HbA1c as a screening tool for detection of Type 2 diabetes: a systematic review

Authors

Errata

This article is corrected by:

  1. Errata: Corrigendum Volume 24, Issue 9, 1054, Article first published online: 21 August 2007

: Dr C. M. Bennett, Senior Lecturer, Department of Public Health, The University of Melbourne, Level 2, 723 Swanston Street, Carlton, Victoria 3052, Australia. E-mail: c.bennett@unimelb.edu.au

Abstract

Aim  To assess the validity of glycated haemoglobin A1c (HbA1c) as a screening tool for early detection of Type 2 diabetes.

Methods  Systematic review of primary cross-sectional studies of the accuracy of HbA1c for the detection of Type 2 diabetes using the oral glucose tolerance test as the reference standard and fasting plasma glucose as a comparison.

Results  Nine studies met the inclusion criteria. At certain cut-off points, HbA1c has slightly lower sensitivity than fasting plasma glucose (FPG) in detecting diabetes, but slightly higher specificity. For HbA1c at a Diabetes Control and Complications Trial and UK Prospective Diabetes Study comparable cut-off point of ≥ 6.1%, the sensitivity ranged from 78 to 81% and specificity 79 to 84%. For FPG at a cut-off point of ≥ 6.1 mmol/l, the sensitivity ranged from 48 to 64% and specificity from 94 to 98%. Both HbA1c and FPG have low sensitivity for the detection of impaired glucose tolerance (around 50%).

Conclusions  HbA1c and FPG are equally effective screening tools for the detection of Type 2 diabetes. The HbA1c cut-off point of > 6.1% was the recommended optimum cut-off point for HbA1c in most reviewed studies; however, there is an argument for population-specific cut-off points as optimum cut-offs vary by ethnic group, age, gender and population prevalence of diabetes. Previous studies have demonstrated that HbA1c has less intra-individual variation and better predicts both micro- and macrovascular complications. Although the current cost of HbA1c is higher than FPG, the additional benefits in predicting costly preventable clinical complications may make this a cost-effective choice.

Abbreviations
ADA

American Diabetes Association

BMI

body mass index

CV

coefficient variation

DCCT

Diabetes Control and Complications Trial

DM

diabetes mellitus

FPG

fasting plasma glucose

FPLC

fast high-pressure liquid chromatography

Hb

haemoglobin

HbA1c

glycated haemoglobin A1c

IDF

International Diabetes Federation

IFCC

International Federation of Clinical Chemistry and Laboratory Medicine

IGT

impaired glucose tolerance

OGTT

oral glucose tolerance test

ROC

receiver operating characteristic

UKPDS

UK Prospective Diabetes Study

WHO

World Health Organization

Introduction

The high prevalence of diabetes mellitus has emerged as a worldwide public health problem in the past 20 years. Type 2 diabetes is the most common form of diabetes, estimated to account for 85–90% of diabetes [1]. Impaired glucose tolerance (IGT) is a predictor for the subsequent development of Type 2 diabetes and is itself a risk factor for macrovascular disease. Type 2 diabetes is often asymptomatic in its early stages and can remain undetected for several years [2]. Increasing evidence shows that half of those with Type 2 diabetes are not aware that they have the condition. Early diagnosis of the condition is important as careful diabetes management can reduce long-term complications, such as blindness, kidney failure, cardiovascular disease and limb amputation [3–5].

There is no consensus on the most accurate screening test for detection of diabetes. The most widely used screening tests include the fasting plasma glucose (FPG) test and the oral glucose tolerance test (OGTT). Both these tests involve measurement of blood glucose. However, the measurement of both OGTT and FPG require patients to fast overnight for at least 8 h and confirmation of diagnosis using FPG requires the test to be repeated at least twice. Furthermore, studies have shown that the sensitivity of FPG for diabetes diagnosis is not as high as expected, with nearly one-third of individuals with diabetes remaining undetected [6]. OGTT is also costly, time-consuming and labour intensive and has low reproducibility that can add confusion and uncertainty to the confirmation of diabetes diagnoses [7]. The accuracy of FPG and OGTT may be reduced by patient non-adherence to fasting, laboratory error and/or use of certain medications [6].

The glycated haemoglobin (HbA1c) test has been suggested as an alternative screening test for Type 2 diabetes. HbA1c levels represent a 2–3-month average of blood glucose concentrations. The accuracy of HbA1c analysis may be influenced by the presence of haemoglobinopathy or renal failure, as well as laboratory error and/or use of certain medications [6], but, compared with the OGTT, HbA1c measurement is quicker and more convenient. HbA1c can be measured at any time of the day regardless of the duration of fasting or the content of the previous meal. HbA1c can also be analysed with a small blood sample using a portable device, although this is an expensive option currently [8]. There is also the potential for blood obtained from a finger prick to be sent to a central laboratory for analysis, allowing screening of individuals in remote areas [9,10].

