A comparative evaluation of the analytical performances of Capillarys 2 Flex Piercing, Tosoh HLC-723 G8, Premier Hb9210, and Roche Cobas c501 Tina-quant Gen 2 analyzers for HbA1c determination

Introduction Haemoglobin A1c (HbA1c) is widely used in the management of diabetes. Therefore, the reliability and comparability among different analytical methods for its detection have become very important. Materials and methods A comparative evaluation of the analytical performances (precision, linearity, accuracy, method comparison, and interferences including bilirubin, triglyceride, cholesterol, labile HbA1c (LA1c), vitamin C, aspirin, fetal haemoglobin (HbF), and haemoglobin E (Hb E)) were performed on Capillarys 2 Flex Piercing (Capillarys 2FP) (Sebia, France), Tosoh HLC-723 G8 (Tosoh G8) (Tosoh, Japan), Premier Hb9210 (Trinity Biotech, Ireland) and Roche Cobas c501 (Roche c501) (Roche Diagnostics, Germany). Results A good precision was shown at both low and high HbA1c levels on all four systems, with all individual CVs below 2% (IFCC units) or 1.5% (NGSP units). Linearity analysis for each analyzer had achieved a good correlation coefficient (R2 > 0.99) over the entire range tested. The analytical bias of the four systems against the IFCC targets was less than ± 6% (NGSP units), indicating a good accuracy. Method comparison showed a great correlation and agreement between methods. Very high levels of triglycerides and cholesterol (≥ 15.28 and ≥ 8.72 mmol/L, respectively) led to falsely low HbA1c concentrations on Roche c501. Elevated HbF induced false HbA1c detection on Capillarys 2FP (> 10%), Tosoh G8 (> 30%), Premier Hb9210 (> 15%), and Roche c501 (> 5%). On Tosoh G8, HbE induced an extra peak on chromatogram, and significantly lower results were reported. Conclusions The four HbA1c methods commonly used with commercial analyzers showed a good reliability and comparability, although some interference may falsely alter the result.


Introduction
Haemoglobin A 1c (HbA 1c ), a major portion of the glycated haemoglobins, is formed by a nonenzymatic interaction of glucose with the N-terminal valine residue of the HbA β chain in two basic steps: first, glucose binds reversibly to haemoglobin (Hb) as an aldimine Schiff base (an intermediate, termed labile HbA 1c (LA 1c ), or Hb pre-A 1c ); and then, the aldimine is transformed via an Amadori rearrangement into an irreversible ketoamine (1). Because HbA 1c reflects a mean blood glucose level over 2-3 months (normal lifespan of red blood cells) with a low short-term variability, it is widely used in the management of diabetes to monitor long-term glycemic control and to assess the risk of developing complications (2,3). In the last several years, the American Diabetes Association (ADA) and other major organizations have endorsed the use of HbA 1c determination for diabetes screening and have suggested the value of 6.5% (48 mmol/mol) as a diagnostic cut-off (4,5).
Analytical methods for HbA 1c quantification have been available since the 1970s. Presently, a num- Wu X. et al. HbA 1c methods' performance analysis ber of methods are used: the capillary electrophoresis (CE) or ion-exchange chromatography method separates and determines HbA 1c from other Hb fractions based on charge differences; the boronate affinity chromatography method separates and quantifies glycated Hb from the nonglycated Hb based on the cis-diol group; and the immunoassay method uses antibodies to recognize the structure of the N-terminal glycated amino acids of the Hb β chain for quantification (6). As different methods for HbA 1c determination exhibit different characteristics and performances, over the past years, major efforts have been made by the National Glycohemoglobin Standardization Program (NGSP) and the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) to standardize HbA 1c determination. The IFCC reference system has been defined as the only valid anchor to implement a standardization of the measurement (7). However, variability between methods is still observed in the presence of other members of the haemoglobin family (e.g., LA 1c , fetal haemoglobin (HbF), and variants) or interfering substances in the samples (e.g., bilirubin, triglyceride, cholesterol, and drugs) (8)(9)(10)(11)(12)(13).
Due to clinical requirements and management demands, the reliability of different methods used to measure HbA 1c and their potential interchange ability represent a key feature in clinical practice. Therefore, the aim of this study was to evaluate the performances (precision, linearity, and accuracy), concordance (method comparison), and influence of the most frequent analytical interferences (bilirubin, triglyceride, cholesterol, LA 1c , vitamin C, aspirin, HbF, and haemoglobin E (HbE)) using the four systems. The results of this study will be helpful for laboratorians to be aware of the limitations of the methods and to select the appropriate one that is less likely to have interference.
The analyzers as well as the associated reagents used for this evaluation were used according to the manufacturers' instructions and calibrated only once according to their routine standard operating procedures prior to any sample analysis. Internal quality controls (both low and high levels) supplied by the manufacturers and purchased from Bio-Rad Laboratories, Hercules, CA, USA (740 Diabetes Controls, lot: 33861/33862) were measured along with the samples on the same day during this study. The same assessment test by the four systems was carried out on the same day by the same technician in our laboratory. The same assessment test by one system was carried out using the same batch accompanied with the same controls.

