Comparative analysis of hemoglobin a1c measurement methods: inter-device agreement and clinical impact Authors: Ergin Taşkin, Ebubekir Bakan, Dilek Taşkin, İrem Arslantürk Published: 2026 ABSTRACT AbstractGlycated hemoglobin (HbA1c) is widely recommended for the diagnosis and monitoring of diabetes mellitus (DM). Minimal changes in HbA1c concentration can influence clinical decisions, making accurate and interference-free measurement essential. To compare the analytical performance and inter-method agreement of three HbA1c measurement techniques—Capillary Electrophoresis, High-Performance Liquid Chromatography (HPLC), and Turbidimetric Immunoinhibition—used in routine clinical practice. Fifty EDTA-anticoagulated patient samples were consecutively analyzed using HPLC (Tosoh G8), Capillary Electrophoresis (Sebia Capillarys 3 Tera), and an autoanalyzer (Beckman DXC 700 AU). Results were stratified into four HbA1c ranges (4–6%, 6–8%, 8–10%, 10–12%) and analyzed using Pearson correlation, Bland–Altman, and Passing–Bablok regression. Capillary Electrophoresis and HPLC showed a high degree of agreement (r = 0.998, bias = 0.07). Capillary Electrophoresis vs. autoanalyzer also demonstrated high correlation (r = 0.997) but a larger bias (0.52) that increased at higher concentrations. HPLC vs. autoanalyzer yielded r = 0.997 and bias = 0.45, with proportional bias at high HbA1c levels. Passing–Bablok analysis confirmed concentration-dependent deviations for the autoanalyzer. Capillary Electrophoresis and HPLC provide highly consistent HbA1c results, while the autoanalyzer may introduce clinically relevant bias, particularly at higher HbA1c levels, which may influence clinical interpretation near commonly used decision thresholds. However, given the limited sample size, these findings should be interpreted with caution.Keywords: Glycated hemoglobin, Diabetes Mellitus, Chromatography, High pressure liquid, Electrophoresis, CapillarySubject terms: Biomarkers, Diseases, Medical research Introduction Diabetes Mellitus (DM) is a chronic metabolic disease that threatens public health worldwide. Its significance is greater in developing countries, as it affects working-age adults. DM is characterized by hyperglycemia and abnormalities in insulin secretion and function1. These abnormalities may result from impaired insulin secretion, resistance to peripheral insulin action, or both. In diabetic patients, chronic hyperglycemia acting in synergy with other metabolic abnormalities can cause damage to various organ systems. This leads to health complications that cause disability and threaten life, the most prominent being retinopathy, nephropathy, and neuropathy, and increases the risk of cardiovascular disease by 2 to 4 times2. The World Health Organization (WHO) reports that approximately 830 million people worldwide have diabetes, most living in low- and middle-income countries, and attributes 1.5 million deaths annually directly to diabetes. The prevalence and case numbers of diabetes have steadily increased over recent decades3. For diagnosed patients, glycated hemoglobin (HbA1c) measurement is widely used to assess blood glucose control. Today, many different methods are available for HbA1c measurement in clinical laboratories4. Some are complex and expensive5. In rural areas, access to such costly tests is limited, which delays diagnosis and treatment, thereby affecting patients’ quality of life. In this study, we aimed to compare the High-Performance Liquid Chromatography (HPLC), Capillary Electrophoresis, and Turbidimetric Immunoinhibition methods used in the clinical biochemistry laboratory for HbA1c measurement. Given the globally increasing prevalence of diabetes according to WHO data, we aimed to contribute to early diagnosis and treatment by evaluating local data from our region. While there have been various large studies that compared the different measurement techniques for HbA1c, many of these have been multicentric and have not put enough attention to local lab variability and inter-device reliability in practice. This study is intended to conduct an evaluation from a single center standpoint and compare three common techniques with their respective biases. Methods Consecutive analyses of HbA1c were carried out using the HPLC [Tosoh G8, Tosoh Corporation, Tokyo, Japan], an autoanalyzer [Beckman Coulter DxC 700 AU, Beckman Coulter Inc., Brea, CA, USA], and Capillary Electrophoresis [Sebia Capillarys 3 Tera, Sebia, Lisses, France] instruments in the Biochemistry Laboratory of Ağrı Training and Research Hospital. The use of all equipment was done according to the manufacturer’s instructions, and calibration was performed using certified reference materials traceable to IFCC guidelines. Daily internal quality controls were carried out using two levels of control material. The methodologies’ underlying principles are explained below: For the Tosoh G8 