Burden of Diabetes and First Evidence for the Utility of HbA1c for Diagnosis and Detection of Diabetes in Urban Black South Africans: The Durban Diabetes Study

Objective Glycated haemoglobin (HbA1c) is recommended as an additional tool to glucose-based measures (fasting plasma glucose [FPG] and 2-hour plasma glucose [2PG] during oral glucose tolerance test [OGTT]) for the diagnosis of diabetes; however, its use in sub-Saharan African populations is not established. We assessed prevalence estimates and the diagnosis and detection of diabetes based on OGTT, FPG, and HbA1c in an urban black South African population. Research Design and Methods We conducted a population-based cross-sectional survey using multistage cluster sampling of adults aged ≥18 years in Durban (eThekwini municipality), KwaZulu-Natal. All participants had a 75-g OGTT and HbA1c measurements. Receiver operating characteristic (ROC) analysis was used to assess the overall diagnostic accuracy of HbA1c, using OGTT as the reference, and to determine optimal HbA1c cut-offs. Results Among 1190 participants (851 women, 92.6% response rate), the age-standardised prevalence of diabetes was 12.9% based on OGTT, 11.9% based on FPG, and 13.1% based on HbA1c. In participants without a previous history of diabetes (n = 1077), using OGTT as the reference, an HbA1c ≥48 mmol/mol (6.5%) detected diabetes with 70.3% sensitivity (95%CI 52.7–87.8) and 98.7% specificity (95%CI 97.9–99.4) (AUC 0.94 [95%CI 0.89–1.00]). Additional analyses suggested the optimal HbA1c cut-off for detection of diabetes in this population was 42 mmol/mol (6.0%) (sensitivity 89.2% [95%CI 78.6–99.8], specificity 92.0% [95%CI: 90.3–93.7]). Conclusions In an urban black South African population, we found a high prevalence of diabetes and provide the first evidence for the utility of HbA1c for the diagnosis and detection of diabetes in black Africans in sub-Saharan Africa.


Introduction
Sub-Saharan Africa (SSA) is experiencing a dramatic increase in diabetes. A consequence of rapid demographic and epidemiological transitions, the number of people with diabetes is projected to more than double to 34.2 million by 2040 [1,2]. An estimated 66.7% of people living with diabetes in SSA are undiagnosed and therefore more at risk of developing harmful and costly complications, the highest proportion of any region in the world [1]. This poses a huge challenge in many SSA countries where over-burdened and under-resourced health systems already have a shortfall of diabetes services [3,4].
Consistent and comparable measures of glycaemia are important for accurate screening and diagnosis of diabetes and for population-level surveillance, including inter-and intra-population prevalence comparisons, and subsequent targeting of services and resources to high-risk populations. Glycated haemoglobin (HbA 1c ) is recommended as an additional tool to glucosebased measures (fasting plasma glucose [FPG] and 2-hr plasma glucose (2PG) during an oral glucose tolerance test [OGTT]) for the diagnosis of diabetes [5][6][7]. However, HbA 1c can provide different diabetes prevalence estimates and identifies a different population as having diabetes compared with FPG and OGTT. This degree of discordance varies between populations, by ethnicity, and according to the burden of clinical conditions affecting HbA 1c, including anaemias, haemoglobinopathies and infection, potentially limiting the utility of HbA 1c for the diagnosis and detection of diabetes [8][9][10]. However, this has not been established in black sub-Saharan African populations.
Given the potential advantages of using HbA 1c for the diagnosis and detection of diabetes in the SSA context [11,12], evidence on the utility of HbA 1c in SSA populations is needed. We therefore assessed the diabetes prevalence estimates, association with established risk factors, and the diagnosis and detection of diabetes based on HbA 1c , FPG, and OGTT in a black South African population.

Study design
The Durban Diabetes Study (DDS) was a population-based cross-sectional study of individuals aged >18 years, who were not pregnant, and residing in urban black African communities in Durban (eThekwini municipality) in KwaZulu-Natal (South Africa), conducted between November 2013 and December 2014. A detailed description of the survey design and procedures has been previously published [13]. Written informed consent was obtained from all participants. The DDS was approved by the Biomedical Research Ethics Committee at the University of KwaZulu-Natal (reference: BF030/12) and the UK National Research Ethics Service (reference: 14/WM/1061).

