Screening for type 2 diabetes: a short report for the National Screening Committee

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 11/18/01. The contractual start date was in August 2011. The draft report began editorial review in September 2012 and was accepted for publication in April 2013. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors' report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Note

this is a ‘short report’ that has been produced with only one-third of the resources that would be available for a full technology assessment report and it cannot be as comprehensive as a full report would be.

Copyright statement

© Queen's Printer and Controller of HMSO 2013. This work was produced by Waugh et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Type 2 diabetes mellitus

People can have type 2 diabetes mellitus (T2DM) with none of the classical symptoms, such as the passing of larger volumes of urine (polyuria) and thirst. Some may have had some symptoms when diagnosed, but have not recognised them as due to diabetes.

Sometimes by the time people are diagnosed with diabetes, they have developed complications such as retinopathy, due to an effect of diabetes on small blood vessels (microvascular disease). About half of the UK Prospective Diabetes Study (UKPDS) recruits had some sign of complications at entry, mostly retinopathy in 36%,1 although the proportion who have retinopathy at diagnosis is now lower than in past studies, with only about 2% having referable retinopathy at first screening. 2 (Note that patients who had established macrovascular complications, such as angina, were excluded from UKPDS.)

However, the main risk to health in undiagnosed T2DM is an increased risk of cardiovascular disease (CVD), in particular ischaemic heart disease (IHD), because of damage to the arteries (macrovascular disease). Indeed, the first manifestation of diabetes could be a heart attack, often fatal.

Early detection of diabetes would lead to measures to reduce the risk of heart disease, such as the use of statins to lower cholesterol, and treatment for raised blood pressure (BP), as well as reduction of blood glucose levels, initially by diet and exercise, and supplemented with glucose-lowering drugs if necessary.

It is known that a proportion of people with T2DM are undiagnosed. In the age group of 52–79 years, the English Longitudinal Study of Ageing (ELSA)3 in 2004–5 found that almost 20% of those with diabetes were undiagnosed, with a higher percentage in men (22%) than women (12%). The overall prevalence was 9.1%, with 1.7% undiagnosed. Predictors of undiagnosed diabetes included body mass index (BMI), waist circumference, systolic blood pressure (SBP) and triglycerides. Diagnosis was based on a single fasting plasma glucose (FPG) value of ≥ 7.0 mmol/l and would therefore miss those whose diabetes is manifested mainly by postprandial hyperglycaemia.

The authors of the ELSA study3 note that the proportion undiagnosed has fallen, and attribute this to increased opportunistic screening in general practice.

The results from the pilot screening programme in England support this, with an overall prevalence of 4.08%, including 0.54% undiagnosed. However, uptake of screening was only 61%4 and the non-responders might have had a higher proportion with undiagnosed diabetes.

Prevalence

The increase in reported prevalence of T2DM depends on a number of factors, including:

• An increase in the incidence of T2DM, related to rising levels of overweight and obesity. Data from the Framingham study show that almost all of the increase in the USA in diabetes prevalence is in the obese category. 5

• Demographic change – half of all people with diabetes are > 65 years of age, so an increase in the number of people over that age will increase the prevalence of diabetes.

• A fall in the age at onset of T2DM – people contracting diabetes earlier in life, probably because of earlier weight gain and reduced physical activity compared with previous generations. 6

• Changes in the definition of diabetes, with the diagnosis made at a lower level of FPG.

• Better survival with diabetes because of improved control of blood glucose level, hypertension and cholesterol level.

• More complete recording of diabetes on general practitioner (GP) computer systems.

• Better detection of undiagnosed diabetes by opportunistic case-finding or practice-based screening, linked with greater public awareness of diabetes.

The York and Humber Public Health Observatory (YHPHO) diabetes report7 estimated that 40% of the rise in crude prevalence was due to changing age and ethnic group structure, and 60% to obesity and overweight.

The prevalence of T2DM is closely linked with that of overweight. The proportion of the population that is overweight or obese (BMI of ≥ 30 kg/m²) has been increasing in recent years.

The importance of BMI in the incidence of T2DM is shown in Figure 1. There is a close relationship between BMI and the incidence of T2DM, and it is worth noting that it starts well below the obesity range.

FIGURE 1.

Age-adjusted incidence rates of diabetes as a function of baseline BMI in 30- to 55-year-olds (both sexes), based on data from Ford et al. (1997). 8

In the Cambridge centre of the ADDITION (Anglo-Danish-Dutch Study in General Practice of Intensive Treatment and Complication Prevention in Type 2 Diabetic Patients Identified by Screening) study, high proportions of people with screen-detected diabetes had risk factors for CVD. 9 Almost all were overweight or obese (mean BMI = 32.5 kg/m²); 86% had hypertension, 75% had dyslipidaemia, and many of those with hypertension and dyslipidaemia did not have good control of BP or lipids. Hence, those detected by screening form a group in which CVD risk can be reduced by combined treatment. The ADDITION study is described in detail in Chapter 4. Another implication is that first-stage screening based on weight would identify most people with undiagnosed diabetes.

Screening

The UK National Screening Committee (NSC) has reviewed its policy on screening for T2DM at intervals. A review in 2006 was underpinned by a Health Technology Assessment (HTA) report that included a systematic literature review and economic modelling of the case for population screening. 10 The HTA report found that the case for screening for undiagnosed diabetes and for IGT was becoming stronger because of greater options for the reduction of CVD and because of the rising prevalence of obesity, and hence of T2DM. (Details of methods are given in Appendix 1.)

