The utilization of HbA1c test in the screening and diagnosis of type 2 diabetes mellitus: An outpatient clinics survey in Botswana

1 Department of Internal Medicine, University of Botswana, Private Bag UB 00713, Gaborone, Botswana. 2 Department of Public Health and Family Medicine, University of Botswana, Private Bag UB 00713, Gaborone, Botswana. 3 Department of Psychology, University of Botswana, Private Bag UB 00713, Gaborone, Botswana. 4 Department of Ophthalmology, University of Botswana, Private Bag UB 00713, Gaborone, Botswana.


INTRODUCTION
The growing incidence of type 2 diabetes mellitus (T2D) is a public health concern in sub-Sahara Africa (SSA) number is projected to double by 2035 (Guariguata et al., 2014).Owing to the increasing prevalence of overweight individuals, obesity and sedentary lifestyles (Ziraba et al., 2009), T2D accounts for over 90% of cases of diabetes globally (Chen et al., 2011).Botswana has a population of about 2.2million people and has been the fastest growing economy in SSA since 2000 (Botswana Report, 2016).The country is currently experiencing high rates of overweight, obesity and other non-communicable diseases risk factors (MOH, 2007).The national prevalence of diabetes was estimated to between 3.6 and 8.25% in 2012 (Unwin et al., 2010; Diabetes Association of Botswana (DAB), 2012).It is estimated that about 50 to 70% of people with diabetes may be unaware or undiagnosed in Africa due to under-resourced healthcare systems resulting in late diagnosis and poor outcomes (Beagley et al., 2014).An effective diagnostic test is therefore important to facilitate early diagnosis and management in order to avoid costly complications.
The current diagnostic criteria for T2D are based on fasting blood glucose (FBG) and a 75 g load of oral glucose tolerance test (OGTT) (Consultation, 1999;WHO, 2006).However, both methods have considerable intra-individual variability, require an overnight fasting of at least 10 h which is inconveniencing to patients, and require a strict quality assurance measures (Libman et al., 2008;Hyltoft Petersen et al., 2001).Although, OGTT is the recognized gold standard diagnostic test for asymptomatic diabetes, FBG is mostly used in clinical practice to screen/diagnose diabetes since it is more accessible.In 2009, an expert committee of American Diabetes Association (ADA) recommended the use of glycated haemoglobin (HbA1c) as an additional test to diagnose T2D (Committee, 2009).T2D is diagnosed when HbA1c threshold of ≥6.5% (48 mmol/mol) is measured and the level between 5.7 and 6.4% (39 to 46 mmol/mol) is considered impaired glycaemia, for individuals at high risk of progression to diabetes (Olson et al., 2010).HbA1c is formed by non-enzymatic glycosylation of erythrocytes, reflecting an average glucose concentration over the lifespan of the erythrocytes (Bunn et al., 1978).
HbA1c testing is highly standardized, requiring no fasting or preparation and has less intra-individual variability (Weykamp et al., 2013).Even though, the HbA1c test was adopted by World Health Organisation (WHO) in 2011 (WHO, 2011), its utility in the screening and diagnosis of T2D remains controversial in many parts of the world.The current diagnostic threshold of ≥6.5% was based on a similar relationship with prevalent retinopathy as that of both fasting plasma glucose and 2h OGTT (Committee, 2009;Colagiuri et al., 2011).However, apart from several conditions that are known to preclude an accurate measurement of HbA1c, age and ethnic variability have been shown to affect diagnostic cut-offs values (Sharma and Signal, 2015).Hence, the recommendation by WHO that long-term prospective validation studies be done in various ethnic groups/ countries to establish the precise glucose and HbA1c levels that are predictive of microvascular complications (WHO, 2011).No such study has yet been done in SSA, raising concerns about its applicability for screening and diagnosis of T2D in this population.
The main aim of this study was to assess the HbA1c performance against a single fasting blood glucose (FBG) for diagnosis of undiagnosed diabetes and impaired fasting glycaemia among general medical outpatients in Botswana; specifically, to assess the correlation between the HbA1c and FBG among Batswana adults of African descent with no previous diagnosis of diabetes and determine the sensitivity, specificity and predictive values of HbA1c against undiagnosed diabetes and impaired fasting glucose.

