Clinical utility of glycated albumin and 1,5-anhydroglucitol in the screening and prediction of diabetes: A multi-center study

Abstract

BACKGROUND

Despite being the gold standard, the use of glycated hemoglobin (HbA1c) and fasting plasma glucose (FPG) for diagnosing dysglycemia is imperfect. In particular, a low level of agreement between HbA1c and FPG in detecting prediabetes and diabetes has led to difficulties in clinical interpretation. Glycated albumin (GA) and 1,5-anhydroglucitol (1,5-AG) may potentially serve as biomarkers for the detection and prediction of diabetes, as well as glycemic monitoring.

AIM

To explore the diagnostic performance of GA and 1,5-AG for screening dysglycemia; assess whether they can be used for glycemic monitoring in Chinese morbidly-obese patients; and examine their predictive ability for incident diabetes in a Chinese community-based cohort.

METHODS

GA and 1,5-AG concentrations were measured in 462 morbidly-obese patients from the Obese Chinese Cohort (OCC). A sub-group of diabetes subjects (n = 24) was prospectively followed-up after bariatric surgery. Differences between baseline and post-surgery biomarker values were converted to percentage change from baseline to assess the response to glycemic control. Predictive ability of the biomarkers was assessed in 132 incident diabetes cases and 132 matched non-diabetes controls in the community-based Cardiovascular Risk Factor Prevalence Study (CRISPS). A prediction model was developed and compared with clinical models based on conventional risk factors.

RESULTS

GA exhibited an excellent diagnostic value with an area under the receiver operating characteristic curve (AUC) of 0.919 (95%CI: 0.884-0.955) for identifying diabetes and a high agreement in the classification of diabetes with both FPG and HbA1c in the OCC. GA demonstrated the fastest response to glycemic control. In CRISPS, the ‘B3A’ prediction model, which consisted of body mass index (BMI) and 3 biomarkers (HbA1c, GA and 1,5-AG), achieved a comparable predictive value [AUC (95%CI): 0.793 (0.744-0.843)] to that of a clinical model comprising BMI, HbA1c, FPG and 2-hour glucose (2hG) [AUC (95%CI): 0.783 (0.733-0.834); DeLong P value = 0.736]. The ‘B3A’ was significantly superior to a clinical model including BMI, HbA1c, FPG and triglycerides [AUC (95%CI): 0.729 (0.673-0.784); DeLong P value = 0.027].

CONCLUSION

GA and 1,5-AG have the potential to act as robust biomarkers for the screening and risk prediction of diabetes. FPG and 2hG may be replaced by GA and 1,5-AG in future diabetes predictions.

Key Words: Diabetes; Biomarkers; Prediction; Glycated albumin; 1,5-anhydroglucitol

Core Tip: This study provides supporting evidence on the effectiveness of glycated albumin (GA) in identifying diabetes and monitoring glycemic control among morbidly-obese individuals. It also highlights the potential clinical utility of the simple ‘B3A’ model, which comprises body mass index, glycated hemoglobin, GA and 1,5-anhydroglucitol (1,5-AG), for predicting diabetes in a community-based cohort. Our findings suggest that both GA and 1,5-AG could serve as supplementary biomarkers, potentially replacing the conventional clinical indicators to predict diabetes. The ‘B3A’ model requires only a single tube of blood sample and simple non-invasive body measurements, making it especially useful for large-scale population screening.

INTRODUCTION

Currently, the diagnosis of dysglycemia relies on the blood glucose criteria, either by fasting plasma glucose (FPG) or 2-hour glucose (2hG) following a 75 g-oral glucose tolerance test (OGTT), or on the glycated hemoglobin (HbA1c) criterion[1]. Despite being the gold standard, the use of blood glucose concentrations and HbA1c for diagnosis is associated with several limitations. Blood glucose concentrations reflect the instant glycemic status, and these measurements can be affected by various factors, such as extreme environmental conditions and alcohol consumption[2-5]. Furthermore, the OGTT has poor reproducibility, and is time-consuming and inconvenient to perform for large-scale screening[6]. HbA1c is determined by the blood glucose concentrations and lifespan of erythrocytes, reflecting the mean blood glucose concentrations over the last 2 to 3 months[7]. Its measurement is inaccurate in conditions with altered erythrocyte turnover, such as anemia and hemoglobinopathies[8]. Moreover, various studies have shown a low concordance rate between HbA1c and FPG in both prediabetes and diabetes detection[9-12], posing challenges in clinical interpretation.

