PDF
Citation
Print
INTRODUCTION
Prediabetes is an intermediate stage in the transition between normal glucose tolerance (NGT) to overt diabetes. Prediabetes is associated with a modest increase in the risk for incident cardiovascular disease [1,2] and accurate diagnosis of diabetes and prediabetes is important to determine the survival outcomes in patients with chronic heart failure [3], or coronary artery disease [4]. Therefore, it is important to perform appropriate tests to detect high-risk subjects to prevent the progression to diabetes and related cardiometabolic outcomes. The diagnosis of diabetes or prediabetes has been traditionally based on fasting plasma glucose (FPG) and/or 2-hour postprandial glucose (2hPG) after 75-g oral glucose tolerance test (OGTT). Additionally, an International Expert Committee recommended measurement of glycated hemoglobin (HbA1c) levels as a diagnostic parameter for diabetes (cutoff of 6.5%) [5] and the American Diabetes Association (ADA) recommended 5.7% to 6.4% as the cutoff for prediabetes [6].
Among different glycemic diagnostic parameters, FPG was contentious in identifying high-risk subjects while impaired glucose tolerance (IGT) was suggested as a strong surrogate marker that is associated with diabetic complications and cardiovascular outcomes [7]. Additional performance of OGTT or HbA1c test can be used to detect high-risk subjects for diabetes or early type 2 diabetes mellitus (T2DM) compared with FPG test alone [8–11]. Although the overall sensitivity of 2hPG by OGTT is higher than HbA1c levels for the diagnosis of diabetes [12–14] or prediabetes [15,16], application of OGTT in clinical practice is a huge challenge. Lack of reproducibility and requiring extensive patient preparation with multiple sampling are barriers to conducting OGTT [17]. By contrast, the advantage of using HbA1c for diagnosis of T2DM includes its stability at ambient temperature, limited biological variability, and convenience to collect single blood samples during non-fasting conditions for diagnosis and treatment. A relatively simple and reproducible HbA1c test can play a complementary role in identifying high-risk subjects for diabetes. However, the usefulness of HbA1c in the screening and diagnosis of diabetes has been widely debated [18–20]. In addition, a low diagnostic concordance between different diagnostic parameters in non-diabetic states shows that a single diagnostic test is not adequate to diagnose high-risk subjects in the general population [21–23].
Central obesity is the main pathophysiologic comorbidity that is related to diabetes and metabolic syndrome. A previous study showed that the concordance between HbA1c criteria and OGTT criteria for prediabetes was affected by the increased body mass index (BMI) and sensitivity of HbA1c test for prediabetes compared with OGTT increased in the obese population [24]. The correlation between BMI and HbA1c varied according to glycemic status in which BMI was significantly positively associated with HbA1c in the non-diabetic but not diabetic population [25]. However, few studies compare the risk of incident diabetes based on complex glycemic parameters that simultaneously evaluated OGTT and HbA1c, and the role of obesity in determining the predictive value of each test. Therefore, it is important to identify subjects who are appropriate for each diagnostic test and their sensitivity and specificity. Thus, we compared the clinical and socioeconomic characteristics according to different glycemic parameters based on OGTT and HbA1c test, and evaluated the prognostic performance of each glycemic parameter (FPG, 2hPG in OGTT, HbA1c) in identifying high-risk people for incident T2DM. Further, the hazard ratio (HR) and diagnostic performance of each test was compared depending on the presence of central obesity.
