Method comparison between Alinity hq and Sysmex XT-4000i in an emergency laboratory

Abstract Due to technological advancements, haematology analysers are becoming increasingly more complex. Before introducing new analyzers, laboratories must compare the agreement between the new and the old instruments. This study aimed to quantify the method agreement between Sysmex XT-4000i and Alinity hq analysers in order to establish whether they can be used interchangeably. A total of 415 complete blood counts (CBC) from adult patients of the Emergency Clinical County Hospital of Târgu Mureș, Romania, were analysed within 4 h from the collection on Sysmex XT-4000i (considered the reference method), then on Alinity hq. Statistical analysis consisted of outlier removal, Spearman Correlation, Bland–Altman test, and Passing–Bablok regression. For each CBC parameter, the analytical difference between methods was compared with the Reference Change Value (RCV) at medical decision levels (MDL). Despite using different technologies, the instruments have a good agreement regarding cell differentiation and counting. Cell counting and haemoglobin measurement showed a good agreement at all (Medical Decision Limits) MDLs. The analytical difference between methods surpassed the (Reference Change Value) RCV with 1.2% at the 14% MDL of HCT and with 0.2% at the 100 fL MDL of MCV. This study can not tell whether Sysmex or Alinity is superior, only if the two methods agree. The poorer agreement observed for RBC indices, especially MCHC, suggests an accumulation of differences caused by the different working principles of the two methods. However, it is reasonable to assume that such small differences will not affect clinical decision-making and patient outcome.


Introduction
Comparability studies between two instruments that perform the same test are regulated in the CLSI EP9-A3 Guideline [1]. Correlation and linear regression remain the traditional statistical tools for method comparison, but other tests can also be used.
The correlation defines a linear relationship between two variables. A high correlation does not necessarily indicate a good method agreement. The coefficient of correlation (r) ranges from À1 to 1. The closer coefficients are to À1 or þ1, the higher the strength of the correlation. The r value can only be interpreted if the p value reaches the statistical significance threshold [2].
The regression model described by Passing and Bablok calculates the slope of the regression line as the median of all possible slopes and is therefore robust towards outliers. The test assumes that the measurements of the two methods have a positive correlation and a linear relationship [3]. Passing-Bablok regression has two types of results: visual (scatter diagram) and numerical (regression equation). The regression equation is expressed as y ¼ aþxb, where 'a' represents the constant difference (intercept) and 'b' represents the proportional difference (slope). The significance of the test is given by the confidence intervals for a and b: if the intercept CI includes the value 0 and the slope CI includes the value 1, there is no significant difference between methods [4].
The Bland-Altman plot describes the agreement between two quantitative measurements, generating agreement limits [5]. The scatter plot's X axis shows the average of two paired measurements and the Y axis represents the difference between the same paired measurements. The visual examination of the plot allows the determination of the global agreement between the two methods. The Bland-Altman analysis is able to quantify the bias and compute an interval of agreement. The bias can be considered significant if the equity line lies outside the confidence interval of the mean difference. The normal distribution of the differences is mandatory for calculating relevant limits of agreement [6].
Reference Change Value (RCV) is used in monitoring serial results from a patient to determine if the change in value is significant and it is designed to be calculated on one analyzer at a time [7]. Although RCV was designed to guide a more patient-centered clinical decision-making, in this study we wanted to test its applicability to method agreement.

Working principle of the two analysers
To contextualise the results of this study, it is imperative to acknowledge the different working principles of the two analysers. Sysmex XT-4000i (Sysmex, Japan), the routine instrument, uses fluorescence flow cytometry to analyse cell complexity and impedance detection method and hydrodynamic focusing to differentiate and count red blood cells and platelets. For the 6-DIFFwhite blood cell differentiation, fluorescent technology is used, more precisely side scatter, forward scatter, and fluorescence. Haemoglobin is measured spectrophotometrically with a non-cyanide reagent. Haematocrit is obtained by accumulating pulse heights of counted RBC [8].
Alinity hq (Abbott, Chicago, USA), the newly introduced analyzer, has MAPSS technology (multi-angle polarised scatter separation) with 8 scattered light detectors to gain information on number, volume, and cell content (nucleus segmentation, cytoplasmic granulations, content of haemoglobin). The instrument detects haemoglobin concentration and measures the mean corpuscular volume (MCV), therefore haematocrit is a calculated value [9].
The aim of this study was to investigate the level of agreement between two haematology analysers, Sysmex XT-4000i and Alinity hq, in order to establish whether they can be used interchangeably in the laboratory.