HbA1c, OGTT and FPG are equivalent as predictors of the development of retinopathy and nephropathy [11–14]. In recent years, the validity of HbA1c as a screening tool for diabetes has also been examined, using OGTT as the gold standard and FPG as the comparison. These studies have been conducted in hospital and community settings in different countries, and in different ethnic groups. However, there is as yet no consensus on a suitable cut-off point for HbA1c in the detection of diabetes. We conducted a systematic review of these primary cross-sectional studies to assess the validity of HbA1c as a screening tool and determine the most appropriate cut-off point for the diagnosis of diabetes.

Research design and methods

Study design

A systematic literature search was conducted to identify published primary research with data on the accuracy of HbA1c in the detection of Type 2 diabetes.

Search strategy

medline, pubmed and embase electron databases (1994–September 2004) were searched using the keywords ‘diabetes mellitus’, ‘screening’, ‘diagnosis’, ‘HbA1c’, ‘fasting plasma glucose test’ and ‘OGTT’. Search engines, such as ‘Google’ and ‘Medscape’, were also used to search for related articles. Reference lists of all publications including original studies, letters, commentaries, guidelines, and reviews were manually checked to identify studies not found through electronic searching.

Inclusion criteria

All peer-reviewed journal articles that related to the validation of HbA1c as a tool for detection of diabetes and were published in English were examined. Papers were analysed further if FPG was used as a comparison, and the 75-g OGTT was used as the reference test. The reference test had to have been performed on at least 80% of the participants in whom HbA1cwas measured. The sensitivity and specificity of the HbA1c and FPG test must either have been reported, or were possible to compute from the data provided. The diagnosis of diabetes must have been based on the diagnostic criteria according to American Diabetes Association (ADA; 1997) or World Health Organization (WHO; 1999) guidelines. All studies had to report Diabetes Control and Complications Trial (DCCT)-aligned HbA1c results, or results that could be converted to DCCT-aligned results using published conversion regression models. If standardization information was not available from the published paper, authors or laboratories were contacted for confirmation of methods used.

Data extraction

Information on study quality, study characteristics and accuracy of HbA1c results were extracted from each selected paper. Study participant characteristics including summary data on age, gender, diabetes risk factors and ethnicity were extracted. Prevalence of Type 2 diabetes in each study paper was extracted directly, or was calculated using the summary data reported in the article.

Information was also examined in relation to study design, verification methods and accuracy of the results, including the cut-off point used, sensitivity and specificity, likelihood ratios and the use of the receiver operating characteristic (ROC) curve analysis to identify the cut-offs. If the accuracy estimates were not provided, raw data were used to calculate these estimates.

Analysis

The methodological rigour and quality of studies were assessed by examining the sampling frame, recruitment methods and sample size, the test measurements of HbA1c and FPG, and the method used for collecting blood samples. The adequacy of the test descriptions was assessed to determine if sufficient to allow potential replication, and whether there was 80% verification with the reference test. Finally, the accuracy of DCCT-aligned HbA1c and FPG tests was compared against OGTT as the standard reference test, based on sensitivity (probability of detecting diabetes or IGT among those who have diabetes/IGT), specificity (probability of a negative test among those without the condition), likelihood ratios [likelihood of having a certain test result (positive or negative) in those with the condition compared with the likelihood of the same test result in those without the condition], and ROC curve analysis. The advantage of using likelihood ratios is their stability in the face of changing disease prevalence, especially when considering the wide variation in the prevalence of diabetes observed between different ethnic groups. Likelihood ratios of more than 1 indicate increased probability of the target disorder [15].

Results

Search results

The initial search strategy identified 63 studies, nine of which were retained in the systematic review after applying the inclusion criteria (Fig. 1). Of the eligible studies, all were published between 1998 and 2004, and six were conducted in European countries and three in Asian countries (see Table 1). All nine studies differed in the population targeted, some were general population screening, others targeted those at increased risk of diabetes for screening. Recruitment methods also differed, with some community based [16–19], and others hospital or health institution based [20–24].

Figure 1.

Studies selected for systematic review.

Table 1.  Studies examining the usefulness of HbA1c as a screening tool for the early detection of Type 2 diabetes
First author, year and countrySettingsRecruitment methodsHbA1c test methodGlucose methodBlood sampleAdequate test descriptionVerification with OGTT
  1. BMI, body mass index (kg/m2); CV, coefficient of variation; DCA 2000: a portable device produced by Bayer (Bayer Diagnostics, Mulgrave, Victoria, Australia) for testing HbA1c by immunological method; DM, diabetes mellitus; FPLC, fast high-pressure liquid chromatography, a kind of ion-exchange chromatography; IGT, impaired glucose tolerance; NGSP: National Glycohemoglobin Standardization Program, sponsored by the American Diabetes Association and others; NHANES III, the Third National Health and Nutrition Examination Survey.