METHODS
Study design and population
We collected data from the Korean Genome and Epidemiologic Study (KoGES) that is a community-based government-funded cohort study involving residents recruited from urban (Ansan) and rural (Ansung) communities. A total of 10,030 participants aged 40 to 69 years were voluntarily enrolled at baseline between 2001 and 2002, with follow-up examination every 2 years. The retention rate was 62.8% at the end of the 6th follow-up phase, and 44.4% (n=4,452) of the baseline participants responded consecutively to all six follow-up phases by 2014 [26]. We excluded subjects who were previously diagnosed with diabetes (n=675), who were undiagnosed but met diagnostic criteria of diabetes based on OGTT or HbA1c (n=580) to include subjects with non-diabetic state. In addition, we also excluded those who had anemia (hemoglobin levels <13.5 mg/dL [men] and <11.5 mg/dL [women]) or hemoglobinemia (hemoglobin levels >17.5 mg/dL [men] and >16 mg/dL [women]) because HbA1c levels were affected by hemoglobinopathies or anemia and disorders associated with accelerated red cell turnover disease (n=1,013) [5]. Finally, 7,512 subjects (3,536 men) were included in this study. Written informed consent by the patients was waived due to a retrospective nature of the study. This study was approved by the ethics committee of Gyeongsang National University Hospital (IRB no. 2020-03-027).
Baseline variables
The baseline survey included evaluation of demographic, socioeconomic (education level, monthly incomes, and family history of diabetes in first-degree relatives), and lifestyle factors (alcohol/smoking intake, physical activity) by trained interviewers. Anthropometric and clinical measurements were obtained including OGTT. After fasting for 12 hours, the glucose levels and lipid profiles were measured using a Hitachi 747 chemistry analyzer (Hitachi Ltd., Tokyo, Japan). The HbA1c level was determined by high-performance liquid chromatography (Variant II, BioRad Laboratories, Hercules, CA, USA). The circulating concentration of C-reactive protein (CRP) was measured by immunoradiometric assay (ADVIA 1650, Bayer Diagnostics, Tarrytown, NY, USA). OGTT was performed by ingestion of standard 75 g glucose load, with glucose and insulin levels determined at 0, 60, and 120 minutes. The insulin resistance was calculated by a homeostasis model assessment of insulin resistance (HOMA-IR), using the following formula: [fasting plasma insulin (μIU/mL)×FPG (mg/dL)/405] [27]. The creatinine clearance was calculated using the chronic kidney disease epidemiology collaboration equation [28].
The presence of central obesity was defined by a waist circumference (WC) with the cut off values of 90 cm or more for men and 85 cm or more for women using standard methods in subjects wearing light clothes [29]. Self-reported smoking habit was classified into three categories: non-, ex-, and current smoker, and the alcohol consumption status was categorized into three groups: non-, ex-, and current intake. Monthly incomes were classified into three categories: <$1,250/month, $1,250 to 3,300/month, ≥$3,300/month. Physical activity was assessed in a binary fashion, and those who performed moderate intensity activity for 150 minutes or more within a week were considered to practice regular physical activity. Blood pressure was recorded three times in the morning with a relaxed state for at least 10 minutes with 5-minute rest period between the measurements. Glucose measurements were assessed in binary fashion, using the following cut points: IGT, 2hPG 140 to 199 mg/dL; high HbA1c, HbA1c levels 5.7% to 6.4%. At baseline, subjects were categorized into four groups based on the results of OGTT and HbA1c tests: (1) NGT/normal HbA1c, (2) NGT/high HbA1c, (3) IGT/normal HbA1c, and (4) IGT/high HbA1c.
Outcomes
All subjects underwent glycemic measurements (both OGTT and HbA1c) at each biannual follow-up visit, and continued consecutively up to six follow-up phases by 2014. Among those with non-diabetic status at baseline, incident diabetes was defined if subjects were newly diagnosed with diabetes clinically or FPG was 126 mg/dL or more and/or 2hPG was 200 mg/dL or more, and/or HbA1c was 6.5% or more. The date of diagnosis was also recorded in this cohort study. The minimum and maximum follow-up duration was 17 and 152 months, respectively.
