In African-born Blacks living in America, we determined by BMI category 1) prevalence of abnormal glucose tolerance (Abnl-GT) and 2) diagnostic value and reproducibility of hemoglobin A1c (HbA1c), fructosamine, and glycated albumin (GA).
Participants (n = 416; male, 66%; BMI 27.7 ± 4.5 kg/m2 [mean ± SD]) had an oral glucose tolerance test with HbA1c, GA, and fructosamine assayed. These glycemic markers were repeated 11 ± 7 days later. Abnl-GT diagnosis required 0 h ≥5.6 mmol/L (≥100 mg/dL) and/or 2 h ≥7.8 mmol/L (≥140 mg/dL). Thresholds for HbA1c, GA, and fructosamine were the values at the 75th percentile for the population (39 mmol/mol [5.7%], 14.2%, and 234 μmol/L, respectively).
Abnl-GT prevalence in the nonobese was 34% versus 42% in the obese (P = 0.124). Reproducibility was excellent for HbA1c and GA (both κ ≥ 0.8), but moderate for fructosamine (κ = 0.6). Focusing on HbA1c and GA in the nonobese, we found as single tests the sensitivities of HbA1c and GA were 36% versus 37% (P = 0.529). Combining HbA1c and GA, sensitivity increased to 58% because GA identified 37% of Africans with Abnl-GT not detected by HbA1c (P value for both tests vs. HbA1c alone was <0.001). For the obese, sensitivities for HbA1c, GA, and the combined tests were 60%, 27%, and 67%, respectively. Combined test sensitivity did not differ from HbA1c alone (P = 0.25) because GA detected only 10% of obese Africans with Abnl-GT not detected by HbA1c.
Adding GA to HbA1c improves detection of Abnl-GT in nonobese Africans.
Introduction
Mathematical models predict that between 2019 and 2045, sub-Saharan Africa will experience a 143% increase in the prevalence of abnormal glucose tolerance (Abnl-GT) (1). This increase in Abnl-GT, which encompasses both prediabetes and type 2 diabetes (T2D), is the highest anticipated increase in the world (1). Slowing this upward trajectory requires strategies for the diagnosis of Abnl-GT that are feasible and effective in Africa.
Another challenge is that 60–80% of Africans living with Abnl-GT are undiagnosed (1). Hence, Africa has the highest prevalence in the world of people living with undiagnosed Abnl-GT (1). Even when undetected, complications from Abnl-GT progress. Furthermore, Abnl-GT is associated with greater susceptibility and higher mortality from infectious diseases, including tuberculosis and coronavirus disease 2019 (2,3). Therefore, lowering the prevalence of Abnl-GT may assist in decreasing the scope and consequences of several important infectious diseases. However, the International Diabetes Federation reports that <20% of African countries have in-country data on the prevalence of Abnl-GT (1).
The challenges of collecting data are magnified by the fact that routine tests used for Abnl-GT screening, such as fasting plasma glucose (FPG) and hemoglobin A1c (HbA1c), have poor diagnostic sensitivities (≤50%) in Africans (4,5). Thus, even when screening programs exist, Africans with Abnl-GT are often misdiagnosed as normal (6).
As a screening test, FPG is suboptimal. Fasting is difficult to achieve. People often travel many miles and wait for extended periods before being seen in clinics. In addition, people carry their food with them so they can easily eat when they are hungry. As going to a medical clinic and being fasted is not standard practice in many medical clinics in African countries, our focus is on the identification of effective nonfasting screening tests. Furthermore, blood samples kept at room temperature for even 30 min undergo extracorporeal glycolysis, resulting in the reporting of spuriously low glucose concentrations (7). HbA1c is also problematic because, even in the absence of factors that adversely affect it, such as nutritional deficiencies, hemoglobinopathies, and anemia, HbA1c detects <50% of Africans with Abnl-GT (6). The diagnostic sensitivity of HbA1c is also <50% in African Americans, Whites, Hispanics, and Arab immigrants to the U.S. (8–10).
Due to the inadequate diagnostic performance of FPG and HbA1c, attention has turned to fructosamine and glycated albumin (GA), both of which are nonfasting markers of glycemia. Fructosamine reflects the concentration of all circulating glycated proteins, including GA, which is formed by the nonenzymatic attachment of glucose to albumin (11).