What should we be screening for?

Diabetes is defined on the basis of high blood glucose levels. The key feature of the classification is that the diagnosis is based on the level at which the risk of retinopathy starts. At the risk of a little oversimplification, people with glucose levels below the threshold do not get retinopathy; those with levels above the threshold are at risk of retinopathy, with the risk increasing as glucose levels rise further. This was based on three studies, described in the report of the expert committee of the American Diabetes Association (ADA). 20 A very large recent study from the DETECT-2 (Evaluation of Screening and Early Detection Strategies for Type 2 Diabetes and Impaired Glucose Tolerance) Group21 also found clear thresholds for moderate retinopathy of 6.5 mmol/l for FPG and 6.5% for HbA_1c. The threshold for the 2-hour plasma glucose (PG) level was less distinct, at around 10.1–11.2 mmol/l.

The DETECT-2 Group21 used ‘moderate retinopathy’. Some retinopathy, such as microaneurysms only, has been reported in IGT from the Diabetes Prevention Program. 22

However, the risk of heart disease increases at lower levels of hyperglycaemia than diabetes. So for public health purposes, there could be advantages in having screening for diabetes done as part of assessment of cardiovascular risk and the English vascular risk assessment programme provides an opportunity for that.

Decision point

Hence, if one aim of screening is to reduce heart disease, we should look not only for diabetes, but also for IGT. Even if we did look only for diabetes, we would identify many with IFG and/or IGT, depending on which test was used and what cut-offs were chosen.

The aims of treatment might be:

• For those with definite diabetes, reduction of the risk of retinopathy and nephropathy, by reduction of PG, initially trying diet and exercise, but using drug therapy when indicated.

• For those with PG levels in the IFG and IGT ranges, prevention of progression to diabetes, by diet and exercise, or by drug therapy (metformin) if indicated.

• For all of the above, measures to reduce cardiovascular risk, by measures other than the glucose control ones already mentioned, such as qualitative improvements in diet, cholesterol-lowering measures (such as statins), BP control and anti-obesity measures.

There could be large implications for workload and costs. About 0.5% of the population may have undiagnosed diabetes, but there may be 10% with IGT. Before any screening was started, there would need to be careful planning of workload, involved in both screening and follow-up. Screening might be introduced in a phased manner in order to avoid overload.

Should screening be selective or universal over the age of 40 years?

It is assumed that screening will be selective by age because T2DM is much less common in younger age groups. The first question is whether or not other selection criteria should be applied. Given the link between T2DM and overweight, the obvious next criterion is BMI.

Organised screening could be a three-stage process, with the first stage being selection from the general population (using general practice registers or self-completed questionnaires) of those likely to be more at risk than average, the second being testing of blood glucose levels, and the third being confirmation (or not) of raised blood glucose level.

Testing only people who are at higher than average risk means that a higher proportion of those who will be tested for glucose will be positive. The number needed to screen to detect each true-positive will be lower, and the whole programme will be more cost-effective. However, the distribution of risk needs to be considered. As the risk threshold is raised, the proportion of those with diabetes will rise, but the absolute number detected may fall. Because of the shape of the distribution curve, most people with undiagnosed diabetes may be in the middle risk band.

Many risk-scoring systems have been used around the world5,42 based on different questionnaires. Risk factors were listed by Paulweber et al. (2010)43 in an analysis to support development of European guidelines. This analysis identified both non-modifiable and modifiable factors, listed in Table 5.

Different risk scores seem to have different specificity and sensitivity. The commonly used Finnish Diabetes Risk Score (FINDRISC) score is based on a sample from the population in Finland. This scorecard predicts the probability of T2DM development over the following 10 years. It uses information such as age, BMI, sex, antihypertensive medication history and some lifestyle factors. This risk score has high sensitivity towards undiagnosed diabetes. It has been validated both in Germany44 and Italy. 45

Risk-scoring systems

There are various scoring systems. The Finnish questionnaire-based system, FINDRISC, includes age, BMI, waist measurement, physical activity, diet (vegetable, fruit and berry consumption), treatment for hypertension, any previous hyperglycaemia and family history. 49

It might be easier if we could use a smaller set of indicators, and there would be little difference in predictive power, as age and BMI provide most of that. 50

One advantage of using a smaller set of risk indicators is that computer systems in general practices will usually have the necessary data – certainly age, drug treatment, comorbidities and, usually, BMI. They are less likely to have family history, and probably will not have waist measurements. But it means that the first stage of any screening system could use existing data at little extra cost.

One scoring system using data that should be available on GP systems was developed by Hippisley-Cox et al. (2009),51 and comprises the following factors: ethnicity, age, sex, BMI, smoking, family history of diabetes, Townsend deprivation score, treated hypertension, CVD and current use of corticosteroids. It was developed using the QResearch database (Version 19; University of Nottingham, Nottingham, UK ) and is known as the QD score. The authors report receiver operating characteristic (ROC) statistics of 0.85 for women and 0.83 for men for detection of diabetes. The results indicate that people from South Asian ethnic groups have different risk factors. These include different lifestyle habits, particularly smoking, along with a family history of diabetes. These factors make them more vulnerable to getting diabetes earlier in life than Caucasians, with a four- to fivefold variation in risk among different ethnicities. So ethnicity should be included in the risk evaluation.