Statistical analysis
Continuous variables are presented as mean±standard deviation for normally distributed variables or median with interquartile range (IQR) for non-normally distributed variables. Categorical variables are presented as counts and corresponding percentages. The baseline variables between four subgroups according to different glycemic parameters were compared using analysis of variance (ANOVA) for continuous variables and Pearson’s chi-square test for categorical variables. Additionally, Tukey’s honestly significant difference (HSD) post hoc analysis was used to for all pair wise comparison and Bonferroni correction method was used for categorical variables. The difference in correlation between 2hPG and HbA1c according to the presence of central obesity was assessed using pairwise comparison with the Least Square Means. Cox proportional hazards regression analysis was performed to calculate HR and 95% confidence interval (CI) for incident diabetes during follow-up according to the complex glycemic status at baseline using NGT with normal HbA1c level as the reference category. The potential confounding variables were adjusted for area, age, sex, estimated glomerular filtration, CRP, total-to-high density lipoprotein (HDL) cholesterol ratio, triglyceride (TG), systolic/diastolic blood pressure (SBP/DBP), and socioeconomic factors (education levels, monthly incomes, alcohol/smoking status, physical activity, family history of diabetes). The follow-up time was defined as the interval in days between the date of baseline examination and the date of last follow-up or the date of incident diabetes. Harrell concordance index (C-index) by the receiver operating characteristic (ROC) analysis was performed to assess the role of additional glycemic parameters in the discrimination of incident diabetes. Furthermore, subgroup analysis was performed according to the presence of central obesity. All statistical analyses were performed using SPSS version 24.0 (IBM Co., Armonk, NY, USA) and all figures were drawn in R version 3.2.5 (R Foundation for Statistical Computing, Vienna, Austria). All statistical tests were two-tailed and the significance level was set at P<0.05.
RESULTS
At baseline, a total of 7,512 subjects with non-diabetic state were enrolled in the present study. The cohort demographics were as follows: mean age, 52 years; mean BMI, 24.5 kg/m2; and mean waist-to-hip ratio, 0.88. Table 1 shows baseline anthropometric, clinical, and laboratory characteristics of the study population according to each glycemic category. The IGT/high HbA1c group (n=876) was older (mean age, 54.8 years), had higher blood pressure (mean SBP/DBP 130/84 mm Hg), higher HbA1c (mean 5.9%), higher FPG (mean, 88 mg/dL), 2hPG (mean, 161 mg/dL), total cholesterol/HDL ratio (mean, 4.8), TG (median, 163 mg/dL), and HOMA-IR levels (median, 1.6) than the other groups (all P<0.05). Meanwhile, compared with isolated high HbA1c group (NGT/high HbA1c group; n=1,886), the isolated IGT group (IGT/normal HbA1c group; n=692) was younger (mean age, 51.5 years vs. 53.6 years, P<0.05), had lower proportion of current smokers (17.9% vs. 32.4%, P<0.05), higher monthly incomes (≥1,250 dollars/month, 55.6% vs. 44%, P<0.05), less physical activity (50.9% vs. 61.8%, P<0.05), lower mean total cholesterol/HDL ratio (4.4 vs. 4.6, P<0.05), and median HOMA-IR levels (1.4 vs. 1.5, P<0.05).
During the median follow-up period of 11.5 years (IQR, 6.0 to 11.8), 1,341 (17.6%) subjects were newly diagnosed with T2DM during follow-up, corresponding to incident density of 22.1 per 1,000 person-years. As expected, the cumulative incident density was the highest in the IGT/high HbA1c group (86.0 per 1,000 person-years) with a declining tendency in the NGT/normal HbA1c group (10.1 per 1,000 person-years) (Table 1).
