Abstract

Objectives

To compare two hematological analyzers—the DxH-800 (DxH; Beckman-Coulter, Miami, FL) and XN-2000 (XN; Sysmex, Kobe, Japan)—with the Cell-Dyn Sapphire (SAPH; Abbott, Santa Clara, CA).

Methods

We analyzed 4,375 samples. Slide reviews were made in the presence of blast, abnormal lymphocyte, and immature granulocyte (IG) flags or nucleated RBC (NRBC) count.

Results

The analyzers exhibited excellent correlations for CBC and neutrophils but displayed a limit correlation for lymphocytes. The XN did not miss circulating blasts (0.5%–95% in microscopy). For NRBCs, the XN demonstrated a sensitivity of 90%; DxH, 74%; and SAPH, 29%. Only the XN demonstrated a correlation with microscopy, permitting a WBC six-part differential until 15% of NRBCs. The XN and DxH gave useful IG counts with a cutoff less than 5% and a WBC level more than 2,500/mm3. For abnormal lymphocytes detection, only XN demonstrated sensitivity of more than 95%, but its specificity of 54% requires adaptation.

Conclusion

The XN increases the sensitivity of abnormal cell detection compared with the other counters, permitting a seven-part differential between predefined levels, decreasing the slide review from 20% to 9%.

Recent hematology analyzers are becoming more efficient, and the number of blood cell parameters reported increases constantly.1 Automated leukocyte differentials show a good correlation with morphology (MO) and, if any abnormal cells are present, demonstrate a more accurate performance to detect these cells due to the smaller number of cells counted in MO; the interobserver variability and the lack of uniform WBC distribution on the blood film are inherent to manual microscopy. Automated digital cell morphology systems have been developed to solve these deficiencies and improve the efficiency and reproducibility of slide review.

Many hematology laboratories draw up rules based on the leukocyte differential and the presence of flags to decrease the number of microscopic reviews and to improve the quality of these reviews, in agreement with the Clinical and Laboratory Standards Institute (CLSI) consensus rules.2,3

In this study, three analyzers—the Cell-Dyn Sapphire (SAPH; Abbott, Santa Clara, CA), DxH-800 (DxH, software 1.1.3; Beckman-Coulter, Miami, FL), and XN-2000 (XN; Sysmex, Kobe, Japan)—are evaluated for repeatability, linearity, and carryover. Then, multiple comparisons among the analyzers themselves are performed for hemoglobin (Hb), mean corpuscular volume (MCV), platelets (PLTs), and WBCs, as well as among the five normal populations of circulating WBCs. We evaluate the performance of the three analyzers, focusing on the capability of flagging in the presence of abnormal cells. Nucleated RBCs (NRBCs) are quantified by these three counters. Immature granulocytes (IGs) are quantified by the XN and by the updated software (2.0) of the DxH. Countings and flagging are compared with results obtained by MO.

Materials and Methods

Repeatability, linearity, and carryover were performed for the three analyzers. During one month, 4,375 routine laboratory samples were randomly run in parallel within 4 hours of collection in K2EDTA tubes on the three analyzers. Slide reviews were realized due to the presence of blast, abnormal lymphocytes (AL), IG, or NRBC flags on at least one analyzer. Moreover, the presence of blast and lymphomatous cells was assessed by flow cytometry (FCM) on the basis of multiple clinical and/or biological criteria. In this context, the probability of missing true-positive immature cells was very low.

In total, 444 samples were flagged and analyzed by a digital microscope (HemaCAM; Horn Imaging, Aalen, Germany) on 230 cells, reviewed by three experienced cytologists (J.H., J.B., and C.S.). The choice of microscopy review by HemaCAM as a reference method is based on the results from a study performed in our laboratory demonstrating the higher sensitivity of HemaCAM to detect abnormal cells than manual microscopy.

Most analyzers have only one flag for atypical or abnormal lymphocytes. In the XN, the WBC differential (WDF) channel differentiates two types of flags: atypical lymphocytes and blast or abnormal lymphocytes. If this last flag is present, a subsequent analysis in the white progenitor cell (WPC) channel allows discrimination between blast cells and abnormal lymphocytes because of their reactivity to a fluorescent dye and their different patterns in biparametrical side scatter/side fluorescence and side scatter/forward scatter.