Study setting and design
Data was extracted from a cross-sectional study originally designed for evaluation of a Finnish diabetes risk Score (FINRISC) between August and October 2014 at two general medical outpatients' clinics in Gaborone and Maun (Omech et al., 2016) Gaborone and Maun are two cities located in the south and north of Botswana, respectively.The study was conducted in compliance with the ethical standards of relevant national and institutional committees on human experimentation and with the Helsinki declaration of 1975.All participants signed written informed consent after all the procedures were clearly explained to them.Permission was obtained from Botswana Ministry of Health and relevant institutional review board of both hospitals.

Participants and procedures
Recruitment and evaluation was done by research nurses with prior training on the study protocol, including anthropometric measurements and good clinical practice certification.The study was conducted simultaneously at both clinics.After obtaining informed consent, all the participants aged ≥20years were selected through a systematic random sampling from each of the two clinics.A sampling frame was made from a list of patients on scheduled visit each study day.On average, 30 patients were scheduled in each clinic.The first patient was picked randomly and evaluated for study eligibility.If eligible, then every sixth patient on the list was subsequently evaluated for study eligibility and enrollment.About five participants were enrolled per day until the sample sized was attained.Patients with conditions that could interfere with HbA1c diagnosis test such as acute illness less than two weeks prior to clinic attendance, established diagnoses of diabetes, anaemia, pregnancy and chronic kidney disease were excluded.Data collected from enrolled participants, included demographics, relevant history, anthropometric measurements and blood pressure (BP) measurements.
BP was measured by the research nurse after the subject had rested for at least 5 min by using the standard mercury sphygmomanometer.With the participant sitting in an upright position, BP was recorded to the nearest 2 mmHg.For each patient, two readings were recorded within a 2-min interval and averaged out.Weight was measured to the nearest 0.1 kg, and standing height was measured to the nearest 0.1 cm using a stadiometer attached to the same medical balance weighing scale (HiCare International, Kerala, India).Body mass index was calculated as weight per square meters (kg/m 2 ).For each patient, venous blood was taken for HbA1c test.Blood samples were transported daily in an ice pack box for processing in one central accredited laboratory NGSP-certified using a high-performance liquid chromatography (HPLC) assay method (Abbot Architect, 2007, Germany).This method is aligned with the Diabetes Control and Complications Trial (DCCT) as recommended by WHO (Diabetes Control and Complications Trial Research Group, 1993).Plasma glucose was analyzed by the research nurse at point of care using oxidase method (Betachek glucometer; National Diagnostic Products Pty Ltd, Sydney, NSW, Australia).The HbA1c results were categorized according to American Diabetes Association (ADA) criteria into normal glycaemia (<5.6%), impaired glycaemia; 5.7 to 6.4% (39 to 46 mmol/mol) and diabetes; ≥6.5% (48 mmol/mol) (Olson et al., 2010).Plasma glucose was categorized according to WHO criteria, into normal glycaemia, <6.1mmol/l, impaired fasting glycaemia, 6.1-6.9mmol/l and diabetes, ≥7.0mmol/l (WHO, 2006)

Statistical analysis
Demographic data, anthropometric and laboratory measurements for individual participants were entered into a data base using IBM SPSS statistics version 24.Participants were grouped into three categories based on HbA1c status.Proportions and means (SD) were computed for categorical variables and continuous variables, respectively.Differences between the groups were ascertained using cross-tabulation.With MedCalc® software (BVBA, 2016), using HbA1c as screening test and FBG as the reference test, the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were computed.Pearson's correlations coefficients with Scatter diagrams between HbA1c and FBG in the overall population and for each gender were computed and results tabulated.A linear regression line with 95% confidence interval and r 2 were provided.Level of statistical significance was taken to be p < 0.05.