There has been an independent validation of the QD score by Collins and Altman,52 which reported that it performed well in distinguishing between those who develop diabetes and those who do not.

Other scoring systems include the Cambridge Risk Score (CRS), again based on data available in GP systems (see Appendix 2). 53 Those in the highest quintile of the CRS had 22 times the risk of diabetes as those in the lowest quintile, and 54% of incident cases were in the top quintile. 50 It takes into account ethnicity. This score is also validated in the Danish population. 54

Body mass index is probably the single most powerful factor, and other factors may add much less to the detection rate. One review of risk scores by Witte et al. (2010)55 for predicting undiagnosed diabetes concluded that a combination of age and BMI was as good as more complex scores. However, if all of the data are on GP systems, then they may as well be used.

A very thorough systematic review of risk-scoring systems by Noble et al. (2011)56 identified 145 risk prediction models, and described 94 in detail. They noted that almost 7 million people had been in risk-scoring studies, with ages ranging from 18 to 98 years, and follow-up ranging from 3 to 28 years. They estimated that risk scores for diabetes were appearing at a rate of one every 3 weeks. Noble et al. (2011)56 set out to look for scores that were ‘sufficiently simple, plausible, affordable and widely implementable in clinical practice’. They noted that some scores included expensive laboratory biomarkers.

Noble et al. (2011)56 applied a set of quality criteria to the scores. These were:

• generalisability external validation by a separate research team on a different population

• statistically significant calibration i.e. that predictions match observations

• discrimination that the score reliably distinguishes high-risk people from low-risk people

• usability defined as the score having 10 or fewer components.

They noted that there were two broad categories of score: those that were completed by people themselves, based on questionnaires, and those that were based on data held by health-care providers such as GPs.

Noble et al. (2011)56 did not recommend any one system, making the point that the choice would depend on local circumstances. However, they did identify seven very good systems. These included two from the UK, the CRS and the QD score. In terms of area under ROC curves (AUROCs), the QD score came out better, with external validation AUROCs of 0.80 for men and 0.81 for women, compared with the Cambridge AUROC of 0.72. It should be noted that AUROCs can be altered by changing thresholds for positivity. The Cambridge score of 0.72 uses a threshold of 0.38.

Both of these scores include age, sex, family history of diabetes, BMI, current use of corticosteroids, treatment for hypertension and smoking. The QD score adds ethnicity, deprivation score and CVD. All of the QD components should be on GP computer systems.

Hence, if screening for T2DM in the UK was to be based in primary care and to use a first stage of selection by risk, the QD score seems best.

Another of the Noble et al. (2011)56 ‘top seven’ scores was FINDRISC, which has been validated in a UK population. So if screening in the UK or part thereof was to be based on self-scoring by completion of questionnaires, then FINDRISC could be used. The cut-off might be 12, based on ROC curves from a study by Martin et al. (2011)57 of Munich, Germany. Martin et al. (2011)57 measured HbA_1c levels in 12,773 blood donors of whom almost 8% scored 12 points or more in the FINDRISC questionnaire and had HbA_1c values of ≥ 5.6%. The 8% and age- and sex-matched controls were invited for an OGTT. Unfortunately, only about 10%, 671 participants, came for the OGTT, some citing reasons such as time costs.

Using a FINDRISC score of 12 as the cut-off, over half had some degree of impaired glucose regulation (IGR) in the OGTT: 9.7% diabetes, 32.6% IFG, 5.7% IGT, 11.8% both IFG and IGT, and 40% having normal glucose levels. Martin et al. (2011)57 compared the percentages of those with diabetes in two FINDRISC subgroups: medium risk, with scores 12–14, and high risk, with scores ≥ 15. Half (31) of the newly diagnosed group were in the high-risk group and 40% in the medium-risk group. The proportions with IFG, IGT or the combination were similar in the medium- and high-risk groups.

Martin et al. (2011)57 state that:

The high acceptance of the FINDRISC questionnaire and the HbA_1c testing contrasted with the rather low number of participants that were willing to participate in the OGTT. This finding indicates that simple screening procedures are necessary to achieve compliance and that the OGTT is not attractive for people who would otherwise be accessible for diabetes screening.

They conclude that a combination of FINDRISC cut-off of 12 and HbA_1c value of 5.9% would be optimal, as that would identify 56 of the 65 people diagnosed with diabetes.

Abbasi et al. (2012)58 used data from the European Prospective Investigation into Cancer and Nutrition (EPIC-NL) study to validate 25 models for predicting future T2DM. Two approaches were used, depending on what data were available, with data from 38,379 people being used for validating ‘basic’ models (with no blood test data) and data from 2506 individuals used for validating prediction models that included biochemical data. Most models performed well, but most overestimated the risk. The measure of capture used was the c-statistic, which is comparable with the AUROC. FINDRISC had a c-statistic of 0.81 at 7.5 years, in both full and concise form. QD score was lower, at 0.76. AUSDRISK scored 0.84 and EPIC-Norfolk37 0.81.

One issue has been raised by Griffin et al. (2000)53 and the Dutch Hoorn group. 59 Griffin et al. (2000)53 wondered about the dangers of reassurance in those who have high-risk scores but who do not have hyperglycaemia – will they feel they are able to persist with unhealthy lifestyles? And in the Hoorn study, Spijkerman et al. (2002)59 found that the group with high-risk scores, but who did not have diabetes on glucose testing, had a CVD risk that was almost as high as those who were glycaemia-positive. As there were more of the risk-positive but glucose-negatives, they had more cardiac events, leading the authors to comment that:

It may be of greater public health benefit to intervene in the screen positive group as a whole rather than only in the relatively small group who on subsequent biochemical testing have an increased glucose concentration.