In this study, these two lymphocyte flags, atypical and abnormal, were assessed together to ensure homogeneity with other instruments. The morphologic criteria to assess abnormal lymphocytes are based on a size larger than a normal lymphocyte; some changes to the shape or chromatin of the nucleus, with nucleoli occasionally present; and the quantity or basophilic color of the cytoplasm, with the presence of vacuoles or azurophilic granules. The probability of finding abnormal cells is expressed into the flag for the SAPH from 0.5 to 0.9 and by the Q-flag for the XN, which is positive from 100 to a maximum of 300. The DxH does not give probabilities associated with flags.

The morphologic criterion used to determine a positive smear finding, in relationship to flagging, for the study was the presence of blast, plasma cells, and NRBCs. We used the threshold of 1% for IGs and 4% or more for ALs, as recommended by the International Society for Laboratory Hematology. Updated software (2.0) for DxH was launched in June 2012, giving an early granulocyted cell (EGC) count for each sample. Using this new software, 38 samples positive in MO for IGs were analyzed.

Statistical Analysis

Multiple comparisons including the three analyzers were performed for Hb, MCV, PLTs, and WBCs and for the seven subclasses of differential with MO. Morphologic flags were compared for each analyzer using the microscopic criteria defined above. Sensitivities, specificities, and efficiencies for each flag were calculated between each analyzer and MO. Statistical analysis was performed by GraphPad Prism 4 (GraphPad Software, La Jolla, CA) using the Wilcoxon signed rank test for paired samples, the Spearman rank test for linear regression analysis, the Bland-Altman plot for systematic differences, and the measure of agreement by the κ coefficient.

Results

Repeatability

The coefficient of variation (CV) was lower than 5% except for results of low WBCs for the SAPH and XN, low PLTs for the XN, IG counts in MO, and NRBC counts in the DxH and MO Table 1.

Table 1

Repeatability Results for Five Repeated Analyses of Each Sample, Except for Repeatability of IGs and NRBCs by Morphology (n = 10)

Linearity

Linearity data for the three analyzers showed excellent correlations (R2 > 0.99) among values for all sets of diluted samples Table 2.

Table 2

Linearity Results for Three Samples With High Values of Hb, WBCs, and PLTs Diluted With Buffer of Each Analyzer

Carryover

Carryover was 0.9%, 0.3%, and 0% for Hb; 0%, 0.1%, and 0% for PLTs; and 0.2%, 0%, and 0% for WBCs for the SAPH, DxH, and XN, respectively.

Comparisons Among the Three Analyzers for Hb, MCV, PLTs, and WBCs

Comparisons of Hb, MCV, PLT, and WBC counts on 500 samples, randomly selected, showed for the four parameters excellent correlations among the analyzers (all R2 > 0.97). A significant biological bias was observed only for PLTs by the Bland-Altman diagrams Figure 1. PLT counts used in SAPH were obtained by an optical method and were slightly higher than those obtained with the DxH and XN, both calculated by the impedance method.

Figure 1

Bland-Altman plots for hemoglobin (Hb), mean corpuscular volume (MCV), platelets (PLTs), and WBCs for the three analyzers on 500 samples. A, Hb SAPH vs DxH. B, Hb SAPH vs XN. C, Hb DxH vs XN. D, MCV SAPH vs DxH. E, MCV SAPH vs XN. F, MCV DxH vs XN. G, PLT SAPH vs DxH. H, PLT SAPH vs XN. I, PLT DxH vs XN. J, WBC SAPH vs DxH. K, WBC SAPH vs XN. L, WBC DxH vs XN.

Figure 1

Bland-Altman plots for hemoglobin (Hb), mean corpuscular volume (MCV), platelets (PLTs), and WBCs for the three analyzers on 500 samples. A, Hb SAPH vs DxH. B, Hb SAPH vs XN. C, Hb DxH vs XN. D, MCV SAPH vs DxH. E, MCV SAPH vs XN. F, MCV DxH vs XN. G, PLT SAPH vs DxH. H, PLT SAPH vs XN. I, PLT DxH vs XN. J, WBC SAPH vs DxH. K, WBC SAPH vs XN. L, WBC DxH vs XN.

