Stability of glycated haemoglobin (HbA1c) measurements from whole blood samples kept at −196°C for seven to eight years in The Malaysian Cohort study

Abstract

Objective

Glycated haemoglobin (HbA1c) is a standard indication for screening type 2 diabetes that also has been widely used in large‐scale epidemiological studies. However, its long‐term quality (in terms of reproducibility) stored in liquid nitrogen is still unknown. This study is aimed to evaluate the stability and reproducibility of HbA1c measurements from frozen whole blood samples kept at −196°C for more than 7 years.

Methods

A total of 401 whole blood samples with a fresh HbA1c measurement were randomly selected from The Malaysian Cohort's (TMC) biobank. The HbA1c measurements of fresh and frozen (stored for 7–8 years) samples were assayed using different high‐performance liquid chromatography (HPLC) systems. The HbA1c values of the fresh samples were then calculated and corrected according to the later system. The reproducibility of HbA1c measurements between calculated‐fresh and frozen samples was assessed using a Passing‐Bablok linear regression model. The Bland–Altman plot was then used to evaluate the concordance of HbA1c values.

Results

The different HPLC systems highly correlated (r = 0.99) and agreed (ICC = 0.96) with each other. Furthermore, the HbA1c measurements for frozen samples strongly correlate with the corrected HbA1c values of the fresh samples (r = 0.875) with a mean difference of −0.02 (SD: −0.38 to 0.38). Although the mean difference is small, discrepancies were observed within the diabetic and non‐diabetic samples.

Conclusion

These data demonstrate that the HbA1c measurements between fresh and frozen samples are highly correlated and reproducible.

Glycated haemoglobin (HbA1c) has been used as a risk marker in various epidemiological studies. However, the quality of HbA1c measurements stored in liquid nitrogen for the long term is still unknown. This study evaluated the reproducibility of HbA1c from the frozen whole blood samples kept at −196°C for more than 7 years. We found a slight variation among the HbA1c measurements of fresh and frozen samples, where the frozen samples measurement are 0.04% lesser than the fresh samples. Still, the HbA1c measurements were correlated and reproducible. The graphical abstract was created with BioRender.com.

1. INTRODUCTION

Type 2 diabetes (T2D) is a complex disease influenced by genetic, environmental risk factors and the interplay of both. ¹ , ² , ³ Lifestyle and diet transition and modernization contributed to the increasing prevalence of T2D globally and locally. ⁴ T2D is predicted to affect 9.4% of the population worldwide by 2030. ⁵ In Malaysia, the burden of T2D increased substantially, with the overall prevalence elevated from 11.2% to 18.3% within 9 years. ⁶

T2D can be diagnosed by several methods, such as fasting plasma glucose and glycated haemoglobin (HbA1c). ⁷ HbA1c, a glucose‐bound red blood cell, can reflect the average blood glucose concentration over the past two to three months. Hence, it was regarded as the standard indication for screening and diagnosing T2D. A person with HbA1c ≥6.3% (IFCC ≥45 mmol/mol) is diagnosed with diabetes according to the Malaysian Clinical Practice Guideline. ⁸

HbA1c was typically measured using fresh blood samples to avoid sample degradation after storage. Although several studies reported the effects of storage temperature and duration on HbA1c's stability, the reproducibility results were inconsistent and mostly kept for a relatively short time at −80°C. ⁹ , ¹⁰ , ¹¹ , ¹² To date, only two studies reported a slightly higher HbA1c measurement for whole blood samples stored for over a decade. ¹³ , ¹⁴

The emergence of biobanks to store samples for a long period with a combination of clinical data as a big data platform promotes a new perspective on understanding the molecular mechanism of diseases in terms of predictive, prognostic biomarkers, therapeutic, and personalized medicine. ¹⁵ This research usually requires comparing biomarkers before and after disease onset. However, the quality of the long storage samples and optimal storage conditions in regard to liquid nitrogen is still unknown. Hence, ensuring the quality and reproducibility of long‐term storage samples is crucial. Reproducibility usually refers to the discrepancy in measurements made on a subject under changing conditions, such as using different methods or instruments, measuring by different observers, or measurements made over time. ¹⁶ Thus, measurement error analysis is essential to identify the quality and stability of the HbA1c sample stored for a long period and under a lower temperature. In this study, we evaluate the quality of archived frozen whole blood samples kept in liquid nitrogen tanks (at −196°C) for over 7 years by assessing the stability of HbA1c measurement across the two‐time period with the fresh samples.

