Performance evaluation of the automated body fluid analysis of the new Sysmex XR haematology analyser

Samples and sample handling

BF samples submitted to the department of Laboratory Medicine of the University Hospital Leuven, Belgium, with a volume of ≥500 µL were used for this study. CSF, pleural fluid, peritoneal fluid, CAPD, synovial fluid and BAL samples collected according to the local laboratory procedures, and <6 h old after sample collection, with no other preselection criteria, were evaluated. Samples were stored at room temperature prior to testing. The target was to include at least 60 samples for each BF type. As CAPD samples are infrequently encountered in our laboratory, they were grouped together with peritoneal fluids. All synovial fluid samples were treated with 1 mg hyaluronidase (Hyaluronidase H3757, 300–1000 U/mg, Sigma-Aldrich) per 1 mL sample, mixed well and incubated at 37 °C for 5 min prior to analysis. Hyaluronidase disrupts hyaluronic acid polymers, rendering the samples less viscous and more homogeneous, which improves the accuracy of the analysis.

Automated BF analysis method

All BF samples with request for haematological cell counting were routinely measured in the open BF mode on the Sysmex XN-Series haematology analyser (Sysmex Corporation, Kobe, Japan). Thereafter, the same samples were processed on the Sysmex XR-Series analyser (Sysmex Corporation, Kobe, Japan). Samples were mixed thoroughly by sample tube inversion before analysis, in accordance with manufacturer instructions. Pre-analytical dilution with Cellpack DCL (Sysmex Europe SE, Norderstedt, Germany) was performed whenever deemed necessary based on the appearance (increased turbidity) and viscosity of the sample. Sample dilution was also performed if the first measurement revealed cell counts that exceeded the analyser linearity range (identical for XN and XR) or when visual inspection of the WDF (white blood cell differential) scattergram revealed sub-optimal cell cluster separation (as judged by the laboratory technician), if residual sample volume was sufficient. In the latter case, both the original undiluted and diluted sample measurements were included. The dilutions ranged from 1 in 5 to 1 in 100. BAL samples were not diluted if linearity was exceeded as our routine processing of such samples is exclusively undertaken to obtain an approximate WBC count. Analytical quality control checks were conducted daily at regular time points using low and high levels of XN-Check BF (Streck, Omaha, USA). Analysis parameters included red blood cell (RBC-BF), WBC (WBC-BF), total nucleated cell (TC-BF) and WBC differential counts (Table 1).

The principle of measurement utilised in the BF analysis is identical for both the XN and XR, albeit with a different reagent system. Hydrodynamically focused DC detection method is used to count RBC-BF, whereas TC-BF, WBC-BF and differential counts are determined using fluorescence flow cytometry in the WDF channel. In brief, forward scatter (cell size), side scatter (cellular complexity) and fluorescence intensity (nucleic acid content) are combined to characterise cells. In so doing, the WBC mononuclear cell (MN) and polymorphonuclear cell (PMN) clusters can be enumerated. The MN and PMN populations are further differentiated into lymphocyte and monocyte, and neutrophil and eosinophil sub-populations, respectively, and provided as research parameters. Furthermore, cells such as macrophages and mesothelial cells, identified by high intensity fluorescence, are reported as HF-BF. Samples with malignant cell infiltration may also exhibit an increase in HF-BF [15]. The latter are excluded from WBC-BF but are included in TC-BF. If the WDF cell clustering pattern does not conform to what is expected, the analyser triggers a WBC abnormal scattergram flag. In such instances, the WBC-related parameters may be unreliable.

Standard analytical performance metrics such as e.g., linearity, carryover, and precision, were not included in this method comparison evaluation.

Management of errors

After each analysis run, results obtained from both analysers were reviewed to identify any samples with technical errors. Repeat testing was performed for all samples where a sample handling or instrument-related error (e.g., inconsistent sample density or clogged instrument flow cell) was identified if there was sufficient residual sample volume, and the sample <6 h old.

Data management

For all samples, date and time of collection and fluid type were recorded. The raw data file export from both the XN and XR was conducted on a regular basis, in accordance with manufacturer instructions.