Community-based studies
Colagiuri 2004, AustraliaForty-two randomly selected urban and non-urban areas (census collector districts)≥ 25 years of age, household survey followed by biomedical examination; 5604 had ≥ 1 risk factor in accordance with Australian screening guidelinesBoronate affinity HPLC, normal range: (3.9–6.2%) CV: not statedGlucose oxidaseVenous plasmaPoor100%
Mannucci 2003, ItalyA suburban community on the outskirts of Florence30–70-year-old self-referred general populationHPLC (HA8121, Menarini Diagnostics; upper limit 5.5%) CV: not statedGlucose oxidase (5.5–6.1)Venous plasmaAdequate100%
Saydan 2002, USANational Center for Health (NHANES III participants)Age 40–74, BMI ≥ 24 kg/m2 with IGTIon-exchange HPLC (Bio-Rad Laboratories; CV 2%; normal range: 4.3–6.1%)HexokinaseVenous plasmaPoor100%
Wiener 1998, UKTwo cities in the UKGP-referred high risk of DM 13–92 yearsIon-exchange HPLC [Daiichi HA-8121(NMGH) or HA-8140(RLH), CV2%] Normal range: (3.8–5.5%)HexokinaseCapillary plasmaAdequate100%
Hospital-based studies
Jesudason 2003, Adelaide, AustraliaEndocrine Test Unit at the Royal Adelaide HospitalVolunteers with high risks of DM (obesity, family history of diabetes and gestational diabetes, symptoms of polyuria and polydipsia)HPLC and DCA 2000 (4.3–6.1%) (4.2–6.3%)HexokinaseVenous plasmaGood100%
Herdzik 2002, PolandOutpatients clinic or the Department of Internal Medicine at Pomeranian Academy of Medicine≥ 18 years of age, suspicion of having diabetes as a result of symptoms or having known risk factors for IGTFPLC (Pharmacia FPLC System) CV (0.47–0.94%) Normal range: 3.8–5.2%OxidaseCapillary whole bloodGood100%
Tanaka 2001, JapanSaiseikai Central Hospital and Juntendo University Hospital in TokyoSuspected of having DM; excluded anaemia (Hb < 12 g/dl) and renal or hepatic dysfunctionHPLC (723Ghb III, Tosoh Corp.) Normal range: 4.3–5.8%Not statedVenous plasmaPoor100%
Tavintharan 2000, SingaporeVolunteers attending the hospital (large proportions were nurses)< 55 years of age, family history, hypertension, hyperlipidaemia, with BMI > 25 kg/m2; excluded renal, liver, or cardiac disease and previous gastrectomyDCA 2000 (CV 3.3%) Normal range: 4.2–6.5%Technicon RA Systems analyser Interassay CV 2.0%Venous plasmaAdequate100%
Ko 1998, Hong KongDiabetes and Endocrine Center of Wales Hospital for diabetes screeningHigh risk for IGT (family history of diabetes, gestational diabetes, history of IGT and obesity)Automated ion-exchange chromatographic method (Bio-Rad Laboratories; CV = 3.1%) normal range: 5.1–6.4%OxidaseVenous plasmaAdequate100%

Methodological quality of studies and characteristics of participants

Methodological qualities and participants’ characteristics for each study are described in Tables 1 and 2. Two community-based studies included random samples from the general population [16,17], whilst the others were based on sampling strategies that enriched the sample with participants with one or more risk factors for Type 2 diabetes [18,19]. All hospital-based studies included subjects at high risk for Type 2 diabetes.

Table 2.  Community-based studies—participant characteristics, presence of baseline data, the prevalence of Type 2 diabetes and results of HbA1c and FPG tests
CriterionColagiuri et al.Mannucci et al.*Saydan et al.Wiener and Roberts*
  • ADA, American Diabetes Association; DM, diabetes mellitus; F, female; FPG, fasting plasma glucose; IFG, impaired glucose tolerance; M, male; NLR, negative likelihood ratio; OGTT, oral glucose tolerance test; PLR, positive likelihood ratio; ROC, receiver operating characteristic; Sn, sensitivity; Sp, specificity; WHO, World Health Organization.

  • *

    HbA1c cut-off points converted to DCCT-aligned equivalent values, see Research design and methods.