2. MATERIALS AND METHODS

2.1. Subjects

A total of 401 whole blood samples with HbA1c information collected between 2012 and 2013 were randomly selected from The Malaysian Cohort (TMC) biobank, a prospective multi‐ethnic cohort study comprising volunteers aged between 35 and 70 years. The details of the TMC project have been described elsewhere. ¹⁷ Briefly, the biospecimen (like blood and urine) and the data on sociodemographic characteristics, medications, and lifestyle were collected from the subjects who consented to participate in the study. To date, the study has successfully had a follow‐up of 42.7% of its participants. ¹⁸ In this study, as many as 191 out of 401 participants were diagnosed with diabetes using fasting blood glucose (FBG ≥7.00 mmol/L) and/or having a medical history of diabetes with medication.

All relevant ethical approvals for TMC project were approved by the Institutional Review and Ethics Board of Universiti Kebangsaan Malaysia (Project Code: FF‐205‐2007) in accordance with the Declaration of Helsinki. In addition, written informed consent was also received from the subject's prior participation in the study.

2.2. Blood sampling and storage

Fasting venous blood was collected in two sterile vacutainer tubes: 3 mL in dipotassium‐EDTA (Becton Dickinson) and 6 mL in acid citric dextrose, ACD (Becton Dickinson), respectively. From the ACD Vacutainer, 700 μL of blood samples were transferred into cryogenic tubes, with an additional 120 μL of Corning™ cellgro™ RPMI 1640 media (Thermo Fisher Scientific) and 30 μL of Hybri‐Max™ DMSO (Sigma‐Aldrich). The mixed whole blood samples in a cryotube were kept in a cryo‐freezing container (Thermo Fisher Scientific) overnight before being stored in a liquid nitrogen tank (−196°C) to avoid peptide degradation and potential interference.

2.3. HbA1c measurement

As for the original protocol for samples in 2012 and 2013, fresh whole blood samples from EDTA tubes were measured for HbA1c using Premier Hb9210 HbA1c analyzer (Trinity Biotech, WW, Ireland) that utilizes boronate affinity technology for high‐performance liquid chromatography (HPLC). ⁷ The quoted machine's reportable values range between 4% and 18% HbA1c. ¹⁹

For this current study (in 2020 and 2021), we retrieved the whole blood samples from the same participants we measured in 2012 and 2013 stored at −196°C in a liquid nitrogen tank at TMC's biobank. These blood samples were aliquoted and frozen from the ACD tube in 2012 and 2013. The frozen samples were thawed at a four‐degree refrigerator and brought to room temperature before re‐measurement of the HbA1c concentration using Variant™ II Turbo analyzer (Bio‐Rad Laboratories). Variant™ II Turbo analyzer employs the cation exchange HPLC method with a measurable range of 3.4%–20.6% HbA1c. ²⁰

Internal quality control (IQC) of the HbA1c assay was performed daily. For Premier Hb9210, IQC is performed using a default build‐in program, while for Variant™ II Turbo, IQC was performed using Bio‐Rad Liquicheck™ Diabetes Controls: Level 1 (low; contains hemolysate) and Level 2 (high; contains hemolysate with purified A1c). Apart from that, external quality control was performed monthly using the Royal College of Pathologists of Australasia (RCPA) glycohemoglobin proficiency test. ²¹ In addition, all instruments used in this study were certified by the National Glycohemoglobin Standardization Program (NGSP) and the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) in 2020 and 2021, as well as standardized to the method used by the Diabetes Control and Complications Trial (DCCT). All assays were performed according to the manufacturer's instructions in an accredited bioanalytical laboratory. ⁷

Some studies used both boronate affinity and cation exchange HPLC systems as the reference when assessing HbA1c concentration. ²² , ²³ The UK Biobank has used and certified the Variant™ II Turbo analyzer as the reference method for HbA1c measurement. ²⁴ Since TMC Biobank uses the UK Biobank protocol as the reference standard, ¹⁷ we then change to Variant™ II Turbo analyzer for HbA1c analysis. The Variant II Turbo analyzer has been shown to have a sigma level of 3, which is higher than the suggested level by the International Federation of Clinical Chemistry (IFCC) Task Force. ²⁵