Glycated albumin (GA) is the product of the non-enzymatic glycation process of serum albumin, which is not influenced by the erythrocyte lifespan[13]. It reflects the short-term (approximately 2 to 3 weeks) average glucose concentrations, attributable to the half-life time of albumin[13,14]. GA levels increase along with blood glucose concentrations[15]. The 1,5-anhydroglucitol (1,5-AG) has a similar structure to glucose. High blood glucose blocks the re-absorption of 1,5-AG through the renal tubule, which reduces its concentration in the circulation[16]. Although GA and 1,5-AG have clear potential to serve as biomarkers for the detection and prediction of diabetes, as well as glycemic monitoring, neither of them are routinely used in current clinical settings due to the lack of standardization and limited clinical validation. Previous studies have demonstrated the diagnostic potential of GA[17-23] and 1,5-AG[23-26]. Moreover, GA has been shown to be a useful indicator for short-term changes in glycemic control and in individuals where measurements of HbA1c are deemed inaccurate[27-30]. Notably, none of these studies has evaluated the clinical utility of GA and 1,5-AG in patients with morbid obesity, who are highly susceptible to the development of dysglycemia. Furthermore, there have been limited investigations into the predictive ability of these potential biomarkers for predicting the development of diabetes in long-term prospective studies. A previous study reported that GA is significantly and independently associated with incident diabetes in Caucasians[31]. However, further investigation is required to determine whether GA can replace FPG and 2hG, which are inconvenient to obtain, for the prediction of diabetes.

The current study has three parts. The first part, which involved morbidly-obese patients recruited from the Obese Chinese Cohort (OCC), aimed to evaluate the diagnostic performance of GA and 1,5-AG in identifying prediabetes and diabetes. The second part aimed to assess whether GA and 1,5-AG can serve as sensitive biomarkers reflecting the efficacy of the therapeutic intervention in controlling glycemia in a subgroup of morbidly-obese patients with diabetes who were prospectively followed up after bariatric surgery. The third part of this study sought to examine the predictive ability of GA and 1,5-AG for incident diabetes and develop a user-friendly prediction model in a prospective community-based cohort with approximately 20 years of follow-up period.

MATERIALS AND METHODS

Study participants

Obese Chinese Cohort (OCC): The first part of the study involved 462 subjects ≥ 18 years old recruited from the OCC, who were screened for eligibility for bariatric surgery at the First Affiliated Hospital of Jinan University from January 2019 to March 2022 (Supplementary Figure 1). Details of this cohort have been described previously[32,33]. The second part of the current study assessed whether GA and 1,5-AG can serve as sensitive biomarkers for glycemic monitoring by analyzing the blood samples of patients with diabetes collected before bariatric surgery (n = 24), and at one (n = 13) and three (n = 14) months after the surgery. As a 75 g-OGTT was not conducted in this cohort, the diagnoses of prediabetes and diabetes were based on the FPG and HbA1c criteria[1]. Prediabetes was defined as FPG of 5.6-6.9 mmol/L, or HbA1c from 5.7%-6.4%, while diabetes was defined as FPG ≥ 7.0 mmol/L, or HbA1c ≥ 6.5%. Written informed consent was obtained from each participant. The study protocol was approved by the Institutional Review Board (IRB) of Jinan University (2016-017) and the University of Hong Kong/Hospital Authority, Hong Kong West Cluster (Approval number: UW 20-700). Further details of this cohort and the inclusion and exclusion criteria of this cohort are described in the Supplementary material.

Hong Kong Cardiovascular Risk Factor Prevalence Study: The third part of the current study involved participants recruited from the Hong Kong Cardiovascular Risk Factor Prevalence Study (CRISPS) cohort[34]. A nested case-control study was conducted on 158 subjects who had developed diabetes by the latest assessment (2016-2018) and 158 age and sex-matched subjects who remained free from diabetes (Supplementary Figure 2). In this cohort, prediabetes and diabetes were defined according to the American Diabetes Association criteria[1]. Prediabetes was defined as FPG of 5.6-6.9 mmol/L, 2hG of 7.8-11.0 mmol/L, or HbA1c from 5.7%-6.4%, while diabetes was defined as FPG ≥ 7.0 mmol/L, 2hG ≥ 11.0 mmol/L, or HbA1c ≥ 6.5%. Ethics approvals were obtained from the IRB of the University of Hong Kong/Hospital Authority, Hong Kong West Cluster (Approval number: UW 10-038). All participants provided written informed consent before any study-related procedures were performed. Further details of this cohort are described in the Supplementary Notes of Supplementary material.