Age (year)> 25 years summary statistics not reported52.2 ± 18.5 (30–70)40–7413–92
No. and gender (M/F)10 44712152844401
 Not stated567/648Not stated208/193
EthnicityAustralia national populationItalianUSAUK
Baseline dataNoYesNoNo
Prevalence of diabetes (%)7.46.66.2 (ADA criteria)44
Results of tests
HbA1c
Cut-off point (%)≥ 5.3 > 6.6 ≥ 5.5≥ 6.0> 6.9> 7.4> 7.6
ItemsDMIGTDMIGTIGT DM  
Sn (%)78.742.0M 98M 5960.016.764.050.641.0
   F 100F 55     
Sp (%)82.888.2M 30M 1955.092.991.098.2100
   F 21F 9     
Likelihood ratio
PLR4.583.56M 1.30 M 0.731.332.357.1128.11
   F 1.28F 0.60     
NLR0.260.66M 0.31 M 2.120.730.900.400.500.59
   F 0F  4.86     
FPG
Cut-off point (mmol/l)≥ 5.5≥ 6.1≥ 7.0≥ 6.1≥ 5.6≥ 6.1> 6.0> 6.9 
ItemsDM DMIGTIGT DM  
 IGT        
Sn (%)79.963.6M 91M 5976.534.989.978.1 
 51.934.6F 100F 55     
Sp (%)79.993.9M 30.0M 19.337.986.965.987.9 
   F 21.6F 9.3     
Likelihood ratio
PLR3.9810.41.300.731.232.662.646.45 
 3.91.280.60     
NLR0.250.390.312.12     
 0.550.6504.86     
ROC curve availableNo Yes (for IGT)Yes Yes  
Diagnose criteriaWHO (1999)WHO (1999)WHO (1999)WHO 2-h OGTT

All the studies used OGTT as the referenced standard test on all participants; however, there was variation between studies in the testing methods for HbA1c and FPG. The majority used venous plasma, while two studies used capillary whole blood or capillary plasma samples [19,21]. In both studies, the OGTT cut-off values were adjusted for the use of capillary rather than venous blood.

For testing HbA1c, most studies used DCCT-aligned methods. Saydan et al. [18], and Jesudason et al. [20] used the US national standard ion-exchange HPLC system (Bio-Rad Laboratories, Hercules, CA, USA). Jesudason et al. [20] also used a standard immunological method tested by a portable device DCA 2000 (Bayer Diagnostics, Mulgrave, Victoria, Australia). Tanaka et al. [22] used DCCT aligned-HPLC(723Ghb) by the Tosoh Corporation (Tokyo, Japan). Two other studies also used the HPLC method, confirmed to be DCCT standardized [16,24]. Tavintharan et al. used the DCCT-aligned DCA 2000 immunological method [2].

Mannucci et al. [17] and Wiener and Roberts [19] tested HbA1c by the HPLC method using the HA8121 or HA-8140 machine [17,19] and their cut-off points were adjusted in this review using the Menarini Diagnostics (Florence, Italy) re-calibration formula [(measured value) = 1.06 (DCCT-aligned value)−1.51][25]. Herdzik et al. [21] used the Pharmacia (Uppsala, Sweden) fast high-pressure liquid chromatography (FPLC) method, and this was converted to DCCT-aligned results using the Federation of Clinical Chemistry and Laboratory Medicine (IFCC) Master Equations [27]. For testing FPG, two studies did not report the specific methods used [22,23], whilst the remaining studies used either hexokinase or glucose oxidase enzymatic methods for the FPG tests.

Results by participant characteristics

The age range of study participants varied across studies. Wiener and Roberts [19] included a sample with an age range from 13 to 92 years, while the other studies only included adults aged 18 years or older. The gender mix also varied across studies. In the studies by Tavintharan et al. [23] and Ko et al. [24], the majority of participants were female, 65 and 80%, respectively. Only three articles provided baseline data [17,21,24]. Participant risk profile and prevalence of Type 2 diabetes varied across studies, ranging from 6.2 to 44.0%. Apart from the exceptionally high diabetes prevalence (44%) reported for the UK enrolled participants [19], the prevalence in hospital-based studies (10–21%) was higher than in community based studies (6.2–7.4%).

Results by cut-off point

Sensitivity is the fraction of individuals at or above the HbA1c cut-off point who have diabetes, whereas specificity is the fraction of individuals with an HbA1c level below the cut-off point who do not have diabetes. For calculations of sensitivity and specificity for IGT, individuals with undiagnosed diabetes were excluded. The ROC curve analysis was performed to assess the best predictive cut-off values for detecting new diabetes and IGT.

Most studies (seven of nine) used the ROC curve to identify the cut-off point for diagnosing Type 2 diabetes or IGT with FPG or HbA1c. Different studies used different cut-off points and reported different sensitivities and specificities (Tables 2 and 3). Even when the same cut-off points were used, different results were reported by different investigators.

Table 3.  Hospital-based studies—participant characteristics, presence of baseline data, the prevalence of Type 2 diabetes and results of HbA1c and FPG tests
CriterionJesudason et al. Herdzik et al.*Tanaka et al.Tavintharan et al. Ko et al.
  • *

    HbA1c cut-off points converted to DCCT-aligned equivalent values, see Research design and methods.

  • ADA, American Diabetes Association; DM, diabetes mellitus; NLR, negative likelihood ratio; FPG, fasting plasma glucose; OGTT, oral glucose tolerance test; PLR, positive likelihood ratio; ROC, receiver operating characteristic; Sn, sensitivity; Sp, specificity; WHO, World Health Organization.