Participant's characteristics
A total of 704 participants were screened; 304 (43.2%) were not eligible because of various reasons such as being acutely ill, already known diabetic or anaemic.Of the 400 patients enrolled, 109 (15.5%) did not turn-up for fasting blood test.Data from the remaining 291 (41.3%) were analyzed (Figure 1).This comprised of 216 (74.2%) women, 75 (25.8%)men and all were Batswana of black African descent residing in the urban centres of Gaborone and Maun.Their mean age was 50.1±11 years; 50.5±11.8years for male and 49.9 ±11.0 years for female.
The majority (71%) of the participants had either completed secondary education level or primary education level, with only <7% having no formal schooling.Regarding employment status, 52% were formally employed and 23% were retired, while only 7% considered themselves unemployed.The most common reason for the visit was regular appointment for either one or more chronic disorders 197 (67.7%), followed by new referrals (26.5%).A majority of the participants: 215 (73.9%) had no symptom at presentation, and the common symptoms were headache, body pains, and shortness of breath.The most common chronic conditions were systemic hypertension; 135 (46.4%) and heart diseases, 37 (12.7%).However, 88 participants (26.9%) had no chronic disease.A total of 114 (39.2%) participants were HIV seropositive, 156 (53.6%) were HIV seronegative and 21 (7.2%) had unknown HIV status.Of those who were seropositive, 87 (76.3%) were on firstline antiretroviral therapy (ART; nucleoside/nucleotide analogues and non-nucleoside combinations).Only nine (7.9%) were on second-line ART (protease-based ART) and 18 (15.7%)were not on ART.
Using ADA diagnostic criteria, the overall population was stratified based on HbA1c diagnoses (Table 1); diabetes (n=42, 14.4%), impaired glycaemia (n=159, 54.3%) and normal glycaemia (n=90, 30.9%).The mean body mass index (BMI) and FBG were significantly higher in those diagnosed with diabetes as compared to those with impaired or normal glycaemia.Although, there were a trend towards higher mean age and elevated cholesterols in the diabetes groups as compared to the other groups, they were not statistically significant.
The outcomes of FBG testing in the various HbA1c glycaemic categories for the overall population were analyzed (Figure 1).Of those classified as diabetes, 37 (88.1%) had normal FBG, 3 (7.1%)had impaired fasting glycaemia and only two (4.8%) had diabetes.In the Impaired HbA1c glycaemia group, 158 (99.4%) had normal fasting glycaemia and only one participant was confirmed to have impaired fasting glucose.All the participants with normal HbA1c status had normal FBG.

Diagnostic accuracy of HbA1c
Table 2 summarizes the sensitivity, specificity, positive and negative predictive values of HbA1c as screening test as compared to FBG.HbA1c had a sensitivity and specificity of 100 and 86.2% for screening of T2D, respectively.Similarly, it had a sensitivity and specificity of 100 and 36.3% for screening of IFG, respectively.The positive predictive value (PPV) for T2D and IFG was 4.8 and 0.6%, respectively.The negative predictive value (NPV) for T2D and IFG was 100% in both cases.

Correlation between HbA1c and FBG
Figure 2 depicts the scatter plots diagram between HbA1c and FBG in the overall population.The Pearson's correlation coefficient, r was 0.546, p=0.001 for the overall population, the fitted regression line showed a slope of 0.43 and r 2 of 0.287.The coefficient r for women was 0.578, p= 0.001 and that for men was 0.336, p=0.003 and both were significant at p< 0.05 (Figure 3A    and B).