Comparison Among the Three Analyzers and MO for Differentials

Correlations between analyzers and MO counts were performed for the 444 samples requiring a slide review. Correlations among the analyzers and MO counts were very good for neutrophils for the three analyzers (R2 > 0.8), displayed a limit (R2 > 0.7) for lymphocytes among MO counts and the DxH and XN, and were not present between MO counts and the SAPH. There was no correlation for monocytes, eosinophils, and basophils among the three analyzers and MO counts.

Correlations between the two new analyzers with SAPH were very good (R2 > 0.9) for neutrophils and lymphocytes but displayed a limit (R2 > 0.6) for monocyte and eosinophil counts. Correlations between the DxH and XN counts were very good (R2 > 0.9) for all classes except basophils.

Blast Flags

Eighteen cases with circulating blast cells were identified by MO and confirmed by FCM (range, 0.5%–94.7% of blasts), seven cases of acute myeloid leukemia, three cases of B-cell acute lymphoblastic leukemia, seven cases of postchemotherapy, and one case of severe infectious disease Table 3. The sensitivities of the SAPH, DxH, and XN to detect blast cells were 89%, 72%, and 100%, respectively, using the three flags of blast, IG, and AL together Table 4. Despite this association of flags, the SAPH and DxH still missed two and five cases, respectively. If considering only the flag “blast,” the sensitivities were 61%, 55%, and 72% for SAPH, DxH, and XN, respectively. The specificities for the SAPH, DxH, and XN were 92%, 89%, and 96%, with efficiency rates of 92%, 88%, and 97%, respectively. The κ coefficients demonstrated an almost perfect agreement between each pair of analyzers.

Table 3

Flags (Blast, AL, or IG) Given by the Three Analyzers for the Cases With Blast Cells in Morphology, Probability for the Cell-Dyn Sapphire, and Q-Flag Values for the XN-2000

Table 4

Sensitivities, Specificities, and Efficiencies to Detect Blast, IG (Software 1.1.3 for DxH), NRBCs, and Abnormal Lymphocytes for the Three Analyzers on 444 Samples Analyzed by Morphology (MO)

NRBC Flags

Fifty-nine cases with more than 0.5% of NRBCs were identified by MO (range, 0.5–35 NRBCs/100 WBCs), with 22 cases having more than 2% of NRBCs. The XN demonstrated the highest sensitivity for NRBC detection (81%) (Table 4), while the DxH had a sensitivity of 63% with four cases (NRBC>2%) missing in comparison to the XN. The SAPH needed to use simultaneously the fluorescent population flag to increase the number of samples detected for NRBCs, resulting in a sensitivity of 29%.

In contrast to the sensitivities, the specificities and efficiency rates for the three devices were 90% or more using both thresholds of NRBCs. Measures of association demonstrated an almost perfect agreement for the DxH/XN association, with a κ coefficient superior to 0.8, and moderate agreement for the two associations of SAPH/DxH and SAPH/XN, with κ coefficients of 0.45 and 0.5, respectively.

Figure 2 illustrates the regression curve of the NRBC counts for the XN and the DxH with MO, demonstrating only a biologically significant R2 between the XN and MO counts. With the updated software (2.0), the DxH still demonstrated a bad correlation with an R2 of 0.19.

Figure 2

Correlations of the percentage of nucleated RBCs (NRBCs) determined by the (A) XN-2000 and the (B) DxH-800 with the manual count method on 200 cells. XN-2000: y = 0.9546x + 0.7278; R2 = 0.77945. DxH-800: y = 0.4343x + 1.032; R2 = 0.32455.

Figure 2

Correlations of the percentage of nucleated RBCs (NRBCs) determined by the (A) XN-2000 and the (B) DxH-800 with the manual count method on 200 cells. XN-2000: y = 0.9546x + 0.7278; R2 = 0.77945. DxH-800: y = 0.4343x + 1.032; R2 = 0.32455.

Immature Granulocyte Flags

More than 1% of IGs were counted in 127 cases by MO, including infectious and hematologic malignant diseases. The sensitivities of flagging were 20%, 29%, and 88% for the SAPH, DxH, and XN, respectively (Table 4). The SAPH and DxH (software 1.1.3) had missed 96 and 86 cases, respectively, in comparison to XN. The specificities for the SAPH, DxH, and XN were 90%, 87%, and 84% and the efficiency rates 70%, 71%, and 85%, respectively. No satisfactory agreement among the three devices was demonstrated for IGs (Table 4).