A previous study of 236 African-born Blacks enrolled in the Africans in America study revealed that combining GA with HbA1c doubled the detection of prediabetes in the nonobese (12). The prevalence of Abnl-GT in the nonobese needs special attention because in low- and middle-income countries globally and in Africa, the prevalence of T2D is rapidly rising in the nonobese (1,13,14). To pursue improved detection of hyperglycemia in Africans, we increased recruitment to the Africans in America cohort and broadened our detection goal from prediabetes to Abnl-GT. With this larger cohort of 416 African-born Blacks living in America, our objectives were to determine by BMI category: 1) the prevalence of Abnl-GT, and 2) the diagnostic value and reproducibility of HbA1c, GA, and fructosamine.
Research Design and Methods
Population
The Africans in America cohort assesses the cardiometabolic health of African-born Blacks living in the U.S. (15,16). Recruitment is by newspaper advertisements, flyers, community event presentations, and relevant websites. The National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Institutional Review Board (ClinicalTrials.gov identifier NCT00001853; Bethesda, MD) approved the study. Written informed consent is obtained prior to enrollment.
During a telephone screening interview, potential enrollees must state they were born in sub-Saharan Africa to two Black parents who were also born in sub-Saharan Africa. Additionally, they must self-identify as healthy and deny a history of diabetes.
A total of 451 African-born Blacks living in metropolitan Washington, DC, successfully completed the telephone interview and had an outpatient screening visit at the National Institutes of Health (NIH) Clinical Center (Supplementary Fig. 1). A history, physical, electrocardiogram, and routine blood tests were performed. Thirty-three individuals did not proceed to an oral glucose tolerance test (OGTT). Exclusion criteria were: anemia (n = 10), elevated liver transaminases (n = 1), hypothyroidism (n = 1), intravenous access issues (n = 4), and scheduling conflicts (n = 17). Two individuals were excluded after the OGTT because hemoglobin electrophoresis revealed hemoglobin type AF. The percent hemoglobin F in one participant was 15% and for the other person 25%. Hemoglobin F at these levels interferes with the determination of HbA1c by the high-performance liquid chromatography (HPLC) used in this study.
OGTTs
OGTTs were performed in 416 individuals (male, 66%; aged 39 ± 10 years [mean ± SD], range 20–65 years; BMI 27.7 ± 4.5 kg/m2, range 18.2–42.2 kg/m2).
After a 12-h fast, participants came to the NIH Clinical Center at 7:00 a.m. On arrival, women provided a urine sample for pregnancy testing, and all were negative. As described previously, weight, height, and waist circumference (WC) were measured (15). Weight was obtained with a calibrated digital scale (Scale-Tronix 5702; Welch Allyn, Skaneateles Falls, NY). Height was measured with a wall stadiometer (Seca 242; Seca Corp., Hanover, MD). BMI was calculated as weight in kilograms divided by height in meters squared. BMI categories were defined according to World Health Organization guidelines (17). WC was measured at the superior border of the iliac crest at the end of expiration using a stretch-resistant tape measure with the person standing with feet hip-width apart and weight evenly distributed. The mean of three values was recorded.
Baseline blood samples were obtained for HbA1c, hemoglobin electrophoresis, glucose, insulin, GA, and fructosamine measurements. This was followed by an OGTT (TRUTOL 75 g; Custom Laboratories) with samples taken at 0.5, 1, and 2 h for glucose and insulin concentrations. Abnl-GT was defined as FPG ≥5.6 mmol/L and/or 2-h glucose ≥7.8 mmol/L (1).
After the OGTT, a computed tomographic scan (Siemens and SOMATOM Force Scanner) was performed to measure visceral adipose tissue (VAT).
Reproducibility of HbA1c, GA, and Fructosamine Values
To determine the reproducibility of HbA1c, GA, and fructosamine, 36% (150 out of 416) of enrollees returned to the Clinical Center 11 ± 7 days after the OGTT.
Participants were divided into two groups (A and B) (Supplementary Fig. 1).
Group A comprised the first 281 consecutively enrolled individuals. In group A, repeat studies were planned only if the initial OGTT met the glucose criteria for T2D (1). Seventeen out of the 18 individuals newly diagnosed with T2D returned for a second visit, and 1 declined. All 17 individuals had HbA1c levels determined, but GA and fructosamine concentrations were available only in the last 5 consecutively enrolled participants.