Material and method
This comparison-of-method study was performed between December 2020 and February 2021 in the Emergency Laboratory of the County Emergency Clinical Hospital of Targu Mures, Romania.

Samples
The study included 415 whole blood samples for complete blood count (CBC) from hospitalized adult patients. Blood was collected in K2 EDTA tubes, stored at room temperature (18-25 C) and processed within 4 h from collection. The following exclusion criteria were applied: incorrect specimen collection, handling or transportation, suspicion or presence of clots. Batches of 20-30 samples were first analysed using the Sysmex XT-4000i (software version 00-18) instrument and then with Alinity Hq (Software version: 4.3).

Quality control
Quality control was performed on 3 levels with commercial whole blood control materials according to the manufacturer's instructions [10]. Calibration of both analyzers was performed with reference materials provided by each manufacturer according to the manufacturer's instructions.

Statistical analysis
The database was created in Excel spreadsheets. Statistical analysis was performed with MedClac Software for Windows. Data distribution was analysed with the Kolmogorov-Smirnov test. For method comparison, Spearman correlation coefficient (r), Bland-Altman plot and Passing-Bablok regression were performed. The statistical significance threshold was set 0.05.
Reference change values (RCV) were computed using the online calculator provided by the European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) [11]. RCV formula contains variables describing both the preanalytical and the analytical phases (preanalytical imprecision -CV P , analytical imprecision -CV A ) and withinsubject biological variation (CV I ).
where Z value was set at 1.96 for 95% confidence [7]. Considering the study design and because the same sample was analysed on both instruments, the preanalytical imprecision is null [7]. Hence, the RCV was calculated with the following formula: For each hypothetically medical decision level (MDL) [12] that would be analysed on the Sysmex instrument, a corresponding value was calculated for the Alinity instrument using the Passing-Bablok regression equation. Then, the analytical difference (D A ) between the two methods was calculated and compared with the RCV. The RCV is designed for monitoring patients on a single analyzer, therefore the coefficient of variation from Sysmex instrument was chosen for calculating the RCV for the investigated CBC parameters at their specific medical decision levels.
The study was approved by the Ethics Committee of the County Emergency Clinical Hospital of Targu Mures.

Results
Outliers were identified with the Grubbs test and removed.  Tables 1 and 2. RCVs were calculated for each QC level of the CBC parameters using the CV A of Sysmex XT-4000i (Supplementary material - Table S1). The RCVs shown in Table 3 for each MDL correspond to the QC level closest to that MDL.

Discussion
The variety of the technologies developed by the manufacturers in the haematological analysers should not interfere with clinical decision making, but support it for the patients' benefit. In an emergency laboratory, with a 24/7 working program, the utilization of two CBC analysers improves the workflow, especially during maintenance procedures. It is important to recognize that each technology has its limitations and operating on two different haematology analysers at the same time can be challenging. The real advantage can be seen when complex cases appear and the differences between analysers become distinct points of view enlarging the perspective on the case. To profit these aspects, it is mandatory for the instruments to function in harmony with one another, and not alter the decision making. In our study Sysmex XT-4000i and Alinity hq have a good agreement regarding cell differentiation and counting. Cell counting and haemoglobin measurement showed a good agreement at all MDLs. There are some analytical differences between methods that surpassed the RCV at some MDLs for HCT, MCV and MCHC.

Spearman linear correlation
Two methods that measure the same sample should have a high correlation. As expected, all but one CBC parameters had a very high, statistically significant correlation. The only exception was MCHC which showed a moderate albeit statistically significant correlation. Similar results were found in the literature: a method comparison study between Alinity hq and another analyser from Sysmex (Sysmex XN-9000) reported fair to good correlation coefficients for all CBC parameters (>0.85, mostly > 0.95) except for MCHC which had a moderate correlation (r ¼ 0.68) [13]. However, a lower correlation for MCHC does not seem to be specifically related to Alinity-Sysmex method comparison studies: Alinity hq vs Cell-Dyn Sapphire (MCHC correlation  coefficient: 0.65) [14], Roche Cobas M511 vs Sysmex XN (MCHC correlation coefficient: 0.56) [15].