However, a recent study by Paddison et al. (2009)60 from the Cambridge MRC group found that people with negative diabetes screening tests were not so reassured that they would have an adverse shift in health behaviours.

In the EPIC-Norfolk study,38 adding a measure of hyperglycaemia, in this case HbA_1c, to the Framingham risk score, added little to the predictive value for CHD. That might imply that glucose testing would not be necessary. However, their focus was on CVD, and detection of diabetes would also lead to reduction of microvascular events, for example by screening for retinopathy.

The tests for blood glucose include:

• casual (non-fasting) blood glucose

• FPG

• glucose tolerance testing, combining fasting and 2-hour levels (OGTT)

• the 50-g glucose challenge test (GCT), which has been used mainly for screening for gestational diabetes

• HbA_1c, which reflects blood glucose over the previous 3 months [assuming red blood cells (RBCs) of normal longevity and the absence of haemoglobin variants].

Casual blood glucose is usually discounted because of its variability and poor sensitivity (at levels which give acceptable specificity). 61

The OGTT is expensive, inconvenient (and sometimes unpleasant – in some people the glucose load causes nausea) and has poor reproducibility. It may also not fit easily into the primary care setting in which most T2DM is diagnosed,62 although some practices have carried out oral glucose tolerance testing. 63

The choice of test depends on what we are screening for. FPG is reliable, in the sense of showing less day-to-day variability than OGTTs, and will identify people with diabetes and IFG. However, it will miss those people with IGT, who have a higher IHD risk than those with IFG.

As with all tests, there is a trade-off between sensitivity and specificity, nicely shown by Hu et al. (2010)64 (Table 6).

Note that although the cut-offs of FPG 6.1 mmol/l and HbA_1c 6.1% give the same sensitivity (they were chosen from ROC curves) they will not identify exactly the same people.

Glycated haemoglobin

An expert group in the UK65 has issued a position statement on the implementation of the World Health Organization (WHO) 2011 guidance on the use of HbA_1c levels in the diagnosis of diabetes. The WHO guidance was that, given good quality assurance (QA) testing, a HbA_1c value of 48 mmol/mol (6.5%) was recommended as the cut-off point for diagnosing diabetes but that a lower value did not exclude diabetes.

The main points in the UK expert group statement are:

• UK laboratories meet QA requirements.

• HbA_1c testing should be based on laboratory measurement of a venous blood sample. Point of care results should always be confirmed in a laboratory.

• One positive HbA_1c result is enough in the presence of symptoms, but should be repeated in the absence of symptoms. (The implication for screening is that repeats would be required.)

• People with HbA_1c values in the range 6.0–6.4% (42–47 mmol/mol) were at high risk of diabetes and should receive lifestyle advice, be warned to report symptoms and should have their HbA_1c level checked annually.

• Some people with HbA_1c values of < 6.0% (42 mmol/mol) may be at high risk of diabetes, and they should be monitored as above.

The ADA Expert Committee66 on the diagnosis and classification of diabetes mellitus summarised the advantages and disadvantages of HbA_1c for the diagnosis of diabetes. The Committee listed the advantages as:

• HbA_1c testing measures average glycaemic levels over a period of 10 weeks or so, and is therefore more stable than FPG testing, and especially 2-hour OGTT.

• Fasting is not required and the test can be carried out at any time of day.

• The precision of HbA_1c testing can be as good as that of PG testing.

• HbA_1c level is the test used for monitoring control of diabetes and correlates well with the microvascular complications; it may be useful to use the same test for diagnosis and monitoring.

• It has been shown by meta-analysis that when using a statistical cut-point of two standard deviations (SDs) above the non-diabetic mean value, HbA_1c testing is as good as FPG testing and 2-hour PG in terms of sensitivity (66%) and specificity (98%).

The disadvantages were identified as:

• Internationally, there had been a profusion of assay methods and reference ranges. However, this can be overcome by standardisation to the DCCT (Diabetes Control and Complications Trial) assay.

• HbA_1c level may be affected by other conditions that affect the life of the RBC; results may then be misleading. This could be a particular problem in ethnic groups in which haemoglobinopathy is common.

• A chemical preparation for uniform calibration standards had become available only recently and was not universally available.

However, with the exception of the other conditions, these disadvantages need not apply in a national screening system that would include quality control measures. Therefore, there is a case for using HbA_1c as the screening test, particularly in view of its correlation with cardiovascular risk across a wide spectrum. As mentioned above, Khaw et al. (2004)37 noted that the rise in cardiovascular events with rising HbA_1c level starts well below the diabetic range. Indeed, they point out that when both diabetes and HbA_1c level are included in the statistical analysis HbA_1c dominates; as Gerstein (2004)67 argues in an editorial:

. . . the glycosylated haemoglobin level is an independent progressive risk factor for cardiovascular events, regardless of diabetes status.