Statistical tests of the IG count between XN and MO were done and demonstrated an R2 of 0.5 for the 127 samples analyzed. When some outlier samples were removed (false positives due to toxic granulations or hyposegmentation of neutrophils), the R2 was increased to 0.65. When we calculated the positive predictive value for the presence of more than 1% of IGs in MO counts, we found the highest value (68%) when the IG rate given by the XN was superior or equal to 2%.

We analyzed the impact of a high “left-shift” flag on the IG rate given by the XN. For leucopenia inferior to 2.5 × 109/L, whatever the IG rate given by the analyzer, samples were false positive in more than 75% of cases if there was a presence of a left-shift flag with a probability level between 200 and 300. With the updated software (2.0) for the DxH, an important increase in sensitivity was observed, 84% instead of 29% with the version 1.1.3 software. The specificity was decreased to 67%. Since an IG count was also given for all blood samples, a statistical test was done with MO counts and showed an R2 of 0.23 for the 38 samples studied, increasing to 0.77 when outlier samples were removed (same criteria as for the XN).

AL Flags

Thirty-one cases were positive in MO for ALs, either for lymphomatous cells (two non-Hodgkin lymphoma [NHL] and five B-cell chronic lymphoid leukemia [B-CLL]) or for more than 4% of abnormal lymphocytes (n = 15) or for the presence of plasma cells (n = 9). For all lymphomatous cell cases, a monoclonal population was found by flow cytometry. For the five cases of B-CLL, the AL flag was present in four cases for the SAPH, in five cases for the DxH, and in four cases for the XN, which gave a false “blast” flag for the fifth case. Two cases of NHL (other than CLL) were analyzed, with the AL flag present in one case for the SAPH and DxH and in the two cases for the XN. Fifteen cases containing 4% or more of ALs were detected, flagging in two cases for the SAPH, in one case for the DxH, and in all cases for the XN. Nine cases contained plasma cells (range, 0.5%–1.2%), with the AL flag in one case for the SAPH, in none for the DxH, and in eight cases for the XN.

The specificities for the SAPH, DxH, and XN were 94%, 92%, and 54%, with efficiency rates of 89%, 87%, and 60%, respectively (Table 4). Measures of association demonstrated a slight or poor agreement for the three associations, with κ coefficients of 0.15 or less.

Discussion

With the SAPH and XN, repeatability was not good for low WBCs (<0.3 × 109/L) and low PLTs (<10.0 × 109/L) with XN (Table 1). For these two analyzers (SAPH and XN), an extended count is possible for low WBC events samples, but it was not used in this study. Without using the extended WBC mode, the CVs were less than 15% for the XN and more than 20% for the SAPH for WBC counts inferior to 0.3 × 109/L, measured by an optical method.

For PLTs, in the SAPH, both optical and impedance methods can be used to give PLT counts; we therefore used the optical counts, with counts slightly higher than impedance counts and leading to the bias previously described. For the XN, the impedance method is used in the first instance, and if an abnormal scattergram is detected by information processing unit software, optical counts or PLT-F (fluorescent platelets, a new channel of the XN series) can be used. The reason for this high CV for PLT counts is also due to the expression of results. For the XN, results increase by a thousand increments, while for the DxH and SAPH, they increase by 100 and 10 increments, respectively.

In MO, neutrophil percentages significantly increased and those of lymphocytes and monocytes decreased in comparison to values of analyzers. These underestimations are explained by the well-known lysis of lymphocytes4 during the smearing, with the presence of nuclei shadows not counted in differentials. Monocytes5 are mainly present on the sides of smearing, an area not scanned by digital microscopy.

The XN showed a higher sensitivity than the SAPH and DxH for all flags of interest, allowing the detection of two samples with blasts, 96 with IGs, 32 with NRBCs, and 16 with ALs not detected by the SAPH, as well as five with blasts, 85 with IGs, eight with NRBCs, and 22 with ALs not detected by the DxH.