Data collection
A detailed questionnaire, adapted from the standardised World Health Organization (WHO) STEPwise approach to Surveillance (STEPS) tool, including information on participant health, lifestyle, and socioeconomic indices was administered by trained study personnel [14]. Family history of diabetes was defined as history of diabetes in first-degree relatives. Current smokers were defined as currently smoking any tobacco product even if not daily. Current alcohol users were defined as having consumed any alcoholic beverage in the last month. Physical activity included both work-related and leisure-time activity and included any combination of walking, moderate, or vigorous intensity activities. Low physical activity was defined as doing physical activity on less than five days a week and for less than 600 metabolic equivalents (METs)-min per week. Low fruit and vegetable consumption was defined as fewer than five servings of fruit or vegetables a day [15].
Weight, height, waist circumference, and hip circumference were measured. Three blood pressure readings were obtained with a calibrated automatic electronic device and taken at least five minutes apart by trained study personnel. The mean of the last two readings was used for analysis. Body mass index (BMI) was used as a measure of total body obesity, and waist circumference and waist-to-hip ratio were used as measures of abdominal obesity. Standard WHO criteria were used to define raised blood pressure and obesity [16,17].
Blood samples were drawn following an overnight fast and were obtained, stored, and tested according to the standard WHO methodology [5,6]. For the OGTT, venous blood samples were collected, in NaF blood tubes, before and 2 hours after ingestion of 75g glucose monohydrate dissolved in 250 ml water for measurement of plasma glucose. In addition, fasting samples were obtained for HbA 1c , in EDTA whole blood tubes, and serum lipids, in plain serum tubes. Blood samples were stored in cold boxes maintained at 4-8°C until transported to a laboratory within six hours of collection. Plasma glucose was measured using the glucose oxidase method (ABBOTT ARCHITECT 2: CI 8200, Abbott Laboratories, Chicago, IL, USA). HbA 1c was measured using ion-exchange high-performance liquid chromatography (HPLC) (BIORAD VARIANT II TURBO 2.0, Bio-Rad Laboratories, Inc., Hercules, CA, USA), using an instrument certified by the National Glycohaemoglobin Standardization Program (NGSP) and International Federation of Clinical Chemistry and Laboratory Medicine (IFCC). The BIORAD VARIANT II TURBO 2.0 method is not significantly affected by HbS-, HbC-, HbEand HbD-trait haemoglobin variants [18]. These traits are rare in South African populations; >90% occur in immigrants from other countries whereas >98% of the DDS study population self-reported as Zulu or Xhosa [13,19,20]. The inter-assay coefficient of variation for HbA 1c was 0.98-2.93% for values of HbA 1c between 4.7-10.8%; all were within NGSP acceptable limits. Serum total cholesterol, low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, and total triglycerides were measured with an autoanalyser (ABBOTT ARCHITECT CI: 6200). Quality control testing was performed at the start and end of each batch or shift. WHO diagnostic criteria for diabetes were used for both glucose-based and HbA 1c -based measures [5,6]. History of diabetes and current use of insulin or oral hypoglycaemic drugs were self-reported from the questionnaire. Dyslipidaemia was defined according to the South African guidelines (based on European guidelines) [21].

Statistical analysis
Statistical analysis was performed in Stata14 software package (StataCorp: College Station, TX, USA). Continuous data are presented as mean with 95% confidence intervals (95% CI) and categorical data as a percentage (95% CI). Age-standardised diabetes prevalences were calculated with the direct method, using the WHO world standard population as the standard. To assess the distribution of risk factors by sex, a χ 2 test was used for categorical variables and a Student's t-test for normally distributed continuous variables or the equivalent non-parametric test (Mann-Whitney U test) where the normality assumption was in doubt. We fitted Poisson regression models with a single potential risk factor to obtain crude estimates of association with diabetes. We then used multi-level Poisson regression models with robust standard errors, adjusted for clustering at the household and planning unit cluster level, and for other potential risk factors and confounders, including anaemia and chronic infection, to obtain the adjusted estimates. BMI and waist circumference were included separately in the fully adjusted models as they are highly correlated and likely to be collinear [22]. Risk ratios (RRs) with 95% CI and p values are presented.
Analysis of the sensitivity and specificity of the diabetes definitions was restricted to participants with no history of previous diabetes diagnosis. This is necessary as participants with previous diabetes diagnosis were likely to be on treatment, which is likely to affect HbA 1c and glucose measurements. Receiver operating characteristic (ROC) analysis was used to assess the overall diagnostic accuracy of HbA 1c , using OGTT or FPG as the reference. Youden's Index (Sensitivity + Specificity -1) was used to determine the optimal HbA 1c cut-off for the detection of diabetes using OGTT or FPG as the reference [23].

Results
Of 1300 individuals invited to join the study, 1204 participated (response rate 92.6%); this analysis includes 1190 subjects (851 women) on whom complete data were available. Table 1 shows Table 1. Characteristics of the total study population by sex (n = 1190).