Data analysis

Sample measurements meeting any one of the following criteria were excluded from the data analysis: >6 h after collection, RBC-BF exceeds upper limit of linearity range (>5 × 10⁶/µL), WBC-BF exceeds upper limit of linearity range (>10 × 10³/µL), raw data file not retrievable, missing measurement on one instrument, and analysis errors as described above. For samples exceeding linearity only the diluted sample measurements were included in the analysis. Data analysis was done using MedCalc^® Statistical Software version 19.8 (MedCalc Software Ltd, Ostend, Belgium). Correlation coefficients, regression slopes and intercepts were estimated for all parameters using the Spearman’s rank correlation and Passing-Bablok regression methods. Bias between the methods was assessed using Bland-Altman plots. CSF, peritoneal, pleural, and synovial fluids were analysed separately per BF type as well as collectively. The WBC abnormal scattergram flagging rate between the analysers, and the impact thereof on the parameter values, was evaluated as described above. BAL fluids were analysed independently of the other body fluids as this BF type is not part of the intended use declared by the manufacturer for measurement on Sysmex analysers. For the research parameters, only absolute counts were included in the analysis.

Ethics approval

The study respected individuals’ rights to confidentiality and was in accordance with procedures supervised by Local Authorities responsible for Ethical Research (Ethics Committee Research UZ/KU Leuven, approval on May 18, 2021; reference S64732).

Study samples

A total of 324 BF samples were collected (86 CSF, 80 peritoneal, 77 pleural, and 81 synovial fluids). One sample was excluded as it was >6 h old at the time of analysis, while 16 samples, which qualified for repeat testing, were eliminated due to insufficient volume. Among the remaining 307 samples, 25 required repeat testing due to analysis errors or missing data. Additionally, 24 samples needed dilution for re-analysis due to exceeding analyser linearity or suboptimal cluster separation as observed in the WDF scattergram. Ultimately, the 307 patient samples provided a total of 356 measurements, with 49 having two measurements.

After the application of exclusion criteria, 330 measurement pairs consisting of 86 CSF, 77 peritoneal, 78 pleural, and 89 synovial fluid samples were compared (Table 2). Where a sample was represented with two measurements, these were either undiluted and diluted, or two different dilutions. Comparison of 20 sample pairs for which both an undiluted and diluted XN and XR measurement was available, showed that the correlation coefficients before and after dilution were similar (data not shown) and thus diluted and undiluted samples were analysed together. For all samples, the cell counts for the diagnostic and research parameters (Table 1) were generated.

Of the 78 BAL samples measured, 16 exceeded the linearity range for WBC-BF and were excluded from further analysis. The remaining 62 BAL sample measurements were analysed separately.

CSF, peritoneal, pleural, and synovial fluid analysis

An excellent correlation coefficient (r) between the two analysers was observed for all diagnostic parameters; WBC-BF: 0.996, RBC-BF: 0.974, TC-BF: 0.995, MN#: 0.994, MN%: 0.830, PMN#: 0.985, and PMN%: 0.835 for all BF types collectively (Table 3). The regression line (Figures 1 and 2) for each comparison was very close to the line of equality for all parameters, and the coefficients showed excellent performance (Supplementary Tables 1 and 2). Bland-Altman analysis (Figures 1 and 2) showed that the difference of means (XN minus XR) for each parameter was zero or close to zero: WBC-BF: 0.06, RBC-BF: 0.0, MN#: 0.03, MN%: −2.2; PMN#: 0.03, PMN%: 1.6, and TC-BF: 0.06. The range for this difference did not reveal a significant bias in either direction; WBC-BF: −0.47 to 0.59, RBC-BF: −0.01 to 0.01, MN#: −0.37 to 0.44, MN%: −33.4 to 29.0, PMN#: −0.26 to 0.32, PMN%: −29.5 to 32.7, and TC-BF: −0.48 to 0.61.

Figure 1:

Correlation of XR and XN body fluid mode for WBC-BF, RBC-BF and TC-BF. Samples from four different body fluids were measured on Sysmex XR and XN analysers. Cerebrospinal fluid samples (n=86) are depicted with an open circle, peritoneal fluid samples (n=77) with an open triangle, pleural fluid samples (n=78) with a closed black circle, and synovial fluid samples (n=89) with closed grey square. Correlation graphs with regression line with confidence intervals, and line of equality (dashed) are shown for white blood cell count (WBC-BF) (A), red blood cell count (RBC-BF) (C), and total nucleated cell count (TC-BF) (E). Bland-Altman plots with difference of means (XN minus XR) (solid line) and range (1.96 SD) (dashed lines) are shown for WBC-BF (B), RBC-BF (D), and TC-BF (F).

