and several other hematological disorders fall within the wide range of clinical conditions that require clinical decision-making on the basis of such analyses [21]. Knowledge of the BV of the various parameters that make up the CBC enables identification of the required APS for analytical systems. In addition, appropriately derived BV data can be used in various ways to enable clinical interpretation and application of laboratory data [22].

For most CBC parameters, previous BV estimates have been derived, some of which are summarized in online 2014 BV database [6]. However, the lack of sex-stratification for these estimates may impact their utility and additionally, the online data lack measures of uncertainty and are derived from publications mostly dated prior to 2000. Such historical data may therefore be less relevant today given the major technological developments in analytical techniques for hematology analyzers and the fact that the analytical performances of current instruments are superior to those of older instruments [23]. It is interesting to note that the APSs for reticulocyte counts of Sandberg et al. [24] were derived using two hematological analyzers, which are no longer available in the market place. Comparison of BV datasets from historical studies and our current study is hindered by the lack of CIs in earlier work and the impossibility of direct methodological comparisons. However, generally, BV estimates delivered by our study are lower than those made available in the online 2014 BV database (Tables 1 and 2). Additionally, our data indicate that for the majority of CBC parameters BV estimates, in most cases either CV_I or CV_G, should be sex-stratified for valid application (Tables 1 and 2). Thus, indices such as the RCV may differ by sex for some parameters. Further complexity may arise upon aging, especially in females. This is because the kinetics of RBC production change with age [25, 26].

For RBCs and reticulocyte parameters mean concentrations in females were as expected lower than those of males. However, CV_Is were not significantly different between sexes (Figure 1; Table 1). Similar observations were made by Buoro et al. [23].

A study evaluating the BV of hematological parameters in adults and elderly subjects reported lower CV_I and CV_G levels in females than males for RBC, Hb and Hct [27]. Critical evaluation of this study identified factors that may have impacted the reliability of the data: The protocol featured collection of only four samples per subject, impacting the power of the study [28]. The cited work also employed a less robust statistical approach than our study; estimates of CV_A were based on quality control samples and not via measurement of replicate subject blood samples, which is the recommended approach. In addition, the CIs of the CV_I and CV_G values were not reported, rendering a comparison of BV data between the two studies difficult.

Different mean concentration levels were also observed for some leukocyte subgroups (Table 2). With the exceptions of the CV_I and CV_G values of basophils, and the CV_G values of lymphocytes and eosinophils, the leukocyte parameters exhibited similar patterns in terms of both CV_I and CV_G, with numerically higher BV estimates for females compared with males. Additionally, the CV_I and CV_G estimates reported by our study are clearly lower than those of the online 2014 BV database [6], and substantially so for eosinophils and basophils. Our new data thus deliver more stringent APSs (Table 3) for imprecision, in particular of eosinophils and basophils (Table 2). Higher CV_A values widen the CIs of BV estimates [28]. Because a wide CI range decreases the utility of BV data, our results emphasize the need for further development of the analytical methods to meet APSs based on BV.

No difference in mean levels or CV_I between sexes was observed for platelets, however, CV_G was significantly lower in the male group. We found lower estimates of CV_I and CV_G for all platelet parameters with the exception of the CV_I of the platelet distribution width (PDW) when compared to the online 2014 BV database [6]. Another study, [27], evaluating CBC parameters stratified by sex, observed similar results for females and males (CV_I 13.7% and CV_G 14% for both genders) in terms of platelets.

Varying sampling intervals may also produce different BV estimates. A recent study derived BV estimates of platelet parameters over 5 consecutive days (day to day variation) and 5 consecutive weeks (week to week variation) using the Sysmex XN analyzer [29]. The CV_Gs were similar in the two situations while the day to day variation CV_I was approximately 50% lower than the week to week variaton CV_I. This can be explained by the short duration (5 days) of the study; this covered a period less than that required for platelet turnover [30]. The platelet group BV data of the present study were in agreement (all CIs overlapped) with the week to week variation reported by Buoro et al. [29]. The only exception was the platelet CV_G, which, in our study, differed significantly between the sexes; only the female CV_G was similar to the value reported by Buoro et al. [29].

The turnover time of CBC parameters varies in vivo; the turnover time for platelets is 7–10 days [31]; that for reticulocytes 1–2 days [32]; that for leukocytes ~3 weeks (subgroups differ in terms of turnover time) [33]; and that for erythrocytes ~4 months [32]. Within our study period of 10 weeks, we covered several multiples of the turnover periods for reticulocytes, leukocytes and platelets, but less than one turnover period for erythrocytes. This may impact the CV_I magnitudes of the CBC parameters and may explain why the CV_I of erythrocyte group tests was dramatically lower than the CV_I of other tests.

Knowledge of BVs enables calculation of the number of samples required to provide an estimate of homeostatic set points within a certain percentage of the true value. This is important, because test results are often evaluated on the basis of a single sample. For the erythrocyte group (Table 3), the result of a single measurement of a single sample is sufficient to predict the homeostatic set point within 10% (with the given APS). For the reticulocyte and leukocyte groups, more samples are needed to predict the homeostatic set point within a reasonable percentage such as 5% or 10%. This is the case particularly for eosinophil and basophil measurands, because they have high CV_I values; additionally, analyzers fail to meet to deliver the APSs required to evaluate imprecision. Under such conditions, replicate measurements are necessary to obtain the required estimates.

Individuality is a marked property of all CBC parameters. The IIs of 12 of 21 parameters were lower than 0.60, which as expected suggests that population-based reference intervals (RIs) are of very limited utility. In such situations, a clinically significant change may occur in a subject whose value remains within the conventional population-based RI, and who may thus be missed. In such situations, the individual is clearly the best point of reference for assessing change, armed with the knowledge of the BV of the measurand. The IIs of certain CBC parameters ranged from 0.60 to 1.25, indicating that RIs may indeed be of some value, but the preferred value of >1.40 was never exceeded.

RCV provides a useful tool for assessment of results when marked individuality compromises the use of conventional population-based RIs, particularly when monitoring patients [20]. Caution should be applied to the use of RCV in certain circumstances. The RCV is calculated from the CV_A and CV_I; RCV may not be an appropriate parameter in tests that do not meet the APSs for imprecision (i.e. eosinophils and basophils).

Conclusions

Because of the concerns raised around the quality and the reliability of the current BV database, EFLM Biological Variation Working Group initiated the EuBIVAS to updated BV estimates for several measurands. In this study strict pre-analytical protocols and appropriate statistical techniques were used to obtain reliable data for the BV of hematological parameters and the presented data is compliant with BIVAC [10]. With some exceptions, the BV estimates of hematological parameters were found to be lower than those of the BVD [6]. Because individuality is characteristic of hematological parameters, sex-related RCVs should be used instead of RIs for patient monitoring.