apparatus, hemoglobin fractioning is accomplished through cation-exchange HPLC since glycation changes the charge of hemoglobin and makes separation possible because of different retention times on the chromatographic column; For the Sebia Capillarys 3 Tera machine, hemoglobin fractionation is done via capillary electrophoresis since glycation changes the charge of hemoglobin and results in distinct migration within an electric field; and For the autoanalyzer device, immunoturbidimetry is used to analyze HbA1c, which requires the use of specific antibodies to detect glycated hemoglobin. This analytical study used blood samples collected in lavender-capped EDTA tubes from 50 patients. The samples were obtained from routine laboratory testing and analyzed retrospectively. Samples with hemoglobin variant flags were excluded due to potential interference with HbA1c measurement. The study population consisted of 50 patients (18 males and 32 females) with a mean age of 52.12 years. Clinical indications for HbA1c testing included routine diabetes screening and follow-up. Patients were divided into 4 groups Group 1: HbA1c 4–6%. Group 1: HbA1c 4–6%. Group 2: HbA1c 6–8%. Group 2: HbA1c 6–8%. Group 3: HbA1c 8–10%. Group 3: HbA1c 8–10%. Group 4: HbA1c 10–12%. Group 4: HbA1c 10–12%. Data were analyzed with SPSS 26.0. Statistical analyses were conducted using IBM SPSS Statistics 26.0 and MedCalc Statistical Software. To evaluate the relationship between different HbA1c measurement methods (Capillary Electrophoresis, HPLC, and autoanalyzer), Pearson correlation coefficients (r) were calculated. While Pearson correlation analysis was used to assess linear relationships, it does not reflect agreement between methods. Therefore, Bland–Altman and Passing–Bablok analyses were considered the primary tools for evaluating method agreement and interchangeability. Passing–Bablok regression was used, with 95% confidence intervals calculated for slopes and intercepts. Bland-Altman analyses were also performed to assess agreement between methods, calculating mean difference (bias), standard deviation (SD), and 95% limits of agreement (LoA). Results were presented in tables and graphs, and variations in differences across concentrations were interpreted. A statistical significance level of p < 0.05 was considered for all analyses. Ethics statement This study was conducted in accordance with the principles of the Declaration of Helsinki and relevant institutional regulations. The study protocol was reviewed and approved by the Ağrı İbrahim Çeçen University Scientific Research Ethics Committee (Decision No: 150, dated 25.04.2024). The study was performed using routine laboratory samples obtained for diagnostic purposes. Patient identities were anonymized and no additional sampling was performed. Therefore, informed consent was waived by the ethics committee. Results Although Pearson correlation analysis showed strong linear relationships between all methods, correlation alone is insufficient to assess agreement; therefore, method comparison was primarily evaluated using Bland–Altman and Passing–Bablok analyses. The Pearson correlation coefficient obtained between Capillary Electrophoresis and HPLC measurements was calculated as r = 0.998. This value may indicate an excellent linear relationship between the two methods. The measurement ranges of both methods were observed to be similar, and the minimum and maximum measurements between the methods were found to be highly consistent (Fig. 1A; Table 1). (A) Correlation analysis between HPLC and capillary electrophoresis. (B) Correlation analysis between the autoanalyzer and capillary electrophoresis. (C) Correlation analysis between the autoanalyzer and HPLC. Table 1Correlation coefficients between HPLC and capillary electrophoresis.MinimumMaximumCapillary Electrophoresis4.8013.70HPLC4.8013.80Correlation coefficient (r): 0.998. Correlation coefficients between HPLC and capillary electrophoresis. Correlation coefficient (r): 0.998. A similarly very high correlation was found between Capillary Electrophoresis and the autoanalyzer (r = 0.997). Minimum and maximum measurement values ranged from 4.80 to 13.70 for Capillary Electrophoresis and from 5.10 to 14.70 for the autoanalyzer. This strong relationship may indicate that measurements between the two devices are largely consistent (Fig. 1B; Table 2). Table 2Correlation coefficients between the autoanalyzer and capillary electrophoresis.MinimumMaximumCapillary Electrophoresis4.8013.70Autoanalyzer5.1014.70Correlation coefficient (r): 0.997. Correlation coefficients between the autoanalyzer and capillary electrophoresis. Correlation coefficient (r): 0.997. In the comparison between HPLC and the autoanalyzer, the Pearson correlation coefficient was found to be r = 0.997. The minimum and maximum measurements for HPLC were 4.80–13.80, while those for the autoanalyzer ranged from 5.10 to 14.70. These