Group B consisted of the 135 subsequently enrolled individuals. All were invited, independent of glucose tolerance status at the initial OGTT, for a repeat study. Two individuals declined repeat studies. For one individual who did return, the blood sample obtained for HbA1c clotted. Therefore, for group B, duplicate HbA1c, GA, and fructosamine were available for 132, 133, and 133 individuals, respectively.
Overall, repeat values were available for HbA1c, GA, and fructosamine in 149 (17 plus 132), 138 (5 plus 133), and 138 (5 plus 133) individuals, respectively.
Metabolic Parameters
Assays
Glucose and hs-CRP were measured in plasma and insulin in serum with a Roche Diagnostics cobas 6000 analyzer.
Hemoglobin, hematocrit, white blood cells, and mean corpuscular volume were measured in EDTA-anticoagulated whole blood using a Sysmex XE-5000.
HbA1c by HPLC
HbA1c values were determined with two different NGSP-certified instruments using HPLC technology manufactured by BioRad Laboratories. HbA1c samples from the first 139 enrollees were measured on the VARIANT II instrument. The next 277 participants had HbA1c measurements performed on a D-10 instrument. The correlation (R2) and mean bias between the VARIANT II and D-10 instruments were 0.9934 and 0.07 (1.21%), respectively.
Fructosamine and GA
Fructosamine and GA were measured in plasma on the cobas 6000. For fructosamine, a colorimetric nitroblue tetrazolium assay was used. Interassay coefficient of variation for fructosamine was 2.9% at 308 µmol/L and 2.6% at 521 µmol/L (12). For GA, the Lucica GA-L enzymatic assay, provided by Asahi Kasei Pharma Corporation (Tokyo, Japan), was used. Albumin was measured with bromocresol purple. GA is reported as percent of albumin concentration. The interassay coefficient of variation for GA was 1.6% at 15.6% and 1.8% at 35.2%.
Determination of Diagnostic Thresholds for HbA1c, GA, and Fructosamine
For detecting Abnl-GT by HbA1c, the standard threshold of 39 mmol/mol (5.7%) was used (1). To identify a diagnostic threshold for GA and fructosamine, the procedure established by the Atherosclerosis Risk in Communities (ARIC) investigators was followed (20). As the diagnostic threshold for detecting Abnl-GT using HbA1c was 39 mmol/mol (5.7%) and corresponded to the upper quartile of our population distribution, we chose the upper quartile for GA (14.2%) and fructosamine (234 μmol/L).
Statistical Analyses
Unless otherwise stated, data are presented as mean ± SD. Analyses included one-way ANOVA with Bonferroni corrections for multiple comparisons, χ2 tests, unpaired t tests, McNemar test for matched pairs, Net Reclassification Improvement (NRI), and κ-statistic for diagnostic reproducibility. κ-Statistic categories were: slight (0–0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), and excellent (0.81–1.0) (21). P values ≤0.05 were considered significant. Analyses were performed with Stata 16.
Results
The African regions of origin of the participants were: West 53% (220 out of 416), Central 18% (74 out of 416), and East 29% (12 out of 416). Three participants from Southern African countries were included in the Central African group. Characteristics by African region of origin are provided in Supplementary Table 1. Of note, there was no difference by African region in sex distribution, age, or age at immigration. West Africans had resided in the U.S. the longest, 14 ± 11 years (P = 0.009). Hemoglobin did not differ by African region. The prevalence of heterozygous hemoglobinopathies (i.e., sickle cell trait and hemoglobin C trait) were highest in West and Central Africa (both P = 0.002).
There was no difference by African region in body size, glucose tolerance, insulin resistance, or insulin secretion (Supplementary Table 1). As these parameters did not vary significantly by African region of origin, participants from West, Central, and East Africa were combined into a single group and evaluated by BMI category and glucose tolerance status (Table 1).