Biomarkers measurements

FPG and HbA1c were measured as a part of standard clinical examination by the accredited clinical laboratories of the Huaqiao Hospital in Guangzhou and the Queen Mary Hospital in Hong Kong. Circulating concentrations of GA and 1,5-AG were measured using the commercially available enzymatic assay kits (GA Cat. No. 51970; 1,5-AG Cat. No. 51990; Immunodiagnostics Co., Ltd, Hong Kong) according to the manufacturer’s instructions. For quality control measures, five different control samples were used to test each cohort as to ensure the consistency and stability of the reagents. The intra-assay and inter-assay coefficients of variation for GA and 1,5-AG were 1.1% to 2.7% and 0.8% to 1.4%, respectively. Further details on the measurement of biomarkers are described in the Supplementary Notes of Supplementary material.

Statistical analysis

All statistical analyses were performed using SPSS Statistics 27 (SPSS, Chicago, IL, United States), GraphPad Prism (version 9), and R (version 4.3.2). Data are presented as mean ± SD or median (interquartile range), where appropriate. The correlation between the two biomarkers was analyzed using Pearson correlation analysis. The agreement of classifications between the different biomarkers was determined by Cohen’s kappa coefficient[35]. To allow the comparison of changes from baseline in the biomarkers with varying units of measure over the 3-month period, the difference between values at baseline and at one and three months after bariatric surgery were converted to percentage changes from baseline. To assess the correlations of the biomarkers with repeated observations over the 3 months, a method that was previously described by Bland and Altman[36] was employed. Receiver operating characteristic (ROC) curve analysis was performed and the area under the ROC curves (AUC) was used to evaluate the discriminatory power. The optimal cut-off value for each biomarker was determined by the Youden index[37]. The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated when assessing the performance at the optimal cut-off values. To develop a user-friendly model, we first identified two clinical models for comparisons, with or without the inclusion of 2hG after OGTT, using the stepwise model selection approach. Variables showing a significant difference between the incident diabetes cases and non-diabetes controls were considered in the stepwise model selection process. The first clinical model comprised body mass index (BMI), HbA1c, FPG, and 2hG (Clinical Model 1) and the second model comprised BMI, HbA1c, FPG and triglycerides (TG) (Clinical Model 2). The new “B3A” prediction model was developed by combining BMI, HbA1c, GA and 1,5-AG in the multivariable logistic regression model. The AUCs were compared using the DeLong test[38]. Additional details on statistical analysis are described in the Supplementary Notes of Supplementary material.

RESULTS

GA and 1,5-AG demonstrated excellent performances in the detection of diabetes among morbidly-obese patients from the OCC

Out of the 462 morbidly-obese subjects (mean age: 31.86 ± 9. 91 years; mean BMI: 38.52 ± 7.92 kg/m²; male: 65.18%), 164 (35.5%) had normoglycemia, 171 (37.0%) had prediabetes and 127 (27.5%) had diabetes (Supplementary Table 1). Subjects with prediabetes or diabetes were older, and had significantly higher BMI, waist circumference, FPG and HbA1c than the normoglycemic controls. The concentrations of GA and 1,5-AG gradually increased and decreased, respectively, with the glycemic statuses (P value < 0.001; Supplementary Figure 3). GA concentrations were significantly and positively correlated with both FPG (r = 0.68; P value < 0.001) and HbA1c (r = 0.69; P value < 0.001). The 1,5-AG showed negative correlations with both FPG (r = -0.51; P value < 0.001) and HbA1c (r = -0.58; P value < 0.001). As expected, FPG and HbA1c were significantly correlated with each other (r = 0.82; P value < 0.001; Supplementary Figure 4).