DISCUSSION
The sensitivity of HbA1c for diagnosis of undiagnosed diabetes and impaired fasting glycaemia were high but its specificity varied widely.HbA1c specificity for impaired fasting glycaemia was very low (36.3%) as compared to undiagnosed diabetes (86.2%).These findings suggest a high proportion of discordancy between the HbA1c and fasting blood glucose based criterion for diagnosis of prevalent undiagnosed diabetes in this population.In addition, the current American Diabetes Associated recommended HbA1c cut-off values poorly predicted both prevalent undiagnosed diabetes and impaired fasting glycaemic status, but performed excellently in predicting normal glycaemic status (Table 2) and hence it can be considered an effective exclusion tool for a new case of undiagnosed diabetes during screening program.The accuracy and predictive values of HbA1c is highly dependent on assays precisions, population's reference values, disease prevalence and the gold standard test.In this study, the HbA1c test was standardized to the currently recommended assay methods certified by NGSP and samples analysis was pooled in one central laboratory which is deemed to have minimized inconsistent findings.The HbA1c optimum cut-off values are known to vary in different races/ethnicities and in between studies which affect its sensitivity and specificity (WHO, 2011).For example, two previous studies in the U.S which included both black and white population showed that blacks had significantly higher mean HbA1c values (5.8%) than whites (5.4%) (p<0.001)(Selvin et al., 2010;Herman et al., 2007).In the SSA context, a cross sectional study in KwaZulu-Natal estimated HbA1c optimal cut-off was 6.0%, with a sensitivity of 89.2% [95% CI :78.6-99.8]and a specificity of 92% [95% CI:90.3-93.7](Hird et al., 2016).Previously, a similar study in Cape Town found a comparable HbA1c optimal value of 6.1% with a sensitivity and a specificity of 80 and 77%, respectively (Zemlin et al., 2011).These findings from South Africa support the need for population derived HbA1c references values in the African context.
A modest, but significant positive correlation was also observed between HbA1c and fasting blood glucose levels (Figure 2).However, the correlation was weaker in men as compared to women possibly due to smaller sample size of men in this study (Figure 3A and B).HbA1c is recognized as a reliable indicator of chronic glycaemic status, as compared to glucose based markers.In this study, the correlation were similar to other studies among non-diabetic Turkish outpatients study (r=0.47,p=0.001) (Giniş et al., 2012), as well as among the nondiabetic Dutch community study (r=0.46,p=<0.001)(van 't Riet et al., 2010).Although, patients with known diabetes were not included in the sample, the correlation between HbA1c and FBG were stronger among known diabetics in the Dutch study (r=0.71),which is consistent with wellknown positive association between HbA1c with glucose levels in established diabetes mellitus (Koenig et al., 1976).HbA1c and FBG discordance in this study may be attributed to intra-individual variability especially in the dysglycaemic phase as well as analytical and preanalytical factors involved in the measurements of FBG (Chen et al., 2011;Sacks, 2011) Despite the current limitations of HbA1c especially in SSA populations, HbA1c is a relatively convenient test, requiring no fasting and a single blood sample, though more costly.Based on the results, HbA1c and fasting blood glucose may still play a dual role in the diagnosis of undiagnosed diabetes in the clinic settings.The high proportions of discordancy, an elevated HbA1c test requires a confirmatory blood glucose test to define true diabetes cases for clinical interventions.This is consistent with most guidelines for screening and diagnosis of diabetes among high risk groups (Committee, 2009;d'Emden et al., 2012).It is an important consideration when planning improved uptake of testing and detection of diabetes, where majority of cases are undiagnosed especially in low and middle income countries (Beagley et al., 2014).
The results of this study should be considered in light of several limitations associated with HbA1c as a screening and diagnostic test that has been discussed in various literature (Sharma and Singal, 2015).For instance, HIVsero positive individuals are considered to have ongoing low chronic haemolysis from viral infections or antiretroviral drugs, affecting the reliability of HbA1c test resulting in under estimation of case diagnosis (Polygreen et al, 2003).Indeed, in this setting, HIV infection was highly prevalent (39.2%) and over 80% were on nonnucleoside reverse transcriptase inhibitors-, particularly efavirenz, which may have affected the HbA1c A B

Figure 1 .
Figure 1.The flow diagram of participants' recruitment and outcomes of HbA1c tests, followed by FBG test under each glycaemic categories.

Figure 2 .
Figure 2. The scatter plots diagram depicting the linear correlation between HbA1c and FBG for the overall population.

Figure 3 .
Figure 3. Scatter plot diagrams depicting the linear correlations between HbA1c and FBG in female (A) and male (B).

Table 1 .
General demographic and clinical characteristic of study participants stratified by HbA1c status (N=291).
Values provided as means ±standard deviations; except for gender which are proportions; † p=<0.05.

Table 2 .
The sensitivity, specificity, positive and negative predictive values of HbA1c as a screening test against FBG.