Participant characteristics according to BMI category1
Parameter2 . | Total (n = 416) . | Nonobese (n = 303) (73%) . | Obese (n = 113) (27%) . | P value3 . |
---|---|---|---|---|
Age (years) | 39 ± 10 | 38 ± 10 | 41 ± 10 | 0.001 |
BMI (kg/m2) | 27.7 ± 4.5 | 25.5 ± 2.7 | 33.5 ± 2.9 | <0.001 |
BMI if Abnl-GT present (kg/m2) | 28.8 ± 4.5 | 26.3 ± 2.3 | 34.1 ± 3.1 | <0.001 |
WC (cm) (n = 415) | 91 ± 12 | 86 ± 9 | 103 ± 9 | <0.001 |
VAT (cm3) (n = 406) | 99 ± 69 | 83 ± 61 | 143 ± 71 | <0.001 |
Fasting glucose (mmol/L) | 5.1 ± 0.8 | 5.1 ± 0.5 | 5.4 ± 1.2 | <0.001 |
Glucose at 2 h (mmol/L) | 7.3 ± 2.4 | 7.1 ± 2.1 | 7.9 ± 2.9 | 0.001 |
AUC-glucose (n = 414) | 544 ± 128 | 533 ± 109 | 572 ± 166 | 0.007 |
ISI (n = 412) | 5.43 ± 3.64 | 6.01 ± 3.73 | 3.89 ± 2.87 | <0.001 |
Insulin secretion (n = 412) | 0.54 ± 0.36 | 0.49 ± 0.30 | 0.67 ± 0.45 | <0.001 |
Oral Disposition Index (n = 412) | 2.25 ± 0.99 | 2.34 ± 0.98 | 2.01 ± 0.97 | 0.002 |
Abnl-GT (%) | 37 (152/416) | 34 (104/303) | 42 (48/113) | 0.124 |
Diabetes (%) | 7 (28/416) | 5 (16/303) | 11 (12/113) | 0.053 |
Prediabetes (%)4 | 32 (124/388) | 31 (88/287) | 36 (36/101) | 0.356 |
Parameter2 . | Total (n = 416) . | Nonobese (n = 303) (73%) . | Obese (n = 113) (27%) . | P value3 . |
---|---|---|---|---|
Age (years) | 39 ± 10 | 38 ± 10 | 41 ± 10 | 0.001 |
BMI (kg/m2) | 27.7 ± 4.5 | 25.5 ± 2.7 | 33.5 ± 2.9 | <0.001 |
BMI if Abnl-GT present (kg/m2) | 28.8 ± 4.5 | 26.3 ± 2.3 | 34.1 ± 3.1 | <0.001 |
WC (cm) (n = 415) | 91 ± 12 | 86 ± 9 | 103 ± 9 | <0.001 |
VAT (cm3) (n = 406) | 99 ± 69 | 83 ± 61 | 143 ± 71 | <0.001 |
Fasting glucose (mmol/L) | 5.1 ± 0.8 | 5.1 ± 0.5 | 5.4 ± 1.2 | <0.001 |
Glucose at 2 h (mmol/L) | 7.3 ± 2.4 | 7.1 ± 2.1 | 7.9 ± 2.9 | 0.001 |
AUC-glucose (n = 414) | 544 ± 128 | 533 ± 109 | 572 ± 166 | 0.007 |
ISI (n = 412) | 5.43 ± 3.64 | 6.01 ± 3.73 | 3.89 ± 2.87 | <0.001 |
Insulin secretion (n = 412) | 0.54 ± 0.36 | 0.49 ± 0.30 | 0.67 ± 0.45 | <0.001 |
Oral Disposition Index (n = 412) | 2.25 ± 0.99 | 2.34 ± 0.98 | 2.01 ± 0.97 | 0.002 |
Abnl-GT (%) | 37 (152/416) | 34 (104/303) | 42 (48/113) | 0.124 |
Diabetes (%) | 7 (28/416) | 5 (16/303) | 11 (12/113) | 0.053 |
Prediabetes (%)4 | 32 (124/388) | 31 (88/287) | 36 (36/101) | 0.356 |
Nonobese: BMI <30.0 kg/m2; obese: BMI ≥30.0 kg/m2.
Data are mean ± SD or percentages.
Comparisons were by unpaired t tests for continuous variables and χ2 for categorical variables.
Denominators are all individuals without diabetes (normal glucose tolerance and prediabetes).
Metabolic Characteristics by BMI Category
Seventy-five percent were nonobese (303 out of 416) (BMI <30 kg/m2) and 25% obese (113 out of 416) (BMI ≥30 kg/m2). The nonobese were younger than the obese participants (38 ± 10 vs. 41 ± 10; P = 0.001). In addition, the nonobese had lower BMI, lower WC, and less VAT (Table 1). Raw data are presented in Table 1; even after adjustment for age, the significant differences between the nonobese and obese remained.