Samples
Whole blood samples (N = 157) and umbilical cord blood (N = 1) were collected in EDTA-containing tubes (2.0 mL, BD Diagnostics, Franklin Lakes, NJ, USA) and kept at 4 ± 2 ºC to be analyzed within 24 h. For precision, method comparison, and analytical interference of Hb variants, the samples were kept at -80 °C (14,15)  This study was approved by the Research and Ethics committee of our institution, and all participants signed their consents prior to the study.

Precision
The precision was evaluated in accordance with the EP15-A2 (16) protocol of the Clinical and Laboratory Standards Institute (CLSI). Two whole blood samples at levels of 5.0% (31 mmol/mol) and 8.5% (69 mmol/mol) by Premier Hb9210 were divided into five aliquots per level and frozen at -80 ºC. One aliquot of each level was thawed daily, and the HbA 1c value was analyzed three times per day during a period of five consecutive workdays (N = 15 per level). Precision was evaluated as the coefficient of variation (CV), which is calculated from the data series mean and standard deviation. The formulas used to calculate CV were as follows: where D is the total number of days (five), n is the total number of replicates per day (three), X di is the result of replicate i for day d, d X is the average of all results for day d, and X is the average of all results.

Linearity
A test for linearity was carried out in accordance with the CLSI protocol EP6-A 8 (20). Linearity was investigated by preparing six different ratios of samples (5:0, 4:1, 3:2, 2:3, 1:4, and 0:5) from two samples with HbA1c results of 4.2% (22 mmol/mol) and 14.9% (139 mmol/mol) by Premier Hb9210 containing the same total Hb concentration (129 g/L). Each sample was assayed in duplicate, and the means were used to examine the linearity. The theoretical HbA 1c values (calculated from the ratios of the mixed packed-cells with high and low HbA 1c values) and the measured values were compared. Polynomial regression analysis was performed for first-, second-, and third-order polynomials.

Accuracy
Four samples used in the method comparison were validated by the Shanghai IFCC Reference Laboratory using the IFCC HPLC/CE reference method (21). The measurements of HbA 1c were made in triplicate. The relative bias was calculated by each of the four system values against the IFCC reference method value for each sample. The proficiency testing acceptance limit ± 6% of College of American Pathologists (CAP) was set as the accuracy limit (relative bias was calculated by NGSP units) (7).

Method comparison
The correlation between systems was assessed by analyzing 93 samples representing a range of HbA 1c values 4.0-13.0% (20-119 mmol/mol) by Premier Hb9210. The boronate affinity HPLC method with the Premier Hb9210 instrument was used as the comparative method. The measurements of HbA 1c were made in triplicate, and the relative bias of each specimen was calculated from the observed value against the baseline value.
The measurements of HbA 1c were made in triplicate. The relative bias of each specimen was calculated from the observed value and the baseline value (NGSP units).

Hb variant
Interference with a Hb variant was tested in several samples containing the most frequent Hb variant HbA/E (N = 6). The boronate affinity HPLC method with the Premier Hb9210 instrument was used as the comparative method as it is not expected to be influenced by the presence of Hb variants (23). For each test method, the results were compared to those obtained using the comparison method.

Statistical analysis
Data analysis was performed using MedCalc version 14.8.1 (MedCalc Software, Ostend, Belgium). The method comparisons were performed by Passing-Bablok regression, and the differences between any two methods were presented in a Bland-Altman plot. For method comparison and analytical interferences, relative bias > ± 7% (calculated by NGSP units) was considered clinically significant (NGSP criterion) (7).

Precision
The within-run CVs at low and high HbA 1c concentrations on all four systems were less than 2.0% and 1.2% (with IFCC units), or 1.3% and 1.0% (with NGSP units), respectively ( Table 1). The total CVs at the low and high HbA 1c levels were less than 2.0% and 1.7% (with IFCC units), and 1.4% and 1.5% (with NGSP units), respectively (Table 1).

Linearity
The linear regression analysis for each of the four analyzers showed a line with a good correlation coefficient (R 2 > 0.99) over the entire range tested (Table 1).

Accuracy
The relative bias against the IFCC targets was less than ± 6% for all four systems tested (NGSP units), indicating a good accuracy of these methods ( Table 2).