results indicate a very strong relationship between the two devices (Fig. 1C; Table 3). Table 3Correlation coefficients between the autoanalyzer and HPLC.MinimumMaximumHPLC4.8013.80Autoanalyzer5.1014.70Correlation coefficient (r): 0.997. Correlation coefficients between the autoanalyzer and HPLC. Correlation coefficient (r): 0.997. In the study, Bland–Altman analysis was performed to assess the agreement between measurement results from the three different devices. For each method pair, the mean difference (bias), 95% confidence intervals (CI), and limits of agreement (LoA) were calculated. In the Bland–Altman analysis between Capillary Electrophoresis and HPLC, the mean difference (bias) was found to be 0.07 (95% CI: 0.018–0.117). The lower and upper limits of agreement were calculated as − 0.27 (95% CI: − 0.353 to − 0.184) and 0.40 (95% CI: 0.318–0.488), respectively. The standard deviation was 0.17. The difference was very small, indicating high agreement between the methods. However, a limited number of measurements fell outside the 95% limits of agreement, indicating occasional deviations (Fig. 2A; Table 4). (A) Bland–Altman analysis between HPLC and capillary electrophoresis. (B) Bland–Altman analysis between the autoanalyzer and capillary electrophoresis. (C) Bland–Altman analysis between the autoanalyzer and HPLC. Table 4Bland–Altman analysis results for HPLC and Capillary Electrophoresis.MinimumMaximumCapillary Electrophoresis4.813.7HPLC4.813.8(Capillary Electrophoresis + HPLC) / 24.8513.7ParameterEstimate95% CISEMean difference0.070.018 to 0.1170.02495% Lower LoA–0.27–0.353 to − 0.1840.04295% Upper LoA0.40.318 to 0.4880.042SD0.17 Bland–Altman analysis results for HPLC and Capillary Electrophoresis. In the comparison between Capillary Electrophoresis and the autoanalyzer, the mean difference (bias) was found to be 0.52 (95% CI: 0.448–0.585). The lower and upper LoA values were 0.05 (95% CI: − 0.071 to 0.166) and 0.99 (95% CI: 0.867–1.103), respectively. The standard deviation was calculated as 0.24. The relatively high mean difference may indicate a systematic difference between these two methods. Furthermore, graphical analysis showed a positive trend—differences increased with higher concentrations—suggesting that device disagreement may be greater at elevated HbA1c values (Fig. 2B; Table 5). Table 5Bland–Altman analysis results for the autoanalyzer and capillary electrophoresis.MinimumMaximumCapillary Electrophoresis4.813.7Autoanalyzer5.114.7(Capillary Electrophoresis + Autoanalyzer) / 2514.15ParameterEstimate95% CISEMean difference0.520.448 to 0.5850.03495% Lower LoA0.05–0.071 to 0.1660.05995% Upper LoA0.990.867 to 1.1030.059SD0.24 Bland–Altman analysis results for the autoanalyzer and capillary electrophoresis. In the comparison between HPLC and the autoanalyzer, the mean difference (bias) was found to be 0.45 (95% CI: 0.368–0.530). The lower LoA was − 0.10 (95% CI: − 0.244 to 0.035) and the upper LoA was 1.00 (95% CI: 0.863–1.142). The standard deviation was calculated as 0.28. Measurement differences showed a positively increasing trend, with particularly pronounced deviations between the devices at higher concentrations (Fig. 2C; Table 6). Table 6Bland–Altman analysis results for the autoanalyzer and HPLC.MinimumMaximumHPLC4.813.8Autoanalyzer5.114.7(HPLC + Autoanalyzer) / 24.9514.25ParameterEstimate95% CISEMean difference0.450.368 to 0.5300.0495% Lower LoA–0.10–0.244 to 0.0350.06995% Upper LoA10.863 to 1.1420.069SD0.28 Bland–Altman analysis results for the autoanalyzer and HPLC. To evaluate linear agreement in inter-device method comparisons, Passing–Bablok regression analysis was applied. For each device pair, the regression equation, slope, and intercept were calculated along with 95% bootstrap confidence intervals (CI). With regard to the capillary electrophoresis vs. HPLC comparison, the slope confidence interval contained the number 1, which shows that there was no proportional bias. On the other hand, the intercept confidence interval did not contain 0, which indicated the existence of a constant bias between the two analytical techniques. This means that even though the two techniques compare well, a difference exists between them. (Fig. 3A; Table 7). (A) Passing–Bablok analysis between HPLC and capillary electrophoresis. (B) Passing–Bablok analysis between the autoanalyzer and capillary electrophoresis. (C) Passing–Bablok analysis between the autoanalyzer and HPLC. Table 7Passing–Bablok analysis results for HPLC and capillary electrophoresis.MinimumMaximumCapillary Electrophoresis4.813.7HPLC4.813.8 Passing–Bablok fit Equation: HPLC = 0.3029 + 0.9714 × Capillary ElectrophoresisParameterEstimateBootstrap 95% CIIntercept0.30290.10000 to 0.4762Slope0.97140.9524 to 1.0000CI based on 999 bootstrap samples. Passing–Bablok analysis results for HPLC and capillary electrophoresis. CI based on 999 bootstrap samples. In the comparison