Additionally, FPG, 2-h glucose, and AUC-glucose were lower in the nonobese group. Similarly, the nonobese were less insulin resistant, had lower insulin secretion, and a higher Oral Disposition Index.
However, the prevalence of Abnl-GT (34% vs. 42%; P = 0.124) was similar in the nonobese and obese, respectively; and the absolute number of nonobese Africans with Abnl-GT (n = 104) was nearly double the number of obese with Abnl-GT (n = 48). Among the nonobese who had Abnl-GT, 30% (31 out of 104) had a BMI <25 kg/m2, and 70% (73 out of 104) had a BMI ≥25.0 kg/m2 and <30 kg/m2.
Reproducibility of Nonfasting Markers of Glycemia
Reproducibility by κ-statistic for HbA1c, GA, and fructosamine was 0.85, 0.83, and 0.60, respectively. This degree of reproducibility is excellent for HbA1c and GA, but only moderate for fructosamine (21). Therefore, only results for HbA1c and GA are presented.
HbA1c and GA Individually and Combined
The term GA-alone represents the added value provided by GA. Some case subjects with Abnl-GT were detected by HbA1c and not GA (Fig. 1, in red). Some case subjects were detected by GA and not HbA1c (Fig. 1, in blue). Some were identified by both tests (Fig. 1, in purple). GA-alone refers to the individuals with Abnl-GT detected by GA and not HbA1c (Fig. 1, in blue).
Successful diagnostic test by BMI category in participants with Abnl-GT. The diagnostic color coding: red for HbA1c, blue for GA, and purple for both tests.
Successful diagnostic test by BMI category in participants with Abnl-GT. The diagnostic color coding: red for HbA1c, blue for GA, and purple for both tests.
Sensitivities and Specificities
Nonobese
Sensitivity for the diagnosis of Abnl-GT by HbA1c and GA was similar, 36% and 37%, respectively (P = 0.999). The sensitivity of the combined tests was 58%, which was significantly greater than HbA1c alone (P < 0.001) (Fig. 2A and Supplementary Table 2). Sensitivity was higher for the combined tests because GA-alone identified 37% (22 out of 60) of the nonobese not detected by HbA1c (Fig. 1). Specificities for the diagnosis of Abnl-GT singly by HbA1c and by GA and then both combined were 80%, 75%, and 60%, respectively (Supplementary Table 2). The NRI for HbA1c plus GA versus HbA1c alone was 0.238 (95% CI 0.018, 0.459; P = 0.034).
Sensitivities for the diagnosis of Abnl-GT by BMI category for HbA1c and GA singly and combined. A: Nonobese: sensitivities for HbA1c, 36%; GA, 37%; and HbA1c plus GA, 58%. Corresponding specificities were 80%, 75%, and 60%, respectively. B: Obese: sensitivities for HbA1c, 60%; GA, 27%; and HbA1c plus GA, 67%. Corresponding specificities were 80%, 94%, and 74%, respectively. C: Whole cohort: sensitivities for HbA1c, 43%; GA, 34%; and HbA1c plus GA, 61%. Corresponding specificities were 80%, 80%, and 63%, respectively. Data are mean (95% CI). ***P < 0.001.
Sensitivities for the diagnosis of Abnl-GT by BMI category for HbA1c and GA singly and combined. A: Nonobese: sensitivities for HbA1c, 36%; GA, 37%; and HbA1c plus GA, 58%. Corresponding specificities were 80%, 75%, and 60%, respectively. B: Obese: sensitivities for HbA1c, 60%; GA, 27%; and HbA1c plus GA, 67%. Corresponding specificities were 80%, 94%, and 74%, respectively. C: Whole cohort: sensitivities for HbA1c, 43%; GA, 34%; and HbA1c plus GA, 61%. Corresponding specificities were 80%, 80%, and 63%, respectively. Data are mean (95% CI). ***P < 0.001.
Obese
For the diagnosis of Abnl-GT, the sensitivity of HbA1c at 60% was greater than the sensitivity of 27% for GA (P < 0.001). As GA-alone detected only 10% (3 out of 32) of the obese not detected by HbA1c, combining HbA1c with GA did not improve detection of Abnl-GT (P = 0.250) (Fig. 2B and Supplementary Table 2). Specificities for the diagnosis of Abnl-GT by HbA1c and GA singly and combined were: 80%, 94%, and 74%, respectively (Supplementary Table 2). The NRI for HbA1c versus GA versus HbA1c alone was 0.419 (95% CI 0.141, 0.696; P = 0.003).