Although this study used two different systems for HbA1c measurement, previous studies indicate no systemic and proportional differences between HPLC systems of boronate affinity and cation exchange. ²⁶ However, we still did a pilot study to assess their correlation using fresh blood samples (n = 41). The Pearson's correlation and interclass correlation coefficient (ICC) between Premier Hb9210 and Variant™ II Turbo analyzers were high (r = 0.993 and ICC = 0.96, respectively), with a regression formula of y = 0.95x + 0.58 (Supplementary material S1). A mean difference of −0.29% (95% CI: −0.24, −0.34) was observed (Supplementary material S1), which was suspected to be the analytical error between the different systems. Hence, the regression formula (y = 0.95x + 0.58) was applied to calculate and correct the fresh HbA1c values for later analyses.

2.4. Statistical analysis

Statistical analysis was performed using the Statistical Package for the Social Sciences (SPSS version 22.0, IBM, NY, USA) and RStudio. ²⁷ First, we calculated and corrected the HbA1c readings from the Premier Hb9210 analyzers using the linear regression formula (y = 0.95x + 0.58), and the corrected values were used for comparisons. A Shapiro–Wilk test was performed to assess the normal distribution of the samples. Then, we compared the characteristics of diabetic and non‐diabetic subjects using either a chi‐square test or a Mann–Whitney U test. The difference between fresh, calculated‐fresh, and frozen samples was also evaluated using a Wilcoxon signed‐rank test. Next, the cumulative sum (CUSUM) test was used to evaluate the linearity among measurements. Passing‐Bablok regression can only be applied when the p‐value for the CUMSUM test is greater or equal to 0.05 (p ≥ 0.05). ²⁸ Then, we compare the measurements from fresh, calculated‐fresh, and frozen samples using Passing‐Bablok regression analysis. The regression analysis was done using the “method comparison regression, mcr” R package. ²⁹ Passing‐Bablok regression analysis is a form of linear regression that does not have assumptions on the sample's distribution and measurement errors. ³⁰ , ³¹ Finally, the Bland–Altman plot evaluated the concordance of HbA1c values. In the Bland–Altman plot, we can indicate whether one method tends to overestimate or underestimate via mean difference; the limits of agreement (Mean ± 1.96 SD) were used to portray how well the fresh and frozen samples agree. ³² The Passing‐Bablok regression and Bland–Altman plot were also used to assess the difference between diabetes and non‐diabetes samples. For all analyses, p < 0.05 was considered statistically significant. In addition, where appropriate, this study was reported by adhering to the Checklist for Reporting Stability Studies (CRESS). ³³

3. RESULTS

In general, almost half of the study subjects had diabetes (47.6%), mostly were males (56.6%), Malay ethnicity (54.1%), living in an urban area (84.5%), overweight, middle age, had impaired fasting glucose and had elevated HbA1c (Table 1). Since we calculated the reading of the fresh samples using the y = 0.95x + 0.58 formula (Supplementary material S1) for later analyses, it was referred to as “calculated‐fresh reading” hereafter. According to the Shapiro–Wilk test, the fresh, calculated‐fresh, and frozen samples were non‐normally distributed (p < 0.001). Overall, there were no significant (p = 0.36) differences between the HbA1c readings of the calculated‐fresh and frozen samples. Still, the diabetic subjects generally have a higher reading than the non‐diabetics (Table 1). In addition, there was a slight (mean difference = −0.02%) but significant (p < 0.001) difference in HbA1c between calculated‐fresh and archived samples stored at −196°C for nearly a decade (Table 1). The HbA1c values from calculated‐fresh and frozen stored samples were strongly correlated (Figure 1). The CUSUM test revealed linearity between measurements (p = 0.05), and the Passing‐Bablok regression yielded the following equation:

$$\%\mathit{HbA}1c_{\mathit{\text{frozen sample}}}=0.82\times \%\mathit{HbA}1c_{\mathit{\text{fresh sample}}}+1.11$$

TABLE 1.

Characteristics of study subjects stratified by diabetes status.