The glucose levels for the diagnosis of diabetes were based on the relationship between PG and retinopathy. Recently, a similar study by Sabanayagam et al. (2009)68 has examined the relationship between HbA_1c level and retinopathy. Using the presence of moderate retinopathy as the indicator of diabetes, the authors suggest a HbA_1c threshold of 6.6%. This is very close to the 6.5% suggested by the DETECT-2 Group. 21

The ADA position statement69 in January 2010 recommended a cut-off of 6.5% for diagnosing diabetes, based on retinopathy risk. They recommended a cut-off of 5.7%, and hence a range of 5.7% to < 6.5%, for identifying those at high risk of diabetes. The arguments in favour of the 5.7% cut-off were that:

• The 6.0% to < 6.5% range misses a lot of patients who have IFG or IGT and who are at increased risk of diabetes. They cite studies reporting that people in the 5.5% to < 6.0% range have a 5-year incidence of diabetes of 12–25%.

• That unpublished NHANES data show that a HbA_1c value of 5.7% corresponds with a FPG of 6.1 mmol/l (i.e. IFG).

• That other unpublished NHANES data show that a HbA_1c cut-off of 5.7% has modest sensitivity (about 40%) but good specificity (81–91%) for IFG and IGT

• That other unpublished analyses indicate that a HbA_1c value of 5.7% is associated with a similar risk of diabetes to the high-risk group in the Diabetes Prevention Program.

As always, there is a trade-off between sensitivity and specificity. If we used a low cut-off of HbA_1c of 5.7%, there would be more false-positives. However, they are at higher risk of CVD than the rest of the population and would benefit from lifestyle measures. The only harm might be from the labelling as ‘pre-diabetic’.

In July 2009, an expert committee appointed by the European Association for the Study of Diabetes (EASD), International Diabetes Federation (IDF) and ADA published a report70 on the role of the HbA_1c assay in the diagnosis of diabetes. The key recommendations of this report were that:

• HbA_1c assay should be used as a diagnostic test for diabetes with a threshold of ≥ 6.5% for defining diabetes.

• That HbA_1c measurement has several advantages (both logistical and technical) over fasting glucose.

• Individuals whose HbA_1c values are close to the 6.5% threshold for diabetes (i.e. ≥ 6.0%) should receive demonstrably effective interventions aimed at preventing progression to diabetes.

• Testing should be by clinical laboratory instruments, not point-of-care instruments.

A cut-off level of 6.0% might pick up most people with IFG, but not all. Selvin et al. (2010)71 (from the ARIC study) reported that a HbA_1c cut-off level of ≥ 6.5% would detect 49% of those with FPG of ≥ 7.0 mmol/l, and a cut-off of 6.0% would detect 75% of those diabetic by FPG. In the band below (HbA_1c level of 5.5% to < 6.0%), only 3% were diabetic by FPG. In this band, the mean HbA_1c level was 5.7% and mean FPG was 5.8 mmol/l. A cut-off of 5.5% would detect 91% of those with diabetic FPGs.

In Germany, Peter et al. (2001)72 examined the value of HbA_1c testing using a cut-off of 6.5% in patients shown to be diabetic by OGTTs. Sensitivity was only 47% and specificity was 98.7%. Using a lower cut-off of 6.1% gave sensitivity of 71% and specificity of 92%.

In Wales, Morrison et al. (2011)73 reported a sensitivity of 67% and specificity of 57% using a 6.5% threshold.

A very useful study by Schottker et al. (2011)74 compared FPG and HbA_1c testing in the ESTHER (Epidemiologische Studie zu Chancen der Verhütung, Früherkennung und optimierten Therapie chronischer Erkrankungen in der älteren Bevölkerung) study in Germany among almost 10,000 subjects aged 50–74 years. They excluded people with diabetes at baseline and re-tested the others at 2 and 5 years. They classified those who had pre-diabetes at baseline using FPG levels of 100–125 mg/dl and HbA_1c levels of 5.7–6.4%. In this group, 23% had pre-diabetes only by FPG testing; 23% had pre-diabetes by both FPG and HbA_1c testing, and 54% had it only by HbA_1c testing. Hence, in total, FPG testing detected 46% and HbA_1c testing detected 77%. They did not carry out OGTTs.

Schottker et al. (2011)74 determined the risks of incident diabetes over the next 5 years. Compared with those who had neither raised FPG or HbA_1c levels, the RRs were:

• IFG alone 4.23

• pre-diabetic HbA_1c alone 3.05

• both FPG and HbA_1c pre-diabetic 7.81.

They concluded that screening for future diabetes should use both FPG and HbA_1c testing. They did not report incidence of CVD.

Moves towards global standardisation of HbA_1c measurement will help. 75 HbA_1c testing has advantages of not requiring people to be fasting, and its diagnostic accuracy now rivals that of PG. However, it should be noted that any cut-off will be arbitrary because for vascular disease there is a continuum of risk, unlike the dichotomy seen with moderate retinopathy.

There are ethnic differences in HbA_1c levels and the cut-off may have to be adjusted for different groups. A study from China by Bao et al. (2010)76 suggested a cut-off of 6.3%.

We need to distinguish the use of HbA_1c testing for diagnosing diabetes from its value in predicting vascular risk. In the latter case, it is correct to say that HbA_1c level is a good predictor of vascular risk on its own, but that once other traditional markers of vascular risk – such as hypertension, smoking, lipid level – are added, HbA_1c testing gives limited marginal benefit. 38

Reservations about the use of glycated haemoglobin

Various concerns about reliance on HbA_1c level as the diagnostic test for diabetes have been raised. Some of these come from the clinical biochemists and, therefore, need to be recognised.

The Association of British Clinical Diabetologists (ABCD) position statement77 on using HbA_1c testing for diagnosis (not screening) lists the advantages and disadvantages (Table 7).