Bland-Altman analysis
In this study, Bland-Altman analysis showed small, statistically non-significant mean differences and, therefore, a good agreement for all analysed CBC parameters. The method agreement interval is calculated with standard deviation of the differences between methods. Therefore, we acknowledge that the new data set (the differences between methods) is better characterized by the median value and standard error and not by the mean value and standard deviation. Visual analysis of the Bland-Altman scatter plots revealed that, for WBC and PLT, the regression line of differences showed an ascending trend which exceeds the CI of the mean of differences. However, given that this only happened at values of about 65 Â 10 9 /L for WBC and 1250 Â 10 9 /L for PLT, these differences have no clinical implications. Trends shown by the regression line could also be observed for other parameters, but these variations occurred within the CI of the mean of differences and are less important: ascending trends for RBC, HGB, and MCH; descending trends for MCV and HCT. The plot for MCHC revealed an abrupt and ascending trend as values are greater. As such, compared with Sysmex XT-4000i, Alinity hq will calculate higher values for the lower part of the MCHC range, and lower values for the upper part of the MCHC range.

Passing-Bablok regression
For WBC, the regression equation indicated a constant difference of À0.025 Â 10 9 /L and a slope of 0.998, but none of them were statistically significant. The regression had no significant deviation from linearity. The data indicated a good agreement between methods for counting white blood cells. The linearity of the RBC equation did not differ significantly (slope 1.008) from the identity line. Although statistically significant, such small constant differences (0.004 Â 10 12 /L) in RBC are irrelevant in clinical practice; overall, the agreement between methods is acceptable. Platelet counting (PLT) on the studied haematology analyser appears to have a significant constant deviation of 3.2 Â 10 9 /L and a non-significant proportional deviation of 0.979. There was no significant deviation from linearity. Despite using different technologies, Alinity hq (flow cytometry) has a good agreement with Sysmex XT-4000i (flow cytometry and impedance) regarding cell differentiation and counting. The thrombocyte counting and differentiating performances of Alinity hq have been recognised in other studies. One such study compared PLT results from Alinity with the reference method and a good agreement was reported [16].
Both Alinity and Sysmex analysers use the cyanide-free method for the haemoglobin concentration assay. According to the Passing-Bablok regression equation, there is a significant deviation from linearity, statistically significant constant error (intercept of À0.28 g/dL), and proportional error (slope of 1.01). Inspecting the scatter diagram, it can be observed that the main deviation from the identity line is for the lower values of HGB.
Haematocrit is quantified differently by the two analyzers: Sysmex measures it directly by accumulating the impedance peaks of RBC's while Alinity calculates it from erythrocytes count and mean corpuscular volume. The regression equation for HCT showed a constant error of 2% and a slope of 0.866, both statistically significant, but with no significant deviation from linearity. The clinical implications of these differences require further investigation.
RBC indices are important tools for screening, diagnosing and monitoring anemias. Analysing the linearity of these indices' regression lines, we found a statistically significant deviation from linearity for MCHC. The constant error was statistically significant for all indices, and the proportional error was statistically significant only for MCV and MCHC. MCV is directly measured on Alinity and used for calculating the HCT and, eventually, the MCHC. Hence, the significant differences seen for MCHC are in fact an accumulation of differences caused by the distinctive working principles of the two analyzers regarding MCV and HCT.

Reference change value
It is of great importance to statistically assess how narrow the agreement interval needs to be for two methods to be considered interchangeable. Nevertheless, such undertaking requires interdisciplinary effort because how big a difference would be acceptable is a matter of clinical judgement [17].
MDLs are laboratory thresholds of clinical importance. Whether laboratory values are above or below the MDL can impact the confirmation or exclusion of a diagnosis, clinical decision making, and patient outcome [18]. For these MDLs the laboratories must be rigorous regarding analytical performances. In our study, the comparison of analytical differences (D A ) between methods and RCVs at MDLs showed a good agreement for cell counting and haemoglobin measurement. The same level of good agreement was not achieved for HCT and RBC indices. As stated above, these differences are most likely a result of the different working principles of the two analyzers. The HCT D A surpassed the RCV by 1.2% at the MDL of 14%. Such situations are rarely seen in clinical practice and, if present, clinical decision making does not rely on HCT values alone, but on other CBC parameters as well (RBC, HGB), for which the D A were within the RCV in this study. Similarly, the MCV D A surpassed the RCV with 0.2% at the MDL of 100 fL, but such small differences are unlikely to affect clinical decision making and patient outcome.

Conclusion
Although in this study the reference method was considered to be Sysmex XT-4000i, it is important to discern that the study evaluated the agreement of methods, not the performances of the two analyzers. This study can not tell whether Sysmex or Alinity is better. The differences found mainly for RBC indices suggest that they are a consequnce of the different technologies used by the two analyzers, and even if these differences surpass the RCV, it is reasonable to assume that such small differences will not affect the clinical decision and patient outcome.

Disclosure statement
No potential conflict of interest was reported by the author(s).