Figure 2:

Correlation of XR and XN body fluid mode for MN#, MN%, MPN#, PMN%. Samples from four different body fluids were measured on the Sysmex XR and XN analysers. Cerebrospinal fluid samples (n=86) are depicted with an open circle, peritoneal fluid samples (n=77) with an open triangle, pleural fluid samples (n=78) with a closed black circle, and synovial fluid samples (n=89) with a closed grey square. Correlation graphs with regression line with confidence intervals, and line of equality (dashed) are shown for mononuclear cell count (MN#) (A), mononuclear cell percentage (MN%) (C), polymorphonuclear cell count (PMN#) (E), and polymorphonuclear cell percentage (PMN%) (G). Bland-Altman plots with difference of means (XN minus XR) (solid line) and range (1.96 SD) (dashed lines) are shown for MN# (B), MN% (D), PMN# (F), and PMN% (H).

Similar results were observed for the five research parameters, with excellent correlation coefficients for all BF samples collectively; NE-BF#: 0.985, LY-BF#: 0.994, MO-BF#: 0.981, EO-BF#: 0.900, HF-BF#: 0.958 (Table 3). Albeit EO-BF# had the widest confidence intervals for its correlation coefficient (0.877–0.918), with a slope of 0.911, these numbers still indicate an excellent concordance between the two systems (Supplementary Figure 1G, Supplementary Table 2). Bland-Altman analysis (Supplementary Figure 1) followed the same trend as for the diagnostic parameters, with the difference of means (XN minus XR) being zero or close to zero (NE-BF#: 0.02, LY-BF#: 0.03, MO-BF#: 0.01, EO-BF#: 0.00, HF-BF#: 0.00) with no significant bias in either direction; NE-BF#: −0.29 to 0.33, LY-BF#: −0.35 to 0.41, MO-BF#: −0.11 to 0.12, EO-BF#: −0.08 to 0.09, and HF-BF#: −0.06 to 0.06.

Analysis of each BF type independently showed that they all had correlations close to the overall study sample group average for the absolute count diagnostic parameters of the BF mode (Table 3). The Bland-Altman analysis showed that the samples that were beyond the limits of agreement for each parameter were mostly pleural and synovial fluids (closed black circle and closed grey square respectively), with the majority showing a negative bias for the XR (Figures 1 and 2). The r values for MN% (0.622) and PMN% (0.639) were however significantly lower for CSF samples, compared to the overall BF group mean values (MN%: 0.830; PMN%: 0.835). These lower r values can be attributed to the CSF samples having very few cells relative to other BF types. A one cell difference has a big impact on percentage count but is insignificant in absolute numbers. Forty-four percent (38/86) of the CSF samples had WBC-BF values below the limit of quantification (LoQ) of the analyser (0.003 × 10³/µL for both XN and XR). When these samples were excluded, the r value for both MN% and PMN% was 0.969.

The research parameters also showed an excellent correlation between XN and XR for the different BF (Table 3). Similar to the diagnostic parameters, a few samples, again mostly pleural and synovial fluids, were beyond the limits of agreement in the Bland-Altman analysis (Supplementary Figure 1). Also, most synovial fluids concentrated above the difference of means for EO-BF# (Supplementary Figure 1H).

WBC abnormal scattergram flags

The flagging rate of 6.06 % (20/330) and 8.79 % (29/330) for XN and XR respectively, were comparable (p=0.1869). The discordance rate was 4.5 %. Correlation analysis of XN and XR analyser sample measurement pairs revealed closely comparable values (Spearman coefficient, slope, mean difference) for the group of samples where (1) no flag was triggered on both analysers (n=278), (2) both analysers flagged the same sample (n=17) or (3) only XN (n=3) or XR (n=12) triggered a flag, for all potentially affected parameters (WBC-BF, MN, PMN, LY-BF, MO-BF, NE-BF, EO-BF, HF-BF and TC-BF) (Supplementary Table 3, Supplementary Figure 2).

BAL fluids

The correlation coefficients for the 62 samples were excellent both for diagnostic and research parameters: WBC-BF: 0.977, RBC-BF: 0.975, MN#: 0.975, MN%: 0.983, PMN#: 0.981, PMN%: 0.983 and TC-BF: 0.990, NE-BF#: 0.979, LY-BF#: 0.959, MO-BF#: 0.973, EO-BF#: 0.899, HF-BF#: 0.992 (Supplementary Figure 3). The largest deviation from the line of equality was observed for EO-BF# (slope 1.240) and HF-BF# (slope 0.859). On average, only four samples were beyond the limits of agreements in the Bland-Altman analysis for each parameter (data not shown).