Age (year)53.8 (19–88)> 18 (20–82)5643.2 (37–50)Mean 55
No. and gender5052348661112877
(M/F)211/294129/105569/29735%/65%565/2312 (19.6%/80.4%)
EthnicityAustralianPolishJapaneseChinese60%, Hong Kong Chinese
Indian20%,
Malays19%,
Others1%
Baseline dataNoYesNoNoYes
Prevalence of diabetes (%)10.71920.61721
Results of tests
HbA1c
Cut-off point (%)≥ 4.7≥ 5.6≥ 6.2≥ 5.0≥ 6.2≥ 6.8≥ 6.4≥ 6.2≥ 5.9 ≥ 6.5≥ 5.9 ≥ 6.8≥ 6.1≥ 6.2≥ 7.0≥ 6.1≥ 5.5
ItemsDMHPLC DMDCA 2000 DMIGTDM DMDCA 2000  DMDM
         IGT      IGT
Sn (%)10085.242.610072.242.673.751.3764995  5281814877.591.7
         7649     80.2
Sp (%)1080.599.111.194.799.693.295.8869867  9784889878.843.6
         7798     48.8
Likelihood ratio
PLR1.114.37106.51.1213.6106.510.8412.25.4324.52.88 17.35.066.75243.661.63
         3.3024.5     1.57
NLR00.180.5800.290.580.280.510.280.520.07 0.490.230.220.530.290.19
         0.310.52     0.41
FPG
Cut-off point (mmol/l)≥ 4.7≥ 5.6≥ 6.4   ≥ 7.0≥ 6.1≥ 7.0 ≥ 7.0≥ 6.1≥ 6.0≥ 5.8≥ 5.8≥ 5.6
ItemsDM     DMIGTDM DM   DMDM
                IGT
Sn (%)1009.659.3   77.762.652 1948525285.188.4
                64.4
Sp (%)23.185.899.1   100100100 10098989084.479.2
                87.7
Likelihood ratio
PLR1.35.6165.89    24265.25.464.25
                5.24
NLR00.240.41   2.233.744.8 8.10.530.490.530.180.15
                0.41
ROC curve availableYes     Yes No Yes   No 
Diagnose criteriaWHO 1999WHO 1999WHO 1999WHO (2-h OGTT) and ADA (1997)WHO 2-h OGTT

Three cut-off points 5.9, 6.1 and 6.2% of HbA1c were advised as optimum cut-offs for detecting diabetes in at least two different studies. At a cut-off point of 5.9%, the sensitivity ranged from 76 to 95%, specificity 67 to 86%[22,23], and at a cut-off point of 6.1%, the sensitivity ranged from 78 to 81%, specificity 79 to 84%[23,24]. At a cut-off point of 6.2%, the sensitivity ranged from 43 to 81% and specificity 88 to 99%[20,21,23]. For FPG at a cut-off point of ≥ 6.1%, two studies showed the sensitivity ranged from 48 to 64% and specificity from 94 to 98%[19,23]. At a cut-off point of 5.6, the sensitivity ranged from 80 to 88%, and specificity from 79.2 to 85.8%[20,24].

At equivalent cut-off points, sensitivity was generally lower in detecting IGT for both HbA1c and FPG tests when it came to their ability to detect IGT, and this was true in both community- and hospital-based studies. Some studies reported different results for the validity of HbA1c and FPG. Wiener and Roberts reported higher specificity of HbA1c, but higher sensitivity of FPG in diagnosing diabetes at different cut-off points [19]. Tavintharan et al. showed HbA1c had higher sensitivity but lower specificity than FPG to diagnose diabetes at different cut-off points [23]. However, at certain cut-off points (7.0% HbA1c and 6.1 mmol/l FPG), Tavintharan et al. reported the same sensitivity (48%) and specificity (98%) for HbA1c and FPG [23].

Saydah et al. found that HbA1c had a lower sensitivity, but higher specificity, for detecting diabetes among non-Hispanic White subjects than for all other ethnic groups [18]. Similar variation in test performance can be seen when comparing findings across studies based on people from Asian [22–24] and European backgrounds [19,20].

Jesudason et al. studied the use of HbA1c and FPG to predict Type 2 diabetes and cardiovascular risk, and compared results for HbA1c measured by HPLC and the portable DCA 2000 device [20]. They concluded that HbA1c (HPLC) of < 4.7% and ≥ 6.2% predicted with certainty the absence or presence of Type 2 diabetes as defined by the OGTT. The corresponding cut-offs were < 5.0 and ≥ 6.8% for HbA1c (DCA 2000). Cardiovascular risk increased at least 2.2 times at HbA1c ≥ 6.2% (by HPLC), 1.8–2.2 times at HbA1c 5.6–6.1%, 2 times at FPG ≥ 6.4 mmol/l, and 1.7–1.9 times at FPG 5.6–6.3 mmol/l. The DCA 2000 and HPLC methods for HbA1c have shown a correlation with a constant bias of approximately 0.2%. Based on the ROC curve, DCA 2000 had higher values than HPLC. An HbA1c cut-off point of 6.2% (DCA 2000) had similar accuracy as FPG ≥ 6.0 mmol/l for predicting Type 2 diabetes (sensitivity 72.7 and 74.1%; specificity 94.7 and 94.5%, respectively).