Each of the first three disadvantages leads to misleading results.

The ABCD choose a HbA_1c range of 5.8–7.2% for intermediate hyperglycaemia and recommend another test, such as FPG or an OGTT, to confirm or exclude diabetes. They suggest that combined HbA_1c and FPG testing could be used for diagnosis.

Schindhelm et al. (2010)78 also warn that laboratory assays for HbA_1c level still show significant variability, noting that coefficients of variance ranged among methods from 1.7% to 7.6%. However, if there was a national screening system with QA systems, this should be less of a problem.

There has been debate about the lower cut-off level for HbA_1c. The SPHN working group79 noted that some groups advocate a HbA_1c range of 5.7–6.4% for defining non-diabetic hyperglycaemia (NDH), whereas others suggest 6.0% as the lower limit. Unfortunately, most studies simply report results for the whole band, whereas what we need is a comparison of the 5.7–5.9% range with the 6.0–6.4% range.

Cederberg et al. (2010)80 used the 5.7% cut-off and compared it with IGT and IFG as revealed by OGTTs. Diabetes was preceded by raised HbA_1c, IGT and IFG levels in 33%, 41% and 22% of cases, respectively, after 10 years. The converse may be more important – diabetes was not preceded by raised HbA_1c in 67%, although if screening were to be introduced in the UK, the repeat testing interval would most likely be 5 years.

Mostafa et al. (2010)81 from Leicester used data on OGTTs and HbA_1c from a cohort of 8696 subjects to compare proportions with abnormal results. Using the OGTT, 3.3% were diabetic, and of these about one-third had a HbA_1c level of < 6.5%. Using a HbA_1c level of 6.5% as the threshold for diabetes increased the prevalence to 5.8%, but on OGTT over half had IGT or IFG. Of 595 people, 198 were identified as diabetic by both OGTT and HbA_1c testing, 93 by only OGTT, and 304 only by HbA_1c. The paper does not give details of how many who were diabetic by OGTT, had the diagnosis made by the FPG or the 2-hour PG, or both. All of those who were recognised as diabetic on OGTT had the OGTT repeated – one-third were not diabetic on the second OGTT. Mostafa et al. (2010)81 noted that a HbA_1c cut-off level of 5.7% would identify 51% of their cohort as abnormal.

A later abstract from the same group82 reported that HbA_1c testing and the OGTT gave similar numbers for incident diabetes – but not the same people.

Borg et al. (2010)83 from Denmark also compared the characteristics of those diagnosed by OGTT and HbA_1c but, again, give no details of the OGTT time points responsible for diagnosis. Using a HbA_1c cut-off level of ≥ 6.5%, 6.6% were identified as diabetic compared with 4.1% using the OGTT. Almost 58% of those recognised as diabetic using the OGTT were not picked up by HbA_1c testing. In terms of cardiovascular risk profile, those people considered diabetic by HbA_1c testing, but not using the OGTT, had as high a risk (actually higher, but not statistically significantly so). Hence, OGTT and HbA_1c testing appear to be detecting groups that only partly overlap.

Lorenzo et al. (2010)84 from the IRAS (Insulin Resistance Atherosclerosis Study) reported that HbA_1c testing was less sensitive than IFG or IGT for detection of risk (not diabetes), but what was meant by this was that HbA_1c testing classified fewer individuals as having abnormal glucose tolerance – it was not about diabetes. No specificity was reported.

Valdes et al. (2011)85 from Asturias found that for prediction of diabetes in those not diabetic when first screened, FPG (2-hour post challenge) and HbA_1c testing had similar ROC curves [area under curve (AUC) 0.83, 0.79 and 0.80, respectively], but that a combination of HbA_1c and FPG testing did better (AUC 0.88).

The incremental risks of higher HbA_1c levels vary among outcomes. Selvin et al. (2010)86 took a HbA_1c range of 5.0% to < 5.7% as the reference range (RR = 1.0). For diabetes, RRs for HbA_1c levels of 5.7% to < 6.5% and ≥ 6.5% were 3.0 and 13.7, respectively, but for CHD the RRs were 1.6 and 1.9, respectively. 86

Skriver et al. (2010)87 from the Danish arm of the ADDITION trial, have provided data on the specificity of HbA_1c testing. A high-risk group (identified by questionnaire and then by a second-stage casual blood glucose or HbA_1c level) underwent OGTTs and then the HbA_1c levels of those with NGT (defined by OGTT) were examined. Only 0.4% had HbA_1c of ≥ 6.5%; 6.7% had HbA_1c levels in the range 6.0–6.49%, and 93% had HbA_1c levels of < 6.0%.

The sensitivity of HbA_1c testing was assessed in a Paris study by Cosson et al. (2010). 88 In a group mainly composed of obese women, they compared FPG and HbA_1c with the 2-hour OGTT. Using FPG testing alone, 70% of people with IGR would have been missed because most had isolated IGT. The sensitivity of HbA_1c testing for detecting an abnormal 2-hour PG using a cut-off of ≥ 6% was only 37%.

From Brazil, Cavagnolli et al. (2011)89 also reported low sensitivity. They carried out OGTTs in 498 people of whom 115 were recognised as diabetic by the OGTT: 26 by FPG alone, 54 by 2-hour PG alone and 35 by both. But only 56 were identified as diabetic by HbA_1c levels of ≥ 6.5%. HbA_1c testing was stronger in specificity. Using a cut-off of ≥ 6.0% identified 167 people, of whom 65 had diabetes, 41 had IFG, 52 had IGT, and only nine had NGT.