ROC curve analysis showed that the AUC of GA on the identification of prediabetes and diabetes were 0.838 (95%CI: 0.795-0.880) and 0.919 (95%CI: 0.884-0.955), respectively (Figure 1 and Supplementary Table 2). At the optimal cut-off determined by the Youden index (13.06%), GA achieved a sensitivity of 68.37% and a specificity of 58.53% for the identification of prediabetes. At the optimal cut-off for diabetes (16.39%), GA achieved a high specificity of 95.68%, sensitivity of 81.60%, PPV of 87.18% and NPV of 93.52%. On the other hand, 1,5-AG showed an AUC of only 0.635 (95%CI: 0.576-0.695) for the identification of prediabetes, with a sensitivity of 61.48% and specificity of 60.00% at the optimal cut-off value (130.39 μmol/L). Nonetheless, 1,5-AG demonstrated a good discriminatory ability for the identification of diabetes, with an AUC of 0.829 (95%CI: 0.790-0.881), a sensitivity of 73.60%, specificity of 83.09% and NPV of 89.64% at the optimal cut-off value of 68.87 μmol/L (Figure 1).

Figure 1 Receiver operating characteristic curve analyses of prediabetes and diabetes in the Obese Chinese Cohort. A: Prediabetes in the Obese Chinese Cohort (OCC); B: Diabetes in the OCC. GA: Glycated albumin; 1,5-AG: 1,5-anhydroglucitol; AUC: Area under the receiver operating characteristic curve.

Combination of HbA1c and GA facilitates the detection of additional diabetes cases with a consistent diagnosis with FPG

In 299/462 subjects (64.72%), the glycemic status classification by FPG and HbA1c was in agreement, with 151 (32.68%) classified as normoglycemia, 60 (12.99%) classified as having prediabetes, and 88 (19.05%) classified as having diabetes by both measures (overall Cohen’s kappa = 0.451) (Table 1). An agreement in the diagnosis of diabetes between FPG and HbA1c was noted in 411/462 subjects (88.96%), including 323 (69.91%) diagnosed as non-diabetes and 88 (19.05%) classified as having diabetes by both tests (Cohen’s kappa for diabetes = 0.702) (Supplementary Table 3).

Table 1 Agreement in diagnostic classification between different biomarkers.

		Normoglycemia	Prediabetes	Diabetes	Total
Concordance in diagnostics classification (HbA1c vs FPG)
		Classification based on FPG
Classification based on HbA1c	Normoglycemia	151	41	6	198
	Prediabetes	71	60	18	149
	Diabetes	7	20	88	115
	Total	229	121	112	462
Concordance in diagnostics classification (GA vs FPG)
		Classification based on FPG
Classification based on GA	Normoglycemia	127	31	9	167
	Prediabetes	90	67	21	178
	Diabetes	12	23	82	117
	Total	229	121	112	462
Concordance in diagnostics classification (GA vs HbA1c)
		Classification based on HbA1c
Classification based on GA	Normoglycemia	127	31	9	167
	Prediabetes	66	99	13	178
	Diabetes	5	19	93	117
	Total	198	149	115	462
Concordance in diagnostics classification (HbA1c/GA vs FPG)
		Classification based on FPG
Classification based on HbA1c/GA	Normoglycemia	112	15	0	127
	Prediabetes	104	77	15	196
	Diabetes	13	29	97	139
	Total	229	121	112	462

The number of subjects with the same diagnosis by both tests are underlined. Optimal cutoffs for prediabetes: Glycated albumin (GA): 13.06%; 1,5-anhydroglucitol (1,5-AG): 130.39 μmol/L; fasting plasma glucose (FPG): 5.43 mmol/L; glycated hemoglobin (HbA1c): 5.65%. Optimal cutoffs for diabetes: GA: 16.39%; 1,5-AG: 68.87 μmol/L; FPG: 6.43 mmol/L; HbA1c: 6.45%. GA: Glycated albumin; 1,5-AG: 1,5-anhydroglucitol; FPG: Fasting plasma glucose; HbA1c: Glycated hemoglobin.