Total Cohort
Diagnostic sensitivities of HbA1c and GA were 43% versus 34% (P = 0.086). At 61%, the sensitivity of the combined tests was significantly greater than when only HbA1c was used (P < 0.001) (Fig. 2C and Supplementary Table 2). By identifying 28% (26 out of 92) of Africans with Abnl-GT not detected by HbA1c, GA-alone contributed to the higher sensitivity of the combined tests in the nonobese category. Specificities for the diagnosis of Abnl-GT by HbA1c and GA singly and combined were 80%, 80%, and 63%, respectively (Supplementary Table 2). The NRI for HbA1c and GA versus HbA1c alone was 0.270 (95% CI 0.091, 0.448; P = 0.003).
Characteristics of Individuals With Abnl-GT Diagnosed by GA-Alone Versus HbA1c
Participants with Abnl-GT detected by GA-alone were younger than those detected by HbA1c (41 ± 9 vs 47 ± 10 years; P = 0.015). BMI, WC, and VAT were also lower (Fig. 3). Differences in body size did not change with age adjustment; therefore, raw data are presented (Fig. 3 and Supplementary Table 3). In contrast to body size measurements, insulin resistance, insulin secretion, and Disposition Index did not differ by diagnostic test (Supplementary Table 3). Similarly, prevalence of heterozygous hemoglobinopathy (sickle cell trait or hemoglobin C trait) (27% vs. 24%; P = 0.789) and albumin (4.04 ± 0.25 vs. 3.98 ± 0.28; P = 0.367) did not differ by diagnostic test (Supplementary Table 3).
Participant characteristics with Abnl-GT according to diagnostic test: A: BMI, B: WC, C: VAT. In each panel: 1 is HbA1c or HbA1c and GA; 2 is GA-alone. Data presented as mean ± SE. *P < 0.05; ***P < 0.001.
Participant characteristics with Abnl-GT according to diagnostic test: A: BMI, B: WC, C: VAT. In each panel: 1 is HbA1c or HbA1c and GA; 2 is GA-alone. Data presented as mean ± SE. *P < 0.05; ***P < 0.001.
Conclusions
This investigation made three findings that could influence the approach to Abnl-GT in Africa. First, there were significant differences by BMI category in the diagnostic capabilities of GA and HbA1c such that the combination leads to improved detection of Abnl-GT in the nonobese. Second, the prevalence of Abnl-GT was similar in nonobese and obese Africans. Third, in contrast to fructosamine, both HbA1c and GA provided highly reproducible results.
Overall, the combination of GA and HbA1c identified more Africans with Abnl-GT than HbA1c. This was because GA increased detection of the nonobese with Abnl-GT by 33% (Fig. 1). As adding GA did not improve the detection of Abnl-GT in the obese, the use of GA could be reserved for the nonobese.
The prevalence of Abnl-GT in African immigrants to the U.S. was 37% and similar in the nonobese and obese. Therefore, clinicians caring for Africans should not identify an individual as “low risk” because they are nonobese (15,22). In short, the threshold for BMI-related risk for Abnl-GT may be lower in Africans than for African Americans, which suggests the need for tailored therapeutic approaches (23). Furthermore, Abnl-GT in the nonobese is observed in India and in many other low- and middle-income countries globally (13,14).
Reproducibility of diagnostic markers is another important finding. With studies done 11 ± 7 days apart and a κ-statistic ≥0.8 for both HbA1c and GA, their reproducibility was excellent. Hence, the biological variability and the interassay variation for both HbA1c and GA are low. In contrast, the fructosamine assay is less optimal and subject to more analytic variables than GA (24). Therefore, it is not surprising that duplicate fructosamine studies revealed only moderate reproducibility (κ-statistic of 0.6). Suboptimal reproducibility may explain why for Abnl-GT, fructosamine is a poor diagnostic test (25).
Relationship Between GA and BMI
Africans with Abnl-GT detected by GA-alone had lower BMI, WC, and VAT than their counterparts detected by HbA1c (Fig. 3). This is consistent with the observation that GA correlates inversely with BMI, WC, and VAT (12,26). In fact, He et al. (27) found that for every 1 kg/m2 increase in BMI, GA decreased by 0.13%. This inverse relationship is often attributed to increased catabolism of albumin from obesity-related chronic inflammation or insulin resistance (28,29). Interestingly, degree of insulin resistance and β-cell function did not differ by diagnostic test (Supplementary Table 3). Therefore, the etiology of Abnl-GT (insulin resistance vs. relative β-cell failure) cannot be ascertained by whether diagnosis was by HbA1c or GA.