Risk for incident diabetes
Both crude and adjusted HRs for incident diabetes were significantly highest in the IGT/high HbA1c group (crude model HR, 10.82; 95% CI, 9.35 to 12.51) (adjusted model HR, 9.94; 95% CI, 8.53 to 11.59) compared with the NGT/normal HbA1c group (reference group) (Table 2). In the subgroup comparison, the isolated IGT group was associated with higher risk (crude model HR, 4.41; 95% CI, 3.67 to 5.28) (adjusted model HR, 4.29; 95% CI, 3.56 to 5.17) than the isolated high HbA1c group (adjusted model HR, 2.79; 95% CI, 2.40 to 3.26; both P<0.05) (Table 2). Otherwise, considering the presence of central obesity, the higher HR in isolated IGT group compared with the isolated high HbA1c group was lost among obese individuals (adjusted model HR, 3.95; 95% CI, 3.01 to 5.18 in isolated IGT group vs. HR, 2.82; 95% CI, 2.30 to 3.46 in isolated high HbA1c group). When the obesity was defined by BMI cut-off value (25 kg/m2), meanwhile, the presence of obesity does not affect the risk of incident diabetes between subgroup with isolated IGT and those with isolated high HbA1c (Supplemental Table S1). Kaplan-Meier curves demonstrated a clear separation of the curves for isolated IGT vs. isolated high HbA1c among centrally non-obese population (log-rank P<0.001) but not among centrally obese population (Supplemental Fig. S1).
The predictive power of models including each glycemic parameter for incident DM
Whether or not the laboratory, metabolic, and socioeconomic confounding factors were included, high HbA1c level (model 2, C-index 0.753) or high 2hPG level (model 3, C-index 0.749) improved the prediction of incident T2DM compared with FPG test alone (model 1, C-index 0.684) (Table 3). Meanwhile, when we stratified subgroups according to the presence of central obesity, the higher predictive power of ‘FPG+OGTT’ model (model 3, C-index 0.785) compared with ‘FPG+HbA1c’ model (model 2, C-index 0.754) among centrally non-obese population was comparable to centrally obese population (C-index 0.754 vs. 0.749). The difference in discriminant ability between these subgroups according to the presence of obesity is similarly observed when BMI is applied as the criterion for obesity (C-index 0.75 for ‘FPG+HbA1c’ model vs. 0.77 for ‘FPG+OGTT’ model, P=0.031) (Supplemental Table S2).
Correlation between glycemic parameters according to the presence of central obesity
In the correlation analysis, the degree of correlation between 2hPG and HbA1c varied with the presence of central obesity. The correlation between HbA1c and 2hPG (converted to mmol/L) was stronger in the subgroup with centrally obese population (r=0.062) than in non-obese population (r=0.041, P<0.001) (Fig. 1). A stronger association between HbA1c and 2hPG in obese population was also observed when obesity was defined according to BMI criteria (r=0.060 [≥25 kg/m2] vs. r=0.037 [<25 kg/m2]; P<0.001). When an OGTT (2hPG) cutoff of 140 mg/dL was adopted to detect prediabetes, the overall sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of HbA1c criteria for prediabetes (5.7% to 6.4%) were 55.9%, 68.3%, 31.7%, and 85.4%, respectively. The sensitivity and PPV of HbA1c-criteria increased to 59.1% (vs. 49.4%) and 35.1% (vs. 29.8%) in obese individuals (n=2,105) compared with non-obese individuals (n=5,401) (Supplemental Table S3).
DISCUSSION
In this population-based cohort study, the different glycemic categories based on OGTT and HbA1c test showed that each glycemic parameter represent different clinical phenotypes. Individuals with high 2hPG had a higher risk of developing diabetes, regardless of the HbA1c level. In addition, 2hPG level had greater discriminatory capacity than HbA1c level. However, when diagnosed as prediabetes through HbA1c test, the number of subjects corresponding to actual IGT increased in obese population as the sensitivity of HbA1c test increases. As a result, the risk of incident diabetes and the predictive capacity between isolated IGT and isolated high HbA1c became comparable in obese population.