Given the excellent diagnostic performances of GA for both prediabetes and diabetes, we then examined its agreement of classification with FPG and HbA1c. The overall Cohen’s kappa values for the classification of prediabetes and diabetes of GA with FPG and GA with HbA1c were 0.389 and 0.529, respectively. GA demonstrated a high agreement in classification for diabetes with FPG, with 397 individuals (85.93%) having the same diagnosis by both tests, including 315 (68.18%) being classified as non-diabetes and 82 (17.75%) being classified as having diabetes by both measures (Cohen’s kappa for diabetes = 0.623) (Supplementary Table 3). There was also a high agreement for the classification of diabetes between GA and HbA1c, with 416 individuals (90.04%) having a consistent classification status, including 323 (69.91%) and 93 (20.13%) cases classified, respectively, as non-diabetes and diabetes by both measures (Cohen’s kappa for diabetes = 0.735) (Supplementary Table 3). We then further evaluated whether the combination of HbA1c and GA could facilitate the detection of additional diabetes cases that showed consistent diagnosis among different tests, which indicates a more reliable diagnosis. The glycemic status classification by FPG and the combination of HbA1c and GA was in agreement in 286 subjects (61.90%), with 112 (24.24%) classified as normoglycemia, 77 (16.67%) classified as having prediabetes, and 97 (21.00%) classified as having diabetes (overall Cohen’s kappa = 0.44) (Table 1). Compared to 88 individuals (78.57%) being classified as having diabetes by both FPG and HbA1c, 97 (86.61%) of the diabetes cases detected by FPG were also classified as having diabetes when combining the HbA1c and GA tests (Cohen’s kappa for diabetes = 0.690). This represented an approximately 8% increase in the detection of diabetes cases with a consistent diagnosis (Supplementary Table 3).

GA responds faster to glycemic control after therapeutic intervention

Plasma samples collected at baseline (n = 24), one (n = 13) and three (n = 14) months after bariatric surgery from 24 morbidly-obese patients with diabetes (mean age: 28.89 ± 8.62 years; mean BMI: 36.73 ± 6.46 kg/m²; male: 28.21%) from the OCC were examined to evaluate whether GA and 1,5-AG can facilitate glycemic monitoring (Supplementary Table 4). Table 2 shows the concentrations of FPG, HbA1c, GA and 1,5-AG at different time points before and after surgery. GA, with a steep reduction of 42.92% one month after the surgery, showed the fastest response that was followed by FPG, which decreased by 35.06%, and 1,5-AG, which increased by 22.86% (Figure 2). In contrast, HbA1c showed a reduction of only 13.05% at one month, and a moderate decrease of 25.75% at 3 months. On the other hand, GA showed a 38.30% reduction at 3 months, which was comparable to the change in FPG (-36.76%) and proportionally slightly higher than the 1,5-AG increase (+31.42%), demonstrating a trend of changes that closely paralleled FPG variation over the 3-month period (Figure 2). Significant correlations were observed among the 4 biomarkers, with FPG attaining a slightly higher correlation coefficient with GA (r = 0.76; P value < 0.001) compared to HbA1c (r = 0.74; P value < 0.001; Supplementary Figure 5). These data suggested that GA may act as a sensitive biomarker accurately reflecting the efficacy of the intervention in controlling glycemia.

Figure 2 Relative change in fasting plasma glucose, glycated hemoglobin, glycated albumin and 1,5-anhydroglucitol levels after bariatric surgery among morbidly-obese patients with diabetes. GA: Glycated albumin; 1,5-AG: 1,5-anhydroglucitol; FPG: Fasting plasma glucose; HbA1c: Glycated hemoglobin.

Table 2 Circulating levels of fasting plasma glucose, glycated hemoglobin, glycated albumin and 1,5-anhydroglucitol of morbidly-obese patients with diabetes at baseline, one and three months after the bariatric surgery.

	Baseline (n = 24)	1-month (n = 13)	3-month (n = 14)	P for trend
FPG (mmol/L)	8.27 ± 2.77	5.37 ± 0.83	5.23 ± 0.86	< 0.001
HbA1c (%)	7.18 ± 1.56	6.24 ± 1.14	5.33 ± 0.52	< 0.001
GA (%)	19.00 ± 5.98	10.85 ± 3.16	11.72 ± 2.97	< 0.001
1,5-AG (µmol/L)	61.09 ± 48.17	75.05 ± 53.77	80.28 ± 40.79	< 0.001

Data are presented as mean ± SD. FPG: Fasting plasma glucose; HbA1c: Glycated hemoglobin; GA: Glycated albumin; 1,5-AG: 1,5-anhydroglucitol.