Characteristics	Diabetic (n = 191)	Non‐diabetic (n = 210)	Overall (n = 401)	p‐value
Gender, %
Male	61.3	52.4	56.6	0.073
Female	38.7	47.6	43.4
Ethnicity, %
Malay	56.5	51.9	54.1	0.544
Chinese	18.3	22.4	20.5
Indian	25.1	25.7	25.4
Locality, %
Rural	11.5	19.0	15.5	0.037
Urban	88.5	81.0	84.5
Age, median (IQR)	51.97 (47.28–57.26)	50.35 (45.06–54.57)	51.36 (46.02–55.86)	0.001
BMI, kg/m2, median (IQR)	26.99 (24.72–30.22)	27.98 (25.74–30.90)	27.59 (25.39–30.70)	0.055
Fasting blood glucose, mmol/l, median (IQR)	7.74 (7.06–9.38)	6.09 (5.68–6.46)	6.55 (5.95–7.69)	<0.001
% HbA1c, fresh samples (Premier Hb9210, 2012–13), median (IQR)	7.00 (6.40–8.50)	6.00 (5.70–6.40)	6.40 (5.90–7.20)	<0.001
% HbA1c, calculated‐fresh a , median (IQR)	6.76 (6.13–8.34)	5.71 (5.39–6.13)	6.13 (5.90–7.20)	<0.001
% HbA1c, frozen samples (Variant™ II Turbo, 2020–21), median (IQR)	6.80 (6.10–7.80)	6.00 (5.50–6.30)	6.20 (5.70–6.90)	<0.001
Difference (= ‘frozen’ – ‘calculated‐fresh’), median (IQR)	−0.14 (−0.77 – −0.17)	0.13 (−0.15–0.49)	−0.02 (−0.38–0.38)	<0.001
Passing‐Bablok regression equation	y = 0.78x + 1.27	y = 1.05x–0.14	y = 0.82x + 1.11	–
Pearson's correlation, r, between fresh and frozen samples	0.877	0.686	0.875	–
Mean difference based on the Bland–Altman plot	−0.28	0.18	−0.04	–

^a Denotes the fresh HbA1c value calculated using the linear regression formula.

FIGURE 1.

Scatterplot of calculated‐fresh versus frozen samples. The dashed line is the 45‐degree “line of agreement”, while the solid line is the Passing‐Bablok regression line with an equation of y = 0.82x + 1.11. The black circles and grey triangles represent subjects with and without diabetes, respectively.

Based on the Passing‐Bablok model, Pearson's correlation coefficient was 0.88, indicating a high correlation between the two measurements. ³⁴ Besides, the slope for the regression line was 0.82 (95% CI: 0.76, 0.89), suggesting a positive relationship between calculated‐fresh and frozen samples.

Bland–Altman plot (Figure 2) was used to evaluate the presence of systematic bias by plotting the difference between assays. The mean differences along the horizontal axis deviate from zero (p < 0.001), indicating a consistent downward bias of 0.04% in the frozen samples. Although a few samples (7.2%, n = 29) fell outside the limit of agreement, there is a reasonable level of agreement between the two samples.

FIGURE 2.

Bland–Altman plot of differences in HbA1c measurements in calculated‐fresh and frozen samples. *denotes the fresh HbA1c value calculated using the linear regression formula. Horizontal lines were drawn for zero difference as reference (solid line), mean HbA1c difference (dashed line), and standard deviation (dotted lines). The black circles and grey triangles represent subjects with and without diabetes, respectively.

Apart from that, since the diabetic samples deviate more than the non‐diabetics, we further stratify the analysis by diabetes status. The Passing‐Bablok regression equations differ by diabetes status (Table 1). Non‐diabetic subjects had a lower correlation coefficient of 0.69, implying that the readings of the calculated‐fresh and frozen samples of the non‐diabetes have less variability. Interestingly, we observed that the mean difference within diabetic and non‐diabetic samples was −0.28 and 0.18, respectively (Table 1), suggesting that the frozen diabetic samples might somewhat deteriorate over time. In contrast, the HbA1c reading of frozen non‐diabetic samples slightly increases in the long run.

4. DISCUSSION

The HbA1c measurements showed a high correlation and agreement between the two periods, indicating that the HbA1c stored at −196°C results were reproducible and stable throughout almost a decade. To our knowledge, this is the first study that assesses the HbA1c stability kept at −196°C in the long term. Previous studies mostly kept their samples at either −70°C or −80°C. ¹⁰ , ¹³ , ¹⁴ Our findings also implied that the quality of the samples stored at −196°C for a long time was not much degraded. However, a slight negative bias (−0.04%) was observed among the frozen samples, contrary to previous reports. ¹³ , ¹⁴ A difference of 0.50% in HbA1c measurement among samples is considered clinically significant. ³⁵ Since our bias is only 0.04%, it indicates that the HbA1c values measured in our frozen samples are highly compliant with its calculated‐fresh samples.