Characteristic
Total (  the characteristics of the total study group by sex. The mean age was 39.7 years (95% CI 38.8-40.7). Mean BMI, waist circumference, FPG, 2-hour plasma glucose, HbA 1c , total cholesterol, LDL and prevalence of HIV and low physical activity were higher in women. Mean systolic blood pressure and prevalence of smoking and alcohol use were higher in men. The age-standardised prevalence of diabetes was 12.9% (95% CI 11.0-14.9) based on OGTT, 11.9% (95% CI 10.2-13.9) based on FPG-alone, and 13.1% (95% CI 11.2-15.2) based on HbA 1c (Table 2). Based on OGTT, the prevalence of impaired glucose tolerance was 3.5% Table 2. Age-specific and age-standardised prevalence of diabetes based on oral glucose tolerance test (OGTT), fasting plasma glucose (FPG) and HbA 1c (n = 1190).    (Table 2). In total, 164 participants had diabetes by any definition, of which 31.1% were previously undiagnosed. For all diabetes definitions, when compared with the non-diabetes group, participants with diabetes were older and had a higher prevalence of total body and abdominal obesity, hypertension, dyslipidaemia, and family history of diabetes (Table 3). After adjustment for clustering and other potential risk factors and confounders, in the fully adjusted models; older age, higher waist circumference, higher BMI, and family history of diabetes were independently associated with diabetes, for OGTT, FPG, and HbA 1c . (Table 4). In the fully adjusted models; sex, blood pressure, haemoglobin, HIV status, lipids, smoking status, alcohol use and physical activity were not significantly associated with diabetes, for OGTT, FPG, or HbA 1c (Tables A-C in S1 Table).

Category of Glycaemia
Analysis of the sensitivity and specificity of the diabetes definitions was restricted to participants with no history of previous diabetes diagnosis (n = 1077); taking into account multiple testing, there were no important differences in prevalence and mean values of risk factors between this sample and the total study population (S2 Table) Table). Partially Adjusted = multivariable poisson regression adjusted for age, sex and clustering at the planning unit cluster and household level. c Fully adjusted = multivariable poisson regression adjusted for age, sex, waist circumference or BMI, blood pressure, haemoglobin, lipids, HIV infection, family history of diabetes, smoking status, alcohol use, physical activity and clustering at the planning unit cluster and household level.