The predictive ability of either GA or 1,5-AG alone was comparable to FPG and HbA1c

The baseline clinical characteristics of the 158 incident diabetes cases and 158 age and sex-matched non-diabetes controls from the CRISPS sub-cohort are shown in Supplementary Table 5. When compared with those without incident diabetes, the incident diabetes cases had significantly higher BMI, waist circumferences, FPG, 2hG, HbA1c, total cholesterol, low-density lipoprotein, TG, diastolic blood pressure, and lower high-density lipoprotein. GA concentrations were significantly higher while 1,5-AG concentrations were significantly lower in patients with incident diabetes (P value < 0.001 for both).

GA alone showed an AUC of 0.680 (95%CI: 0.622-0.738) for predicting incident diabetes. At the optimal cutoff of 11.80%, GA achieved a sensitivity of 75.3% and a specificity of 53.2%. The 1,5-AG alone showed a similar predictive power, with an AUC of 0.664 (95%CI: 0.604-0.723) (DeLong P value > 0.05). At the optimal cutoff at 122.73 μmol/L, 1,5-AG demonstrated 70.9% sensitivity and 58.90% specificity. Notably, both GA and 1,5-AG showed AUCs similar to those of HbA1c [AUC (95%CI): 0.635 (0.574-0.697)] and FPG [AUC (95%CI): 0.650 (0.589-0.710)] (DeLong P value > 0.05 for both). GA also showed an AUC comparable to that of 2hG alone (DeLong P value = 0.067), which had in turn an AUC of 0.749 (95%CI: 0.695-0.803) and significantly outperformed HbA1c, FPG, and 1,5-AG (DeLong P value < 0.05 for all comparisons; Supplementary Figure 6). Notably, both GA [odds ratio (OR) (95%CI): 1.29 (1.16-1.45); P value < 0.001)] and 1,5-AG [OR (95%CI): 0.99 (0.98-0.99); P value < 0.001] remained significantly and independently associated with incident diabetes even after adjustments relevant clinical predictors (Supplementary Table 6).

The ‘B3A’ prediction model for diabetes

We next assessed the predictive ability of clinical models comprising conventional diabetes indicators, and evaluated also a new model including AG and 1,5-AG. Clinical Model 1, which consisted of BMI, HbA1c, FPG, and 2hG, achieve an AUC of 0.783 (95%CI: 0.733-0.834), with 72.78% sensitivity and 73.72% specificity (Table 3 and Figure 3). Clinical Model 2, which comprised BMI, HbA1c, FPG and TG, demonstrated an AUC of 0.729 (95%CI: 0.673-0.784), with 61.15% sensitivity and 75.00% specificity (Table 3 and Figure 3). To construct a simple model for diabetes prediction that does not require patients to fast or undergo uncomfortable OGTT procedures, we investigated whether GA and 1,5-AG could replace FPG and 2hG in Clinical Model 1, or FPG and TG in Clinical Model 2. As with HbA1c, GA and 1,5-AG concentrations are not influenced by fasting status, and obtaining blood samples for their measurement is convenient in routine clinical practice. This makes them potential alternatives to the fasting parameters and the 2hG measurement after an OGTT. When FPG and 2hG in Clinical Model 1 were replaced by GA and 1,5-AG, the new ‘B3A’ model, which comprised BMI, and the 3 non-fasting biomarkers (HbA1c, GA and 1,5-AG), achieved comparable predictive performance with an AUC of 0.793 (95%CI: 0.744-0.843; DeLong P value = 0.736; Table 3 and Figure 3). When the two fasting parameters (i.e., FPG and TG) in Clinical Model 2 were replaced by GA and 1,5-AG, the ‘B3A’ model significantly outperformed the original model (AUC 0.793 vs 0.729), demonstrating a 6.4% improvement in the AUC (DeLong P value = 0.027; Table 3 and Figure 3). The ‘B3A’ model achieved also a reasonable sensitivity of 77.22% and a specificity of 72.44%, which were similar to those of Clinical Models 1 and 2.

Figure 3 Receiver operating characteristic curve analyses of the different models. Clinical Model 1: Body mass index (BMI), glycated hemoglobin (HbA1c), fasting plasma glucose (FPG), 2-hour glucose; Clinical Model 2: BMI, HbA1c, FPG, triglycerides; ‘B3A’ Model: BMI, HbA1c, glycated albumin, 1,5-anhydroglucitol.

Table 3 Predictive performance of the different prediction models.