between Capillary Electrophoresis and the autoanalyzer in this analysis, the intercept’s confidence interval included 0, indicating that constant bias was negligible. However, the slope’s confidence interval did not include 1, suggesting the presence of proportional bias. Furthermore, it was observed that the difference increased with concentration. This suggests that agreement between the devices may weaken, particularly at higher concentrations. This suggests that interchangeability between the methods may be limited. These findings support the positive trend observed in the Bland–Altman analysis (Fig. 3B; Table 8). Table 8Passing–Bablok analysis results for the autoanalyzer and capillary Electrophoresis.MinimumMaximumCapillary Electrophoresis4.813.7Autoanalyzer5.114.7 Passing–Bablok fit Equation: Autoanalyzer = 0.1239 + 1.049 × Capillary ElectrophoresisParameterEstimateBootstrap 95% CIIntercept0.1239–0.1959 to 0.3945Slope1.0491.016 to 1.082CI based on 999 bootstrap samples. Passing–Bablok analysis results for the autoanalyzer and capillary Electrophoresis. CI based on 999 bootstrap samples. In the comparison between HPLC and the autoanalyzer the analysis showed that the intercept’s confidence interval included 0, indicating no significant constant bias. However, the slope’s confidence interval did not include 1, revealing a proportional difference between the two methods. It was noted that measurement differences between HPLC and the autoanalyzer increased particularly as concentration rose, which may reduce agreement between the devices. (Fig. 3C; Table 9). Table 9Passing–Bablok analysis results for the autoanalyzer and HPLC.MinimumMaximumHPLC4.813.8Autoanalyzer5.114.7Equation: Autoanalyzer = − 0.216 + 1.08 × HPLCParameterEstimateBootstrap 95% CIIntercept–0.2160–0.4459 to 0.03333Slope1.081.048 to 1.111CI based on 999 bootstrap samples. Passing–Bablok analysis results for the autoanalyzer and HPLC. CI based on 999 bootstrap samples. Discussion In this study, we evaluated the agreement and biases between three different analytical methods—Capillary Electrophoresis, High-Performance Liquid Chromatography (HPLC), and an autoanalyzer—for measuring HbA1c levels, which is one of the key parameters widely used in the diagnosis and monitoring of diabetes. Our aim was to assess the agreement and discrepancies between these methods both statistically and in terms of potential clinical relevance. In our findings, the high correlation coefficient (r = 0.998) and low mean difference (bias = 0.07) observed between Capillary Electrophoresis and HPLC suggest a high degree of agreement between the two methods. This result appears to be consistent with a 2015 comparative study, in which a statistically significant and strong correlation (R² = 0.99) was reported between Capillary Electrophoresis and two different HPLC systems. That study also noted that certain hemoglobin variants could cause deviations in some HPLC systems, while Capillary Electrophoresis was more successful at separating these variants4,6. In our study, the constant bias of approximately 0.3 units observed in the regression analysis suggests a small difference between the methods that may be clinically relevant in certain situations. Our results are also supported by a multicenter study conducted in China that examined the effect of five different hemoglobin variants on HbA1c measurements7. In that study, HbA1c values obtained by HPLC were found to be significantly lower than those measured by Capillary Electrophoresis, suggesting that hemoglobin variants may influence HbA1c measurement results in HPLC systems, which may influence clinical interpretation during diabetes management5,7. Similarly, Yadav and Mandal reported that Hb variants could interfere with HbA1c measurements and noted that while ion-exchange chromatography methods may be affected by genetic variants, mass spectrometry was not, suggesting that it may serve as a useful analytical approach for HbA1c assessment. They emphasized that due to differences in glycation profiles of variant hemoglobins, different elution patterns could contribute to variability in HbA1c measurements8. When comparing Capillary Electrophoresis with the autoanalyzer, although the correlation coefficient was again high (r = 0.997), the mean difference was significantly larger (bias = 0.52), and the difference increased with higher concentrations. This suggests that inter-device differences may become more apparent at higher HbA1c ranges. A multicenter study comparing Capillary Electrophoresis, enzymatic, and turbidimetric methods to HPLC found statistically significant but clinically negligible differences, with Capillary Electrophoresis producing values closest to HPLC. However, the same study stressed that achieving standardization in HbA1c measurement may contribute to more consistent clinical interpretation and follow-up5,9,10. The apparent tendency