Different glycemic categories (measurements of FPG, 2hPG, and HbA1c) probably reflect diverse pathophysiological features of glucose metabolism and the different clinical, socioeconomic characteristics. In this study population, the subgroup with isolated IGT had fewer risk factors associated with diabetes compared with those carrying isolated high HbA1c levels (5.7% to 6.4%). Based on different clinical/anthropometric characteristics according to each diagnostic criteria, neither HbA1c test nor OGTT alone facilitated the detection of high-risk population and the accompanying risk factors to predict the risk of diabetes. Two tests (HbA1c and OGTT) can be used to measure different pathophysiological aspects of dysglycemia and a single positive result does not establish the accompanying risk factors and clinical phenotypes of diabetes.
The concordance between 2hPG and HbA1c levels was higher in obese population regardless of whether the BMI criterion or the WC criterion was adopted in our study. In addition, the sensitivity and PPV of HbA1c for prediabetes compared with OGTT also increased in obese individuals than in non-obese individuals. Increased oxidative stress in obesity [30] might explain elevated HbA1c levels at a given glycemic level in obese population [24,31]. In the Insulin Resistance Atherosclerosis Study [32], the sensitivity of HbA1c criteria (5.7% to 6.4%) for prediabetes increased with increasing BMI (67.8% in obese subjects [BMI ≥30 kg/m2] vs. 19.9% in overweight subjects [BMI 25.0 to 29.9 kg/m2]), but this trend was not observed in IGT criteria (60.0% in obese subjects vs. 58.1% in overweight subjects). In the Spanish non-diabetic community study, the prevalence of prediabetes was affected by BMI when the HbA1c criteria was applied but not in FPG criteria [33]. A pooled analysis of global surveillance of diabetes prevalence showed that the relationship between glucose-based and HbA1c-based prevalence of diabetes was partly related to participants’ age and the presence of obesity, and the prevalence based on HbA1c increased with age, national income, and mean BMI [34].
Interestingly, among centrally obese population, the risk of incident T2DM in prediabetes defined by HbA1c test alone became comparable to that of OGTT alone. In addition, the additional predictive performance of HbA1c and OGTT tests for diabetes was not different. Considering the general higher sensitivity of OGTT compared with HbA1c test in defining the high-risk population for diabetes, a higher accordance between HbA1c and 2hPG based on OGTT in obese subjects might increase the predictive power of HbA1c-based model compared with OGTT-based model. However, the findings do not warrant substitution of HbA1c for OGTT to determine the high risk for diabetes in obese population. Otherwise, it is necessary to recognize that the predictive power and the risk of diabetes based on HbA1c criteria may differ depending on the presence of central obesity. In interpreting the HbA1c test, an additional OGTT may be especially needed for risk analysis in non-obese subjects, even though the HbA1c is high (5.7% to 6.4%). However, if the obese subjects are considered at risk based on the HbA1c test, appropriate lifestyle modifications and drug therapy can be considered without additional confirmatory tests.
We acknowledge both the strengths and limitations of the study. This study has several strengths. A community-based longitudinal cohort study of middle-aged adults was conducted and the association between different glycemic parameters and incident diabetes was evaluated. In addition, information based on a large number of confounding factors including socioeconomic, anthropometric, and laboratory traditional risk factors were adjusted in the study. Glycemic parameters such as HbA1c test and OGTT were evaluated simultaneously and the performance of HbA1c was compared with 2hPG instead of FPG. Furthermore, based on this nationwide sample of urban/rural area population, subjects with anemia or hemoglobinopathies were excluded to validate the accuracy of HbA1c, which was affected by red cell metabolism.
However, the study findings from the Korean national cohort cannot be generalized to other races/ethnicities. We included subjects aged 40 to 69 years and excluded younger population. In addition, the longitudinal cohort observational study did not elucidate the causal relationship between different glycemic indices and incident diabetes. A further study is needed to evaluate the causal relationship.
In conclusion, the overall risk of future diabetes and the diagnostic performance of HbA1c test for identifying high risk subjects were inferior to OGTT, but became comparable among obese subjects as the sensitivity of HbA1c test increases. Depending on whether or not subjects are accompanied by central obesity, it may be helpful in determining whether an additional confirmatory test should be performed after the HbA1c test.
XML Download