Discussion
In an urban black South African population, we found a high prevalence of diabetes, and show evidence for the utility of HbA 1c for the diagnosis and detection of diabetes. Based on OGTT, FPG, and HbA 1c , diabetes was independently associated with established risk factors, including age, family history of diabetes and obesity. Our findings highlight the need to evaluate the potential role for HbA 1c in screening and diagnosis of diabetes in health services in the region.
The 12.9% prevalence of diabetes in the DDS is amongst the highest reported in SSA at more than triple the current International Diabetes Federation (IDF) prevalence estimate for SSA (3.8%), but is similar to that recently reported in urban black Africans in Cape Town (13.1%) [1,24]. The prevalence of diabetes in the DDS is more than double that found in a previous study in Durban in 1984 (5.3%), and greater than the increase reported in the Cape Town study [25]. Our study highlights the dramatic increase in the prevalence of diabetes in the past 30 years, and confirms that the diabetes epidemic is well established in urban South African populations. This high prevalence of diabetes is likely to be a result of the increasing burden of established risk factors, including obesity and family history of diabetes, in this population. The prevalence of diabetes was markedly higher in women than in men. This is consistent with previous reports of diabetes prevalence in South Africa; however, in some SSA countries men are consistently found to have a higher prevalence of diabetes [1,24]. Women in the DDS population had significantly higher levels of risk factors for diabetes, including low physical activity and measures of obesity, compared to men, which is, in part, likely explain this disparity.
To our knowledge, the DDS is the first population-based study in a black sub-Saharan African population to assess the utility of HbA 1c for the diagnosis and detection of diabetes. Our findings are broadly consistent with a recent pooled analysis of 96 population-based studies which found HbA 1c 48 mmol/mol (6.5%) to have consistently high specificity and low-tomoderate sensitivity to detect diabetes (using FPG or OGTT as the reference) in populations across 38 countries [8]. However, the sensitivity of HbA 1c for detection of diabetes was markedly higher in the DDS population than the pooled analysis [8]. This may be due to methodological differences in HbA 1c testing between studies in the pooled analysis or might be a result of true physiological differences in red blood cell turnover and glucose regulation between populations, affecting the relationship between HbA 1c and glucose measures in different populations. The optimum HbA 1c cut-off for the detection of diabetes in the DDS (42 mmol/mol [6.0%]) was consistent with those suggested in other populations, including prospective studies, most of which reported values lower than 48 mmol/mol (6.5%) [8,9]. This suggests that lowering the HbA 1c threshold would increase sensitivity whilst maintaining high specificity.
Some studies have found sizable differences in diabetes prevalence estimates based on HbA 1c and glucose-based definitions in populations of African origin outside of SSA [9,10]. This has led to concerns about the use of HbA 1c for population-level surveillance in these populations due to a lack of comparability with glucose-based prevalence estimates [8,11]. One study, comparing six populations of different ethnicities, included a black Kenyan population and found the prevalence of diabetes using HbA 1c to be less than half that found using OGTT [26]. However, this study used a small (n = 296) selected sample and HbA 1c was measured using a point of care test and not the standardised laboratory HPLC method recommended for HbA 1c -based diagnosis of diabetes [6]. By contrast, we found the prevalence of diabetes to be similar using OGTT, FPG, and HbA 1c , indicating that prevalence estimates using laboratory based HbA 1c are comparable to those of glucose-based measures in this population. This is important for the reliable comparison of diabetes burden and distribution in population-based health surveys and for disease surveillance using multiple measures of glycaemia.
The strengths of our study include the population-based sample with OGTT and HbA 1c performed in all individuals, as well as extensive assessment of potential confounding factors. Furthermore, laboratory measures were performed uniformly, including the use of validated NGSP and IFCC certified, laboratory-based HPLC assay for HbA 1c measurement. Limitations of this study include the use of glucose-based measures as a reference by which to assess the HbA 1c definition; glucose-based measures have considerable intra-individual variability that may lead to random misclassification [9,27]. No single measure of glycaemia captures the phenotypic complexity of diabetes and the risk of its microvascular and macrovascular complications. As such, diagnosis of diabetes in clinical practice is a sequential analytical process including repeated measurement of one or many measures of glycaemia, depending on each patient's characteristics. The low proportion of men in the DDS is consistently observed in population-based studies in South Africa and may limit the generalisability of the findings [24,28]. Likely explanations include high levels of unemployment in the townships sampled leading to men moving away for work (migrant labour system) [13,29].
HbA 1c may have several advantages for use in SSA populations. Unlike FPG and OGTT, HbA 1c does not require fasting overnight or immediate laboratory handling and samples can be easily stored and transported [11,12]. HbA 1c also appears to be more strongly associated with risk for macrovascular complications and may have potential utility for combined cardiovascular and diabetes risk assessment [30]. However, this is not established in SSA populations. There is a critical need for prospective studies to assess the relationship between HbA 1c and glucose-based measures and risk of diabetes and diabetes complications in SSA populations.
HbA 1c is also affected by conditions including haemoglobin variants, anaemia and chronic infection (including HIV and malaria), which may distort HbA 1c measurements and estimates of prevalence [11]. These conditions can be broadly divided into those that interfere with HbA 1c measurement, such as haemoglobin variants which affect the accuracy of the measurements, and those that affect the interpretation of the HbA 1c results, such as anaemia and chronic infection. In South African populations, the prevalence of haemoglobin variants and malaria is low and usually restricted to high risk populations, such as immigrants from other countries, for haemoglobin variants, and in populations near the northern border of the country, for malaria [19,20,31]. Other studies have shown that anaemia and HIV can falsely raise or lower the HbA 1c measurement [32,33]. However, in the DDS study population, HIV and anaemia were not independently associated (or inversely associated) with diabetes based on OGTT, FPG or HBA 1c definitions. Further studies, including prospective studies and studies in populations with a higher prevalence of haemoglobin variants and malaria, are needed to assess the effects of anaemia, erythrocyte abnormalities, and chronic infection on HbA 1c measurement and the utility of HbA 1c for the diagnosis of diabetes. Furthermore, evaluation of whether the potential advantages of HbA 1c result in earlier diagnosis and improvement in outcomes are needed and, in the context of extremely limited access to NGSP and IFCC certified HbA 1c laboratory testing across much of SSA [34,35], the feasibility and cost-effectiveness of large-scale implementation requires investigation in SSA.
Supporting Information S1 Table. A. Risk factors associated with diagnosis of diabetes by oral glucose tolerance test (OGTT) (n = 1190). B. Risk factors associated with diagnosis of diabetes by fasting plasma glucose (FPG) (n = 1190). C. Risk factors associated with diagnosis of diabetes by HbA 1c (n = 1190). (DOCX) S2 Table. Prevalence and mean estimates of risk factors in participants included in ROC analysis (n = 1077) compared to total study population (n = 1190). (DOCX) S3 Table. Sensitivity and specificity of HbA 1c cutoffs for detection of diabetes using oral glucose tolerance test (OGTT) and fasting plasma glucose (FPG) as the reference in participants with no history of previous diabetes diagnosis (n = 1077). Writingoriginal draft: TRH.