Blast Cells

The XN, thanks to its new channel WPC, increases the blast rate detection in comparison to the two other devices and, moreover, avoids false-positive flags due to monocytosis, as unfortunately observed in the SAPH and DxH. Our limited population of 18 samples with circulating blast cells does not allow reaching a significant difference between the pairs of analyzers with the κ coefficient, despite the difference in sensitivity for the three analyzers. Further investigations with a higher number of positive cases should be done.

NRBCs

Today, it appears clearly that the new dedicated NRBC channel (WNR channel) of the XN is the best actual method to enumerate NRBCs, according to Rosenthal et al6 and our study, in the range of 0.5% to 15% of NRBCs. This channel performs the WBC count and specific differentiation and enumeration of basophils and NRBCs using a patented cell-specific lysing agent, which preserves basophil integrity, and a new fluorescent dye, which stains stronger WBCs than NRBCs.

The DxH uses seven parameters—impedance, radiofrequency, and five laser light scatter measurements (“VCS” technology)—to enumerate NRBCs, but the results obtained are disappointing in comparison to those given by the XN (Figure 2). The moderate agreements between SAPH/DxH and SAPH/XN reflect the bad sensitivity of the SAPH for NRBCs.

IGs

Automated IG counts are now available on the XN and DxH (software 2.0). For the XN, the combination of a lysing product and a fluorescent stain in the WDF channel allows differentiating the IG cluster right above the neutrophil cluster in the biparametric histogram side scatter/fluorescence. The enumeration of EGCs is possible on the DxH by a novel combination of volume, conductivity, and light scatter parameters, which provide increased separation between EGCs and other WBCs.

Results of sensitivity for IGs with the XN are considerably better then those of SAPH and DxH (software 1.1.3), but the absence of a perfect correlation with MO (R2 = 0.65) requires some algorithmic rules. The new software (2.0) of the DxH improves the sensitivity to detect IGs. Preliminary results demonstrated a good correlation with MO, with an R2 superior to the XN (R2 = 0.77). This weak correlation is explained in different ways, including the smaller number of cells counted by MO, the subjectivity involved in the morphologic classification (especially the difficulty in separating metamyelocytes from band cells),7 the inhomogeneous repartition of cells on blood smears, and the well-known important CV for these rare events on blood smearing.8

A previous study had reported a better correlation coefficient between the XE-IG Master (XE-2100 and XE-5000) and FCM method (R2 = 0.92) than with MO (R2 = 0.60).9 In another article,10 the same authors have shown that for all samples with an automated IG count of more than 3%, IGs were found by morphology, unless the WBC count was less than 0.5 × 109/L. Although the software is the same in the XN series, we did not find the same results, because the left shift and leucopenia demonstrated a great impact on the predictive positive value.

Further investigations must be done to evaluate interferences in the IG count, such as the left-shift flag and granularity index.11–13 Nevertheless, the advantage given by the XN and DxH (software 2.0) to get a “seven-part” leukocyte differential provides great time savings for a laboratory when a small IG count is detected, allowing the validation of a small amount of IG, with a “cutoff” of 5% or less and with a leukocytosis of more than 2.5 × 109/L, without slide review if any other flag is present. Rosenthal et al14 determined the same optimum IG cutoff of 5% or less to use without MO verification, with a decrease of 30% of smear reviews in their laboratory, confirming our results.

ALs

During our study, we used the AL flag without any threshold of probability for the SAPH or the Q-flag for the XN. Combining the new WDF channel with the WPC channel of the XN, six cases of mature lymphocytic neoplasms out of seven were detected by the XN. Despite this good sensitivity, its poor specificity needs further investigation of the AL flags in the WPC channel. The XN demonstrates good capability to flag the presence of plasma cells, in contrast with the two other devices, which are present in the AL flag and do not require a reflex test in the WPC channel. The poor and slight agreements between the pairs of analyzers reflect the different technologies of each analyzer to detect ALs.

Conclusion

In conclusion, the XN, thanks to its new combination of fluorescent dyes and specific lysing agents, provides significantly more efficient results than the DxH and the SAPH for NRBCs and IGs, with the possibility of using a WBC seven-part differential with a range from 1% to 5% for IGs when the leukocytosis is 2.5 × 109/L or higher and a range of NRBCs from 1% to 15%.