The DCA 2000 method was also used to test HbA1c in a study conducted by Tavintharan et al. [23]. Both Jesudason et al. [20] and Tavintharan et al. [23] recommended HbA1c with a cut-off level of 6.2% to be the best predictor for diabetes.

Saydah et al. [18] stratified by age and body mass index (BMI) for the sensitivity and specificity of fasting glucose and HbA1c in identifying IGT among US participants between 40 and 74 years of age. They concluded that age and BMI influence the sensitivity of both tests in detecting IGT. Both tests had relatively higher sensitivity and specificity when used to screen people aged 60 to 74 years than for people between 40 and 59 years, but the differences were not substantial. Sensitivity also increased when BMI increased from ≥ 24 to ≥ 27, or to ≥ 30 kg/m2.

Discussion

This systematic review did not find clear evidence to suggest that one test, HbA1c or FPG, was superior to the other in screening for diabetes or IGT. On the whole, HbA1c had slightly lower sensitivity but higher specificity than the FPG in detection of diabetes, but neither was effective in detecting IGT. This is not surprising as neither FPG or HbA1c involve a glucose challenge.

It is rarely possible in screening tests to have both high sensitivity and specificity. In the case of diabetes, we have a relatively common disease and so efficiency of screening, and therefore the specificity of the test used, is arguably more important. For screening purposes, a test that produces a large number of false positives would pose major problems to the health department and may be costly in resource terms. However, to fully assess these tests, it is important to consider the relative abilities of these tests to not only detect both diabetes and IGT, but also to predict the long-term prognosis.

First, it is worth considering the choice of test that is used in this review as the ‘gold standard’ test for diagnosing diabetes. In individuals with Type 2 diabetes who have insulin resistance, the insulin secretory response can initially compensate for the insulin resistance. However, eventually, first-phase insulin secretion is lost, and second-phase secretion is impaired, causing postprandial hyperglycaemia, one of the earliest markers of disease progression. Abnormalities in hepatic, pancreatic and muscle metabolism all result from long-standing hyperglycaemia. By the time most patients experience symptoms significant enough to cause them to seek medical attention, Type 2 diabetes has often been present, unrecognized for years, hence the presence of complications at diagnosis in some subjects [28].

Therefore, from the view of the pathophysiology of Type 2 diabetes, postprandial hyperglycaemia (PPG) is one of the earliest markers of disease progression. FPG is influenced by insulin secretion, carbohydrate absorption and hepatic and peripheral glucose metabolism. Elevated FPG levels result from unrestrained hepatic glucose production secondary to resistance to insulin action in the liver, and every diabetic individual has their own internal and external causal factors that trigger this disease. The difficulty is in preventing and treating long-term micro- or macrovascular complications which magnify the economic cost of diabetes.

HbA1c is related to both elevated OGTT and FPG [10], and the various complications [29]. HbA1c can therefore be used for assessing the risk of complications of diabetes as well as for monitoring glycaemic control. Both the DCCT (Type 1 diabetes) [30] and UK Prospective Diabetes Study (UKPDS; Type 2 diabetes) [31] studies have shown a reduction in the risk of complications in diabetic patients associated with a reduction in HbA1c. In the DCCT, a sustained reduction in HbA1c from 9.0 to 7.0% over 6.5 years led to a reduction in the risk of developing retinopathy of 76%, 39% risk reduction for microalbuminuria and 60% for neuropathy. The risk of developing any microvascular complication of diabetes was reduced by 25% in the UKPDS, where HbA1c was lowered from 7.9 to 7.0% over 10 years through intensive diabetes management.

Some studies in this review reported similar sensitivities and specificities for both HbA1c and FPG, whereas others demonstrated different sensitivities and specificities for the same cut-off points. There are many confounders that potentially influenced the results of the primary studies, making the comparison of HbA1c and FPG among studies difficult and complex.

First, different ethnic groups were found to have different sensitivity and specificity for HbA1c which may be related to genetic differences in the concentration of haemoglobin (Hb), the rates of glycation and the lifespan or amount of red blood cell. Anand et al. [15] stratified their findings according to the three ethnic groups included in their study population, East Asian, Chinese and European, and reported ethnic variation in the sensitivity and specificity for both HbA1c and FPG tests. Similar ethnic differences in sensitivity and specificity at equivalent cut-off points were observed across the studies in this review.