BMI Category to Guide Use of GA
In detecting Abnl-GT, the diagnostic utility of GA appears to depend on BMI category. We speculate that for normal-weight people, GA-alone may be sufficient to detect Abnl-GT. In overweight people, GA should be combined with HbA1c, while HbA1c may be satisfactory in obese people.
Studies from East Asia reveal that the performance of GA as a single diagnostic test was equivalent to or better than HbA1c (30,31). GA may be an effective diagnostic test for Abnl-GT because the BMI in East Asians with T2D is typically <25 kg/m2 (31).
In the Africans in America cohort, two-thirds of the nonobese participants with Abnl-GT were overweight (BMI 25.0–29.9 kg/m2), and one-third were normal weight (<25.0 kg/m2). Considering that the Africans in America cohort represents mainly the overweight with Abnl-GT, we found that combining HbA1c with GA optimized detection.
For obese people with Abnl-GT, HbA1c alone may be sufficient. In a South African study of mixed-ancestry adults in whom the mean BMI of the group with T2D was 32.5 kg/m2, Zemlin et al. (32) reported that the combination of GA and HbA1c was no better than HbA1c alone. Our findings were similar in the obese participants in the Africans in America cohort (Fig. 2B). However, NRI suggests that the combined tests may be beneficial in the obese as well (Supplementary Table 4). While this would be an excellent development, there is concern that the assumptions made in the calculation of NRI may be overly optimistic in predicting improvement (33).
Feasibility of Obtaining GA
The enzymatic method used to measure GA has been evaluated in many clinical studies and is approved for use in the U.S., Japan, Korea, Indonesia, and China (11,30). The assay is reproducible, precise, and easily performed on automated analyzers that can measure glucose or electrolytes. Analyzers of this type are widely available in both clinical and research settings in Africa (M. Nyirenda, personal communication).
Methodology for Determining Diagnostic Thresholds for GA
Thresholds for FPG, 2-h glucose, and HbA1c were determined by their relationship to diabetic retinopathy (34). The methods used to determine diagnostic thresholds for GA rely on glucose concentrations obtained during the OGTT. The two most common approaches are: 1) identifying the ability of GA to predict Abnl-GT by calculating the area under the receiver operating characteristic curve and applying the Youden Index to define the optimal cut point (30–32), and 2) use the GA cutoff at the upper 75th, 95th, or 97.5th percentile of the population being evaluated (20). For example, the ARIC investigators determined the upper fraction of their cohort who had HbA1c ≥5.7%. As 5.7% occurred as the cutoff for the 75th percentile, they used the GA threshold at the 75th percentile (20). With these two approaches, GA thresholds range from 13 to 16% depending on whether the outcome is prediabetes, T2D, or a combination (i.e., Abnl-GT) (20,30–32). We used the approach taken by the ARIC investigators, and our threshold for Abnl-GT was 14.2%.
To systematically resolve which diagnostic threshold to use for GA, the way forward may be a two-step process with the establishment of an International Working Group followed by an International Consensus Panel. The International Working Group could survey existing studies and determine: 1) end point criteria for GA (retinopathy or glucose or other), 2) the statistical methodology for determining threshold (AUC–receiver operating characteristic or percentile), and 3) size and diversity of the cohorts needed to test recommendations. Within these cohorts, BMI and race/ethnicity would be specifically considered. If the International Working Group decides that current data are insufficient, parameters for study design for future determination by International Consensus Panel could be set.
Strengths and Limitations
Our study is the largest exploring the diagnostic value of nonfasting markers of glycemia in African-born Blacks. In addition, it is the first investigation to explore the diagnostic reproducibility of GA, fructosamine, and HbA1c. Furthermore, studies evaluating the diagnostic efficacy of these markers often state they cannot rule out the effect of confounding factors (6). However, our study provides documentation of the performance of GA, fructosamine, and HbA1c in the absence of cirrhosis, thyroid disease, hemoglobinopathies, anemia, nutritional deficiencies, hypoalbuminemia, and renal failure.