Prediction models	AUC (95%CI)	DeLong P value	Sensitivity (%)	Specificity (%)
Clinical Model 1 (BMI, HbA1c, FPG, 2hG)	0.783 (0.733-0.834)	0.736	72.78	73.72
Clinical Model 2 (BMI, HbA1c, FPG, TG1)	0.729 (0.673-0.784)	0.027	61.15	75.00
‘B3A’ Model [BMI and 3 biomarkers (HbA1c, GA, 1,5-AG)]	0.793 (0.744-0.843)	Reference	77.22	72.44

¹Natural-log transformed before analysis.

BMI: Body mass index; FPG: Fasting plasma glucose; HbA1c: Glycated hemoglobin; 1:5-AG: 1:5-anhydroglucitol; GA: Glycated albumin; TG: Triglyceride; AUC: Area under the receiver operating characteristic curve; 2hG: 2-hour glucose.

DISCUSSION

Ongoing research is being conducted to identify alternative or complementary biomarkers to further improve the accuracy of diagnosis and prediction of diabetes. In the current study, the potential of GA and 1,5-AG for the detection of diabetes was systematically explored in morbidly-obese patients who are at high risk of developing dysglycemia. In addition to showing an excellent discriminatory power for identifying diabetes, GA also displayed a high concordance with both FPG and HbA1c. In the prospective follow-up analysis of the morbidly-obese patients with diabetes who underwent bariatric surgery, GA demonstrated the fastest response to improvements in glycemic control, indicating its potential to act as a sensitive biomarker for glycemic monitoring. Moreover, highlighting the clinical utility of GA and 1,5-AG in the prediction of diabetes, we demonstrated that a simple prediction model incorporating those biomarkers (‘B3A’ model), which would not require the patients to fast or undergo OGTT, could achieve comparable or even better prediction performance than the conventional clinical models. Although GA and 1,5-AG have been recognized as biomarkers for diabetes, they are not been widely used in clinical practice. This is attributable to the lack of standardization and limited clinical validation studies across different populations. We are the first to conduct a systematic evaluation of the diagnostic potentials of GA and 1,5-AG among morbidly-obese Chinese patients who are susceptible to developing dysglycemia.

GA has demonstrated satisfactory correlations with both HbA1c and FPG. In line with previous studies[17-19,21-23], in our analysis GA was found to be useful in diagnosing dysglycemia, as demonstrated by its high discriminatory power for both prediabetes and diabetes. We here report, for the first time, a high agreement between GA and either HbA1c or FPG in the detection of diabetes. GA may thus serve as a useful biomarker to validate the diagnosis by FPG and HbA1c in situations where the use of these established biomarkers is unfeasible or inappropriate. Furthermore, in agreement with previous studies which showed that the combination of HbA1c and GA could improve the detection of diabetes[21,23], we showed that these two markers combined allowed the identification of additional diabetes cases similarly diagnosed as per the FPG classification alone, which facilitated the clinical interpretation of the diagnosis. In future studies, other potential emerging biomarkers, such as advanced glycation end-products[39] and certain microRNAs[40], may also be explored for their diagnostic performances independently or in combination with GA and 1,5-AG.

Attributable to the long lifespan of erythrocytes, HbA1c is considered an unsuitable biomarker for the assessment of short-term variation of glycemic control[41]. In contrast, GA is not influenced by the half-life of erythrocytes and is therefore particularly useful for such evaluation of short-term glycemic control in situations where early diagnosis and timely intervention are required, such as gestational diabetes[42]. To our knowledge, the use of GA or 1,5-AG for glycemic monitoring after bariatric surgery has not been previously reported. The current study showed that in morbidly-obese diabetic patients, GA is a sensitive biomarker that readily responds to short-term changes in blood glucose concentrations following bariatric surgery. Our findings are in line with a previous study which demonstrated that GA showed a more rapid reduction compared to HbA1c during intensive insulin therapy in type 2 diabetes patients[27]. Compared to HbA1c and 1,5-AG, GA also demonstrated a trend in variation that was closely correlated with that of FPG over the 3-month recovery period after bariatric surgery. Therefore, assessment of GA would allow clinicians to adjust the treatment regimen in a timely manner.