The HbA1c stability for frozen samples stored in different conditions and times was assessed by several previous studies. ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ We classify them based on their storage period, with a cut‐off of 1.5 years. To date, only three studies ¹⁰ , ¹³ , ¹⁴ reported long‐term storage (more than 1.5 years, at −70°C and −80°C respectively) effect on HbA1c stability. Two reports from the same research team suggested that the HbA1c measurements between fresh and frozen samples are reliable and inter‐correlated, with a slightly higher HbA1c measurement in frozen samples. ¹³ , ¹⁴ Interestingly, they found that a freeze–thawed‐refrozen cycle would increase HbA1c measurements. ¹⁴ However, they could not justify the consistent positive bias in the frozen samples.

Another study performed by Rolandsson and colleagues revealed no significant difference between HbA1c values in fresh and frozen samples when stored for up to 14 years, ¹⁰ which resonates with our findings. Nevertheless, it is noteworthy that the study only includes non‐diabetic samples. Our stratified analysis resonated well with that, where we found the frozen non‐diabetic samples have slightly higher readings (mean difference = 0.18) than its calculated‐fresh samples (Table 1). We also proved that the diabetic samples contribute to the negative bias among fresh and frozen samples in our study. This is because frozen diabetic samples tend to fall outside the agreement limit (±1.96 SD) and generally have lower‐than‐mean HbA1c measurements (Figure 2). One possible explanation for this situation is that richer samples degrade at a higher rate after a single thawed‐freeze cycle. ¹¹ In addition, a similar pattern was observed previously, ¹³ but they have a higher HbA1c measurement overall. This might be because their study has more non‐diabetic subjects (64.0%); compared to our distribution between diabetic and non‐diabetic which is quite balanced (47.6% vs. 52.4%).

In terms of short‐term storage (<1.5 years) of whole blood samples, HbA1c was stable for up to a month when stored at −20°C ¹² and up to 1.5 years when stored at −80°C. ¹¹ No significant difference in HbA1c measurements was observed ⁹ , ¹⁰ , ¹² among the samples. Besides, short‐term frozen samples highly correlate with fresh samples, ¹⁰ , ¹¹ even after a round of freeze–thaw cycles. ¹¹ However, unnecessary freeze–thaw cycles are not encouraged when handling archived samples because the error in HbA1c measurements increases at each freezing step. ¹¹

Despite exciting findings, we would like to highlight certain limitations of this study. All biospecimens collected from the recruitment sites, including East Malaysia or Borneo, are processed at our headquarters located in Kuala Lumpur. ¹⁷ Hence, an undetectable freeze–thaw cycle might occur during transportation, influencing the HbA1c measurements. Moreover, there is a change in the instrumentations to assay HbA1c concentration. The fresh samples were assayed using boronate affinity HPLC, while the frozen ones were assayed using cation‐exchange HPLC. Although several studies ²² , ²⁶ had cross‐validated both assay methods and a high correlation (r = 0.993) and interclass correlation coefficient (ICC = 0.961) (File S1) were observed in our samples, assay instrumentation changes might contribute to the variation in HbA1c measurements. In addition, other confounding factors, such as different blood collection tubes and the additives added to preserve the biospecimens, might influence the measurement. Hence, the reference significance of this study is limited.

5. CONCLUSION

Our data show that the HbA1c measure in samples stored at −196°C for over 7 years is highly reproducible (r = 0.875). However, we observed that the frozen samples are, on average, 0.04% lower than their respective fresh samples. The slight negative bias is attributable to the frozen diabetic samples in the study, where we found the samples, in general, 0.28% lower than the fresh samples.

FUNDING INFORMATION

This work was supported by the Ministry of Education Malaysia (grant number PDE48); additional funding, including infrastructure and utilities, was provided by Universiti Kebangsaan Malaysia.

CONFLICT OF INTEREST STATEMENT

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Supporting information

Supplementary information S1.

Abdullah N, Goh Y‐X, Othman R, et al. Stability of glycated haemoglobin (HbA1c) measurements from whole blood samples kept at −196°C for seven to eight years in The Malaysian Cohort study. J Clin Lab Anal. 2023;37:e24898. doi: 10.1002/jcla.24898

DATA AVAILABILITY STATEMENT

The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request.