Genetic variants (e.g. HbS trait, HbC trait) and chemically modified derivatives of haemoglobin (e.g. carbamylated Hb in patients with renal failure, acetylated Hb in patients taking large amounts of aspirin) can affect the accuracy of HbA1c measurements. Any condition that shortens erythrocyte survival or decreases mean erythrocyte age (e.g. recovery from acute blood loss, haemolytic anaemia) will falsely lower HbA1c. For example, iron-deficiency anaemia is reported to increase test results. In some patients with haemoglobinopathies such as D Punjab, sickle cell disease, very high fetal haemoglobin or abnormal red cell turnover, HbA1c measurement may not be appropriate. Only two of the studies included in this review considered the possible influence of these factors, such as anaemia, renal or hepatic dysfunction, and excluded affected patients [22,23].

The lack information provided on the detection and exclusion of variant haemoglobins in the majority of studies is a limitation of this review, however, haemoglobinopathies are unlikely to be present in numbers that will impact on the overall findings, given the ethnicity of the source populations and the sample sizes (1.7% prevalence found in the UK screening programme [8]). Clinically silent variants have now also been raised as a potential influence on HbA1c analysis [32]. However, this observation was based on four cases drawn from a Caucasian population of about 500 000, so the overall chances of inclusion of undetected variant haemoglobin cases in the studies under review remain extremely small. Therefore, variant haemoglobins, clinically important or silent, are unlikely to impact on the diagnostic test outcomes reported here. However, the potential presence of variant haemoglobin will need to be considered when these tests are applied at the individual level, especially for individuals and/or populations at high risk of variant haemoglobins.

Second, variation in the prevalence of risk factors may also have affected the results of the primary studies. Only two studies recruited from the general population through random or self-referred sampling [16,17], whilst the others enrolled participants with at least one major risk factor for Type 2 diabetes. The exceptionally high diabetes prevalence (44%) reported for the UK community-based sample [19] would likely have been related to referral bias, as this study included only those patients referred for OGTT. Only three studies [17,21,24] provided data on gender distribution, and one of these studies examined the estimates stratified by gender [17]. This study observed a higher sensitivity but a lower specificity among females than males in diagnosing diabetes for both FPG and HbA1c. Saydah et al. [18] demonstrated higher sensitivity and specificity of fasting glucose and HbA1c for the 60–74 age group compared with the 40–59 year-old group. Other studies with a wide age distribution did not provide age-specific sensitivity and specificity estimates.

It is known that variation in collection, storage and processing time of blood can influence the results of blood glucose assays. It is recommended, therefore, that blood for fasting plasma glucose analysis should be collected after the individual has fasted overnight and that plasma be separated from the cells within 60 min. If whole blood is used, the sample should be kept at 0–4 °C or centrifuged or assayed immediately. Glycolytic inhibitors such as sodium fluoride do not inhibit glycolysis initially and the glucose levels will be 0.5–1.0 mmol/l lower than if glucose was measured immediately [33]. Such methodological detail ideally should have been compared between the studies in this review, but this level of detail was not provided in any of the reviewed studies.

The studies in this review reported different optimum DCCT-aligned cut-off points for HbA1c for the early detection of Type 2 diabetes. However, three studies observed the same cut-off point of ≥ 6.2% as optimum [19,20,23]. Ko et al. suggested HbA1c ≥ 6.1% as the optimum cut-off point in their cross-sectional study [24]. Further follow-up of their cohort with IGT over 18 months provided further confirmation of 6.1% as the optimum cut-off point for early detection of diabetes as further cases were detected [34]. These findings are consistent with previous follow-up studies that also identified a cut-off point of ≥ 6.1% for HbA1c as the optimum cut-off point for detection of diabetes [35,36].

In 1994, McCane et al. examined the relationship between diabetes complications and concomitant results of OGTT, fasting glucose and HbA1c, and reported that all three significantly predicted the development of retinopathy and nephropathy [13]. In 2000, Ito et al. showed a high correlation between all the three measures and again HbA1c ≥ 6.1% was reported as the optimum cut-off point [12].

The consistency of results is not surprising given that FPG, OGTT and HbA1c tests reflect different underlying aspects of the same pathology of hyperglycaemia. The FPG depends on baseline insulin secretion and hepatic glucose output, whilst the OGTT, especially the 2-h glucose value, is influenced by peripheral insulin resistance, and HbA1c is a combination of glucose and protein. All these test outcomes are potentially useful for directly predicting increased morbidity and premature mortality.

Glucose can react with many different proteins and cause structural alterations and subsequently impair protein and tissue function. According to Jesudason et al. [20], HbA1c has much lower intra-individual coefficient variation (CV) than OGTT and FPG, and HbA1c may better reflect the risk of long-term micro- and macrovascular complications than OGTT and FPG. One study in the USA has shown one-quarter of those with previous IGT revert to normal glucose tolerance [37]. In another follow-up study, Little et al. [38] reported that 68% of those with elevated HbA1c (> 6.1%) at baseline progressed to diabetes compared with 28% of those with normal HbA1c at baseline, suggesting that, in subjects with IGT, HbA1c may be useful in determining who will likely progress to diabetes [39].