Analytical interferences
Assessment of interferences is shown in Table 3 A glucose concentration less than 277.78 mmol/L (LA 1c < 10.9%) did not interfere with HbA 1c quantification on any of the four analyzers. In addition, no significant bias was observed with the results analyzed by the Capillarys 2FP analyzer in the presence of vitamin C up to 250 mg/mL. In contrast, using the Premier Hb9210 and Tosoh G8 analyzers, vitamin C concentrations of more than 50 and 150 mg/mL, respectively, interfered with the HbA 1c measurements.
Moreover, the HbA 1c value was falsely increased only with the Tosoh G8 analyzer at an aspirin concentration exceeding 26.64 mg/mL in the samples.
Compared with the native sample, on the Capillarys 2FP analyzer, significant biases were observed in samples with 10-15% HbF (relative bias > ± 7%). On the Premier Hb9210 analyzer, the results were not affected in the two groups of samples with 5 -10% and 10-15% HbF (relative bias < ± 7%). On the Tosoh G8 analyzer, the HbA 1c results were obtained by a correction via excluding the HbF peak from the total integrated area, resulting in an insignificant interference in the samples with HbF < 30% (relative bias < ± 7%). For all samples analyzed by the Roche c501 analyzer, the HbA 1c results were influenced by elevated HbF values > 5% (relative bias > ± 7% for all groups).
The Capillarys 2FP analyzer gave a perfect separation of the variant HbE; on the Tosoh G8 analyzer, an extra peak was present between the A 1c and A 0 peaks ( Figure 2). A rather good agreement was noticed between the Capillarys 2FP/Roche c501 and Premier Hb9210 analyzers (relative bias < ± 7%); meanwhile, a significant negative bias of HbA 1c values was observed systematically on the Tosoh G8 system in comparison with the Premier Hb9210 analyzer (relative bias > ± 7%).

Discussion
The data presented here are comparative evaluations of the analytical performance of four commonly used systems: Capillarys (15,(24)(25)(26)(27) and manufacturer claims.
In the HbA 1c comparison study, Passing-Bablock regression analysis highlighted a good correlation between any two methods. Moreover, the Bland-Altman plot showed that the HbA 1c values between any two of the four systems were in good agreement.
The measurement of HbA 1c by the Capillarys 2FP, Tosoh G8, and Premier Hb9210 analyzers was not subjected to common interferences such as bilirubin and triglycerides/cholesterol, and these findings confirmed the results presented previously (15,24,25,28). However, very high levels of triglycerides/cholesterol led to a falsely low HbA 1c value in the immunoassay using the Roche c501 system.  tions. The isoelectric points of LA 1c and its stable counterpart are similar, which may lead to little or no separation between them by some methods that rely on molecular charge for separation. In this study, no effect was reported on the quantification of HbA 1c by the two methods based on the principle of molecular charge. The current study showed that treatment with glucose up to 277.78 mmol/L in vitro, much higher than the concentration reported previously (15,24,29) and the manufacturer's claims, did not interfere with the HbA 1c measurement, suggesting that the blood LA 1c contents may not affect the quantification of HbA 1c by the four tested methods.
In our study, for the concentration of vitamin C or aspirin at a dose higher than clinical relevance, it is presumed that the concentration of vitamin C or aspirin in patient samples does not affect HbA 1c results tested by the four analyzers in clinical practice. Chronic ingestion of aspirin or vitamin C in high doses may promote acetylation of HbA 1c chains or inhibit Hb glycation, thereby affecting HbA 1c results (13,(30)(31)(32)(33). Thus, the effects of aspirin and vitamin C on HbA 1c assays are uncertain, and they may be of a biological nature rather than an analytical interference, although this needs to be investigated in a future study.
With the Capillarys 2FP instrument, at an increased proportion of HbF greater than 10%, which is in excess of that reported by Jaisson et al. (24), the separation of HbF and HbA 0 could not be accomplished, resulting in a false result. As the HbF amount was excluded from the total integrated amount through using the Tosoh G8 instrument software, it may not interfere with the HbA 1c measurement at a concentration < 30%, in agreement with a previous evaluation and NGSP reports (7,8). Elevated HbF > 15%, in agreement with boronate affinity methods reported by the NGSP, can affect the HbA 1c results using the Premier Hb9210 system, which might be due to a lower glycation rate of HbF compared with that of HbA (7). The HbA 1c concentrations can be misestimated with the Roche c501 instrument at a higher HbF level ( > 5% in our study). However, there are no manufacturer claims of HbF interference. According to the NGSP, HbF levels > 10-15% can interfere with the Tina-quant Gen2 assay (7). Elevated HbF levels can occur under some pathological conditions such as beta-thalassemia, delta/beta-thalassemia, or a hereditary persistence of HbF (HPFH). The four instruments would lead to inaccurate results using different concentrations of HbF; therefore, it is important for laboratories to consider this fact in areas of a high prevalence of thalassemia or HPFH.
Over 1000 different Hb variants have been discovered. Quantification of HbA 1c in the presence of an Hb variant is an analytical challenge in the clinical laboratory. The most common Hb variants worldwide in descending order of prevalence are HbS, HbE, HbC, and HbD. As the second most prevalent hemoglobinopathy worldwide, HbE is mostly found in the Far East and Southeast Asia; thus, it has been included in the present study (11). A good agreement of HbA 1c concentrations was observed between the Capillarys 2FP, Roche c501, and Premier Hb9210 analyzers; therefore, it was concluded that the HbA1c measurements on these three systems were not affected by the presence of the HbE variant, confirming the results reported previously (9,10,15,(23)(24)(25)28). However, on the Tosoh G8 instrument, the presence of HbE induces an extra peak on the chromatogram, which is not reportable as shown previously (10,23).
In summary, the four HbA 1c methods commonly used with commercial analyzers showed a good comparability and reliability, although some interference may impede the results.