The study used a convenience sample. However, for three reasons, the Africans in America cohort appears to be representative of African-born Blacks living in the U.S. First, as most participants were from West African countries, the sample size of 416 was large enough to reflect known immigration patterns (35). Second, compared with East Africa, the prevalence of sickle cell trait and hemoglobin C trait was higher in West and Central Africans. Therefore, the sample size was large enough to detect known genetic differences by African region of origin (36). Third, the prevalence of T2D was 7%, which was comparable to the 8% prevalence of T2D in African-born Blacks living in Canada (37). Similar data are not available in the U.S.
The potential limitations of our investigation are intrinsic to our study design. One limitation is the applicability of our results to populations in African countries. However, we designed the study to use resources available in America to provide a proof of concept about the potential value of GA in Africans. Therefore, our investigation provides justification: 1) for the funding of studies to be conducted in Africa, and 2) for the study to be used for power analyses for population-based prospective studies that have optimization of GA and HbA1c thresholds built into their study design. To assist in these analyses, we provided the total number of participants with Abnl-GT and then for the two components of Abnl-GT, diabetes and prediabetes (Table 1). The current study was not large enough to examine diabetes and prediabetes separately.
Another challenge for African countries is that in the absence of point-of-care options for GA, initial studies will have to be conducted in urban or semiurban areas where there is access to clinics that have both automated analyzers and the opportunity to detect confounders such as nutritional deficiencies, infections, and hemoglobinopathies.
In addition, we focused on sensitivity over specificity. Results for specificity for each testing paradigm are provided only in the legend to Fig. 2 and Supplementary Table 2. However, the sensitivity of HbA1c as a single test in the nonobese is <40%. Therefore, HbA1c is not a viable option for effective screening for a disease that is reaching epidemic proportions in Africa. After improved detection is achieved, future studies can determine if the benefits of more optimal detection offset the lower specificity.
Conclusion
Between 2019 and 2045, a steep increase in Abnl-GT prevalence in Africa, especially in the nonobese, is anticipated. Innovative and improved diagnostic tools may be the best way to limit the epidemic, improve data collection, and inform treatment paradigms. GA is a nonfasting, easily obtainable, highly reproducible test that, in combination with HbA1c, provides valuable diagnostic information about Abnl-GT in nonobese Africans living in the U.S. Data from the Africans in America cohort provide a proof of concept and can serve as primary data for sample size calculations for population-based prospective studies in Africa. Research on the diagnostic value of GA may lead to better screening, earlier interventions, and ultimately less medical and social consequences from Abnl-GT.
Clinical trial reg. no. NCT00001853, clinicaltrials.gov
A.F.H. and N.H.O.-T. are co-first authors.
This article contains supplementary material online at https://doi.org/10.2337/figshare.12690023.
Article Information
Acknowledgments. The authors thank Paulina Stallcup, Clinical Center, National Institutes of Health, Bethesda, MD, for the laboratory expertise, commitment, and time.
Funding. The study was funded by the intramural research program of two NIH institutes, NIDDK and National Institute on Minority Health and Health Disparities (NIMHD), and the NIH Clinical Center. A.F.H. and A.E.S. are supported by the intramural programs of both NIDDK and NIMHD. N.H.O.-T., T.H., E.M.S., C.W.D., L.S.M., J.H., A.S., and S.T.C. are supported by the intramural program of NIDDK. D.B.S. is supported by the NIH Clinical Center.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. A.F.H., N.H.O.-T., and A.E.S. did the literature search. D.B.S. and A.E.S.designed the study. A.F.H., N.H.O.-T., T.H., E.M.S., C.W.D., L.S.M., and A.E.S. contributed to enrollment. A.F.H., N.H.O.-T., T.H., E.M.S., C.W.D., L.S.M., S.T.C., and A.E.S. collected the data. A.F.H., N.H.O.-T., T.H., E.M.S., J.H., A.S., S.T.C., D.B.S., and A.E.S. analyzed the data. A.F.H., N.H.O.-T., and A.E.S. made the figures. A.F.H., N.H.O.-T., E.M.S., and A.E.S. made the tables. A.F.H., N.H.O.-T., and A.E.S. wrote the first draft. A.F.H., N.H.O.-T., T.H., E.M.S., C.W.D., L.S.M., J.H., A.S., S.T.C., D.B.S., and A.E.S. provided critical rewrites of the manuscript. A.E.S. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation. This study was presented in abstract form at the 79th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 7–11 June 2019.