In the current study, GA alone or 1,5-AG alone showed comparable predictive values with both FPG and HbA1c. Except for GA, which showed a comparable predictive ability with 2hG, the predictive values of all other biomarkers were significantly inferior to that of 2hG. Selvin et al[31] showed that GA was independently associated with incident diabetes, even after adjustment for HbA1c or FPG. However, their study did not determine whether GA can replace conventional biomarkers used for the prediction of diabetes. Still, and in line with their findings, we also demonstrated that GA and 1,5-AG were independently associated with incident diabetes. In view of the inconvenience and possibly unpleasant experience associated with the measurement of FPG and 2hG after OGTT, we developed a more practical ‘B3A’ model, which incorporates several non-fasting and easy-to-collect parameters, for the prediction of diabetes. This model was comparable to a clinical model that included the strongest predictor, 2hG, showing a similar predictive performance in terms of AUC, sensitivity, and specificity. This simple model also significantly outperformed another clinical model that replaced 2hG with TG, showing a significantly higher AUC (0.793 vs 0.729). These observations suggested that both GA and 1,5-AG may act as supplementary biomarkers and replace some conventional clinical predictors (e.g., 2hG, FPG, and TG). The ‘B3A’ model requires only a single tube of blood sample and simple, non-invasive body measurements, making it especially useful for large-scale population screening or in situations where estimating FPG or 2hG after OGTT is unsuitable.

A major strength of this study was the use of multiple well-characterized cohorts. The morbidly-obese patients in the OCC involved a large proportion of prediabetes and diabetes subjects, which favored the comparison of prediabetes and diabetes detection by different biomarkers. On the other hand, the long follow-up period of the CRISPS sub-cohort allowed for a robust assessment of diabetes prediction. Moreover, when compared with the existing tests, such as HbA1c and FPG, the use of GA and 1,5-AG for diabetes screening will be particularly beneficial in low-resource settings. This is due to the simplicity of measuring these biomarkers using commercially available ELISA kits, which do not require fasting blood samples, as well as the cost-effectiveness reflected by the relatively low cost of the kits and reagents required for measuring these two biomarkers[43].

We acknowledge several limitations of the present study. First, due to the OCC study design, which aimed to screen morbidly-obese participants for bariatric surgery, the assessment of OGTT was not included in the study protocol. The lack of OGTT as the reference standard for diagnosing prediabetes and diabetes has therefore hindered our ability to comprehensively compare the performance of GA and 1,5-AG with 2hG in this cohort. Nonetheless, we were able to confirm the good performance of both GA and 1,5-AG in the detection of diabetes and assessment of glucose monitoring among individuals with morbid obesity. Second, the sample sizes of both the OCC and CRISPS sub-cohort were relatively small, which may affect the robustness of statistical inference and the generality of the results. Nonetheless, based on these cohorts, we explored and successfully demonstrated the possibility of using GA and 1,5-AG for the detection and prediction of diabetes, as well as the monitoring of glycemic control. Further studies with larger sample sizes and in different ethnic groups will be necessary to validate our findings. Third, the current study lacks long-term follow-up data to evaluate the potential clinical utility of GA and 1,5-AG in predicting the progression and development of diabetic complications, such as diabetic nephropathy and diabetic retinopathy. Future studies with long-term follow-up data would allow to assess the predictive ability of these biomarkers on the progression and development of diabetic complications over time, further enhancing the clinical value of these biomarkers in long-term management. Fourth, the nested case-control study design of the CRISPS sub-cohort hindered us from developing a proper risk score for predicting diabetes. Future studies in a large population-based cohort would be of aid in validating the usefulness of the ‘B3A’ model. Finally, our study was confined to individuals of Chinese ancestry, thereby limiting the international relevance and potential for broader application on a global scale. Future studies in diverse ethnic groups, encompassing individuals with and without obesity, will be useful to explore the clinical utility of these biomarkers across different demographics and validate our findings. In addition, further studies to establish a standardized method for measuring GA and 1,5-AG will be necessary, as to ensure consistency and reliability in different clinical settings and lay the foundation for future large-scale applications.

CONCLUSION

In conclusion, this study demonstrated that GA and 1,5-AG could potentially be useful in clinical applications, in supplement to the current gold standards, for the identification of diabetes. The user-friendly model herein developed, incorporating BMI, HbA1c, GA and 1,5-AG, may be a robust alternative to simply and accurately predict diabetes. Further independent validations to confirm our findings are warranted.

ACKNOWLEDGEMENTS

The authors thank all the study participants, and the clinical and research staff of the OCC and the CRISPS cohort studies for their contribution to this research.