Earlier studies also reported HbA1c ≥ 7.0% as the diagnostic criterion to trigger pharmacological intervention for diabetes [40,41]. Data from Peters et al. showed that using an HbA1c > 7.0% as a cut-off to diagnose diabetes mellitus (DM) had a 99% sensitivity, and a specificity of 99.6%[39]. The results of the DCCT showed that an HbA1c of 7.0% was associated with an increased risk of microvascular complications, such as diabetic retinopathy and nephropathy. For macrovascular complications, a new 6-year large prospective study showed that HbA1c (ranging from 5.0 to 6.9%) significantly predicted all-cause mortality, coronary and cardiovascular disease independent of age and other classic risk factors [41]. In order to minimize the risk of developing complications, International Diabetes Federation (IDF) 2005 guidelines advise that people with diabetes should maintain a DCCT-aligned HbA1c < 6.5%.

The findings of this review combined with previous studies [12,35,36,38,42] suggest the optimum cut-off point of HbA1c to be ≥ 6.1% or ≥ 6.2% for the diagnosis of diabetes. McCarter et al. [29] showed that HbA1c has genetic inter-individual biological variation, but low intra-individual variation which will be useful in the early detection of diabetes and prediction of risk of long-term complications. HbA1c therefore is not only a marker of hyperglycaemia via measuring the mean blood glucose (MBG), but also reflects the risk of both long-term macro- and microvascular diabetic complications. HbA1c results arguably have the advantage of being more meaningful for clinical workers, and even for non-diabetic patients.

As in the studies included in this review, all HbA1c results should be reported according to an international standardized reference. Studies using non-DCCT standardized methods should report DCCT-aligned results, or DCCT traceable results, and best practice may be to include conversion equations to facilitate cross-study comparisons. The future basis for international standardization will likely be the reference system developed by the IFCC Working Group on HbA1c Standardization. The IFCC method is more specific and accurate than other methods as there is no interference from abnormal haemoglobins such as HbS and HbC, and no interference from acetylated or carbamylated Hb. Excellent inter- and intra-laboratory precision has been demonstrated [42].

An international working group was established in 2004 (ADA/EASD/IDF Working Group of the HbA1c Assay) to facilitate standard global reporting. This group also recommend that the IFCC method be the gold standard reference method for calibration of all instruments and methods used in assaying HbA1c[33]. This group continues to advocate for consistency in existing methods and reporting. Other new approaches are also being considered based on mean blood glucose equivalents [8]. These methods need to be linked to existing methods including HbA1c, and their diagnostic value determined in relation to diabetes, as well as the influence of other cofactors, such as age, sex and ethnic group.

As stated previously, the sample sizes in the reviewed studies make it unlikely that they will have included participants with variant haemoglobins in numbers that would be likely to influence the review findings. However, HbA1c results are likely to be influenced by the presence of variant and abnormal haemoglobins [8], so the conclusions in relation to cut-off levels from this review may or may not be relevant to this subgroup. Therefore, the diagnoses of diabetes should not be based on HbA1c alone for those identified as being at increased risk of haemoglobinopathy, including mildly anaemic individuals or those from an ethnic background.

Conclusions

The following conclusions can be drawn from this systematic review. First, HbA1c and FPG are both equally effective as screening tools in early detection of Type 2 diabetes, but neither of the tests is effective in detecting IGT. An OGTT is therefore still required to diagnose IGT. However, previous studies have demonstrated a low intra-individual variation for HbA1c and that this test is a good predictor for both the micro- and macrovascular complications of diabetes. Currently, the cost of HbA1c is higher than FPG, but the additional benefits of the HbA1c test in predicting costly clinical complications may make this a cost-effective choice.

Second, this review reinforces the need for the standardization of HbA1c measurements worldwide to allow meaningful comparison of results across laboratories. The current advisable method is ion-exchange HPLC or immunological assay based on the DCCT/UKPDS where the normal range is approximately 4.0–6.1%. The cut-off for HbA1c is recommended as either ≥ 6.1% (≥ 2 sd above the normal mean) or ≥ 6.2%, as found to be the optimum cut-off point in most studies. In future, it is likely that the IFCC reference method will become the international standard [33], but population-specific cut-off points should also be considered, as the optimum cut-offs have been shown to vary with ethnicity, age, gender and the population prevalence of diabetes.

Finally, further investigation of the validity of all three tests in predicting diabetes-related mortality and morbidity, especially in relation to macrovascular disease, is urgently needed to draw firm conclusions on the best screening tool for Type 2 diabetes.

Competing Interests

None to declare.

Acknowledgement

The authors would like to particularly thank Dr Alicia Jenkins for her comments and advice during the preparation of this paper, and Dr Carmel Apicella for assistance in contacting international authors.

Ancillary