Pajunen et al. (2010)90 (from the DPS) carried out two OGTTs to be sure of correctly diagnosing people with IGT. They then monitored participants for the development of diabetes. They found that the sensitivities of HbA_1c levels of ≥ 6.5% for detecting new diabetes were 35% in women and 47% in men compared with two OGTTs. A cut-off level of 6.0% gave sensitivities of 67% and 68%. Specificities were better; using a cut-off HbA_1c level of < 6.5% gave specificities of 90% in women and 91% in men.

Turning to the usefulness of HbA_1c testing for detecting IGT at baseline, Pajunen et al. (2010)90 report that HbA_1c levels in patients with IGT confirmed by two OGTTs were:

• < 5% 8.5% of participants

• 5.0–5.9% 69.5% of participants

• 6.0–6.4% 14.3% of participants

• 6.5% or > 7.7% of participants.

So a HbA_1c cut-off level of 6.0% missed most people with IGT.

Pajunen et al. (2010)90 compared those who developed diabetes and had a HbA_1c level of ≥ 6.5% with those who developed diabetes and had a HbA_1c level of < 6.5%. The former group had significantly higher weight, BMI and FPG, but there were no differences in BP or cholesterol levels. They then reclassified the DPS patients as diabetic or not, based on a HbA_1c level of ≥ 6.5% and found that the results of the study, in terms of preventing progression to diabetes, would not have been statistically significant.

Older studies were reviewed by Bennett et al. (2006). 91 They identified nine studies that assessed the value of HbA_1c testing in screening for diabetes, using the OGTT as the reference standard. They concluded that the HbA_1c cut-off level should be 6.1%, which, for diabetes, had a sensitivity ranging from 78% to 81% and specificity of 79–84%. FPG testing, with a cut-off level of > 6.1 mmol/l, had poorer sensitivity (ranging from 48% to 64%) but better specificity (ranging from 94% to 98%). HbA_1c and FPG testing had, at those cut-off levels, only 50% sensitivity for detecting IGT. Bennett et al. (2006)91 noted that the need for cut-off levels varied among different populations.

Glucose challenge test

The 50-g GCT has been used extensively in screening for gestational diabetes but has seldom been studied in screening for T2DM. It can be used for people who have not fasted, which makes it more convenient.

Abdul-Ghani and De Fronzo (2009)92 have reviewed the evidence on relationships of FPG and PG at all time points after the 75-g OGTT. They make a convincing case for the 1-hour PG being the strongest predictor of later diabetes. This would suggest that it would be worth researching the value of the 50-g 1-hour GCT in screening for IGT and diabetes.

A study published since that review provides further support for the superiority of the 1-hour PG, although it was based on the 75-g OGTT, not the 50-g GCT: Joshipura et al. (2011)93 reported that among a group of non-diabetic people aged 40–65 years in Puerto Rico, the 1-hour PG had stronger associations with metabolic factors – such as BMI, waist circumference and body fat percentage – than the 2-hour PG.

In a recent study, Abdul-Ghani et al. (2011)94 compared various measures of blood glucose for predicting future T2DM. A HbA_1c level of 5.65% had an AUROC of 0.73, and FPG (126 mg/dl) had an AUROC of 0.75. However, the 1-hour PG (155-mg/dl cut-off) had a greater AUROC of 0.84. Adding HbA_1c testing to the 1-hour PG increased the AUROC slightly, to 0.87. The 2-hour PG had an AUROC of 0.79.

Phillips et al. (2009)95 compared the non-fasting GCT with HbA_1c testing and the OGTT (undertaken 1 week later) and found AUROCs of 0.90 for diabetes, and 0.79 for pre-diabetes (defined as IGT of IFG based on the 6.1-mmol/l threshold). For HbA_1c testing (probably > 6.0%, although not clear) the AUROCs were 0.82 for diabetes and 0.68 for pre-diabetes.

From the same group, Chatterjee et al. (2010)96 examined the cost-effectiveness of screening by random PG, the 1-hour 50-g GCT and the 75-g OGTT, compared with no screening. Their model included costs of testing and treatment (with metformin). They concluded that the GCT would be the most cost-effective test and also that screening would be cost-effective compared with no screening. This study is discussed in more detail in Chapter 6.

Jones et al. (2013)97 from Exeter reported results from a group of 2253 people aged > 40 years and not known to have diabetes. They were participating in the ‘Exeter 10,000’ study. They had two risk scores applied: Leicester and Cambridge. The latter was more sensitive (69%) but less specific (57%) than the Leicester one (62% and 69%). However, the interest here is in the results of HbA_1c and FPG testing; both result in diabetes being diagnosed in about 16%, but half of those diagnosed as having diabetes or at high risk by HbA_1c testing were not diagnosed by FPG testing and vice versa.

Summary

There is no perfect screening test. HbA_1c testing is accepted for diagnosis in people who are suspected to have diabetes, using a level of 6.5%, but is less reliable for screening because of its sensitivity. But it is convenient and a good marker of cardiovascular risk.

The OGTT has higher sensitivity but is inconvenient, and uptake is poorer.

The 50-g GCT looks good, and does not require fasting, but the evidence base is sparse.