toward positive bias seen in the immunoturbidimetric assay, especially at higher glycemic values, can be explained by non-specific interferences and matrix effects characteristic of latex-based procedures. This is because these techniques employ an immune complex reaction between antigen and antibodies; therefore, antibody specificity and cross-reactivity with other glycated hemoglobin forms, along with certain physiological conditions like uremia and hyperlipidemia, may cause a positive bias in uncontrolled diabetes patients. The performance of capillary electrophoresis (CE) can be attributed to its high resolving power, which enables separation of HbA1c from other hemoglobin fractions through electrophoretic mobility and electro-osmotic flow through electrophoretic mobility and electro-osmotic flow. Although CE demonstrated good analytical agreement, the use of CE in practical applications could be restricted because of its relatively high cost and lengthy time for analysis, unlike the more affordable turbidimetric approach. While previous large-scale studies have included larger cohorts, our single-center study supports these findings within our own patient population and highlights the need for standardized calibration across different analytical platforms. A key strength of our study is the direct comparison of three commonly used methods—HPLC, capillary electrophoresis (CE), and immunoturbidimetry allowing us to evaluate method-specific biases in a real clinical setting11,12. Even small differences in HbA1c values may influence clinical interpretation, particularly around the commonly used 6.5% diagnostic threshold, where variability between methods may become more noticeable. In addition, previous studies have shown that high correlation between methods does not necessarily mean good clinical agreement13,14. Therefore, method-dependent bias may need to be considered when monitoring patients over time or when changing analytical methods. The upward trend observed in Bland–Altman analysis further suggests that device differences become more pronounced at higher HbA1c levels, which may influence the interpretation of treatment response during diabetes follow-up. Small differences between methods are often clinically negligible; however, at high HbA1c levels, inter-method variability may influence clinical interpretation. Standardization and quality control programs may help improve inter-method agreement. Although there is generally high agreement and low bias between different HbA1c measurement methods, method-dependent differences may become more apparent in the presence of hemoglobin variants and at higher HbA1c levels. Method selection and standardization may contribute to more reliable HbA1c interpretation in diabetes management. In light of all these findings, the selection of HbA1c measurement methods may have implications not only for analytical accuracy but also for clinical interpretation. Previous studies suggest that capillary electrophoresis may provide improved variant separation with reduced analytical interference15. In this study, Capillary Electrophoresis and HPLC appear to provide a high degree of agreement in HbA1c measurement. In contrast, measurements obtained with the autoanalyzer suggest method-dependent differences that may influence clinical interpretation, particularly at higher HbA1c levels. Consistent with the literature, our findings highlight the need for verification of inter-device agreement using standardized methods, particularly near commonly used diagnostic thresholds, and suggest that careful interpretation may be warranted when switching between analytical platforms. Previous studies suggest that Capillary Electrophoresis systems may offer advantages in terms of sensitivity and reduced susceptibility to hemoglobin variant interference (15). Nonetheless, it is recommended that each laboratory conduct its own method validation and reference interval studies tailored to its patient population. There are some limitations associated with our research. One limitation of this study is the relatively small sample size (n = 50). Thus, our findings may not be applicable to larger samples with various hemoglobin types. Future studies should be conducted to further evaluate inter-method differences in larger cohorts. Author contributions E.T. wrote the main manuscript text. E.B. provided language support for the article. D.G. She provided biochemical samples. İ.A. She edited the article. All authors reviewed the manuscript. Data availability The datasets generated and/or analyzed during the current study are not publicly available due to patient confidentiality but are available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The datasets generated and/or analyzed during the current study are not publicly available due to patient confidentiality but are available from the corresponding author on reasonable request.