The new software (2.0) of the DxH gives good preliminary results for using EGCs in routine differential results with the same criteria for leukocytosis, but the efficiency of NRBC recognition seems to be not improved compared with the 1.1.3 version. Moreover, the AL flag in the XN seems to increase significantly the detection of mature lymphoid neoplasms with the restriction of bad specificity. Further tests are needed with a better “tuning” of the Q-flag for ALs.

Finally, the XN allows reducing, in a general and oncologic hospital, the level of slide reviews for blast, IGs, and NRBCs, with an increased sensitivity of detecting these abnormal circulating cells. For the first time, we have decreased the slide review for our laboratory from 20% with the SAPH to 9.3% with the XN.

References

1.
Kang
S-H
Kim
HK
Ham
CK
et al
.
Comparison of four hematology analyzers, CELL-DYN Sapphire, ADVIA 120, Coulter LH750, and Sysmex XE-2100, in terms of clinical usefulness
.
Int J Lab Hematol
 .
2008
;
30
:
480
486
.
2.
Clinical and Laboratory Standards Institute (CLSI)
.
Reference Leukocytes (WBC) Differential Count (Proportional) and Evaluation of Instrumental Methods: Approved Standard
 .
2nd ed
.
Wayne, PA
:
CLSI
;
2007
.
CLSI Document H20-A2
.
3.
Clinical and Laboratory Standards Institute (CLSI)
.
Validation, Verification, and Quality Assurance of Automated Hematology Analyzers: Approved Standard
 .
2nd ed
.
Wayne, PA
:
CLSI
;
2010
.
CLSI Document H26-A2
.
4.
Good Practice Guidelines on the Reporting of Smudge Cells
 .
Toronto, Canada
:
QMP-LS
;
2001
.
5.
Kratz
A
Bengtsson
H-I
Casey
J
et al
.
Performance evaluation of the CellaVision DM96 system
.
Am J Clin Pathol
 .
2005
;
124
:
770
781
.
6.
Rosenthal
N
Connell
B
Brown
B
et al
.
Evaluation of a new method for the enumeration of Nucleated Red Blood Cells on the new Sysmex XN Automated Hematology Analyzer
.
Int J Lab Hematol
 .
2012
;
34
(
suppl 1
):
1
180
,
PM52
.
7.
Cornbleet
PJ
Novak
R
.
Lack of reproducibility of band neutrophil identification despite the use of uniform identification criteria
.
Lab Hematol
 .
1995
;
1
:
89
96
.
8.
Henry
JB
.
Clinical Diagnosis and Management by Laboratory Methods
 .
20th ed
.
Philadelphia, PA
:
Saunders
;
2001
.
9.
Briggs
C
Kunka
S
Fujimoto
H
et al
.
Evaluation of immature granulocytes counts by the XE-IG Master: upgraded software for the XE-2100 automated hematology analyser
.
Lab Hematol
 .
2003
;
9
:
117
124
.
10.
Briggs
C
Linssen
J
Longair
I
et al
.
Improved flagging rates on the Sysmex XE-5000 compared with the XE-2100 reduce the number of manual film reviews and increase laboratory productivity
.
Am J Clin Pathol
 .
2011
;
136
:
309
316
.
11.
Zimmerman
M
Cremer
M
Hoffman
C
et al
.
Granularity index of the SYSMEX XE-5000 hematology analyser as a replacement for manual microscopy of toxic granulation neutrophils in patients with inflammatory diseases
.
Clin Chem Lab Med
 .
2011
;
49
:
1193
1198
.
12.
Le Roux
G
Vlad
A
Eclache
V
et al
.
Routine diagnostic procedures of myelodysplastic syndromes: value of a structural blood cell parameter (NEUT-X) determined by the Sysmex XE-2100
.
Int J Lab Hematol
 .
2010
;
32
:
e237
e243
.
13.
Furundarena
JR
Araiz
M
Uranga
M
et al
.
The utility of the Sysmex XE-2100 analyzer’s NEUT-X and NEUT-Y parameters for detecting neutrophil dysplasia in myelodysplastic syndromes
.
Int J Lab Hematol
 .
2009
;
32
:
360
366
.
14.
Rosenthal
N
Connell
B
Brown
B
et al
.
Automated immature granulocyte counts on the new Sysmex XN Automated Hematology Analyzer
.
Int J Lab Hematol
 .
2012
;
34
(
suppl 1
):
1
180
,
PM51
.