The combination of HbA_1c and FPG testing might be a compromise option, but will miss some people who diagnosed as diabetic on the 2-hour OGTT.

Suggested conclusions

The first step in screening for diabetes and IGT should be selection by risk factor score.

The second stage should use HbA_1c level as the screening test, with 6.0% as the cut-off.

In the third stage, the diagnosis of diabetes should be confirmed by a second test of blood glucose: either FPG or HbA_1c testing. Two HbA_1c results of ≥ 6.5% would confirm the diagnosis.

Given the lack of agreement on the use of HbA_1c testing alone, we recommend that if the first HbA_1c value is in the 6.0–6.49% range then the third stage should meantime use both HbA_1c testing and either FPG testing, with 7.0 mmol/l used as the diabetes cut-off as in the standard definitions, or a 2-hour OGTT. It is unlikely that people with HbA_1c values in the range 6.0–6.49% will have FPG values of ≥ 7.0 mmol/l; this recommendation can be reviewed in the light of experience and FPG dropped if it does not contribute. However, more people may have postload hyperglycaemia and we might expect some with HbA_1c values in the 6.0–6.49% range to have 2-hour PG in the diabetic range.

We recommend research into the usefulness of the 50-g non-fasting GCT as a screening test.

Recent papers: cohort studies

The ADDITION study102 followed up 20,916 participants for a median of 7 years, all of whom had a HbA_1c level of > 5.8% or a random PG level of > 5.5 mmol/l. A HbA_1c level of > 6.5% was associated with increased HRs for all-cause mortality in the NGT groups, and in groups with IGT but not IFG. This relationship was present in both the subsets, with high and low heart risk scores at baseline. In people with IFG but not IGT, increasing HbA_1c level was not associated with significantly changing HR. Limited comparisons can be made between the effectiveness of HbA_1c, FPG and IGT for predicting cardiovascular risk factors because the population investigated was selected partly on the basis of HbA_1c risk score.

The ARIC Study71,103 followed a cohort of 11,057 people, without diabetes or heart failure at baseline, for a median of 14.1 years. This study found that a HbA_1c level of between 6.0% and 6.4% is associated with a doubling of risk – adjusted for age, race and sex – for incident heart failure in comparison with a HbA_1c level of between 5.0% and 5.4% (Table 8). The HR is reduced to 1.4 after adjustment for a range of CVD risk factors and fasting glucose levels. Compared with the group with HbA_1c levels of 5.0–5.4%, the group with HbA_1c levels of 6.0–6.4% were more likely to be smokers (31% vs 18%), had higher BMIs (29.7 kg/m² vs 26.6 kg/m²), were more likely to be treated for hypertension (42% vs 23.5%) and had a much higher proportion of African American people (51% vs 12%).

A FPG of 6.1–6.9 mmol/l was not indicative of an increased risk of incident heart failure above a reference category of 5.0 to 5.5 when adjusted for covariates and HbA_1c levels (see Table 8). Of the people with elevated HbA_1c levels (6.0–6.4%), the greatest risk for incident heart disease was observed in those with lower FPG:

• HR 3.57 (1.45–8.79) for FPG of < 5.0 mmol/l

• HR 2.19 (1.40 to 3.43) for FPG of 5.0–5.5 mmol/l

• HR 1.29 (0.86 to 1.94) for FPG of 5.6–6.0 mmol/l compared with reference category HbA_1c 5.0–5.4%, FPG 5.0–5.5 mmol/l.

Each 1% increase in HbA_1c level was associated with a 39% increase in heart failure risk after adjustment for covariates.

Hence, HbA_1c level predicted heart failure but FPG level did not.

The Strong Heart Study104 followed a cohort of 4549 American Indian adults for a median of 15 years. There was a non-significant trend towards higher risk of CVD in people with elevated but non-diabetic HbA_1c and FPG. This trend was stronger for HbA_1c than FPG but both were non-significant. Elevated HbA_1c of over 6.5% was associated with an increased HR for CVD 1.40 (95% CI 1.02 to 1.93) when correcting for a range of risk factors including FPG although the CI almost intersects zero (Table 9). Elevated FPG of over 126 mg/dl (7 mmol/l) was associated with a non-significant trend towards an increased HR for CVD 1.25 (95% CI 0.97 to 1.62) when correcting for a range of risk factors including HbA_1c (see Table 9).

These two cohort studies provide evidence that when considering cardiovascular risk, HbA_1c is more predictive than FPG.

Other studies have reported different results. In the AusDiab cohort study, Barr et al. (2009)31 followed 10,026 people without diabetes for 7 years. They found that in a model with FPG the effect of a SD increase in HbA_1c level (0.4%) on all-cause mortality of 1.1 (95% CI 0.9 to 1.2) and CVD mortality of 1.0 (95% CI 0.8 to 1.2) was not significant, but in a model containing HbA_1c a SD increase in FPG level (0.7 mmol/l) was associated with a significant increase in CVD mortality 1.3 (95% CI 1.0 to 1.6) but not all-cause mortality 1.1 (95% CI 0.9 to 1.2). However, in terms of application to a population screening programme, comparing SD increases provides less useful information than making comparisons above and below a cut-point that may be implemented. The differences in results between the studies may also be associated with the differing length of follow-up time. Barr et al. (2009)31 found that one SD increase in 2-hour PG levels (2.2 mmol/l) increased the HRs for all-cause mortality even with FPG in the model 1.2 (95% CI 1.1 to 1.3) or with HbA_1c in the model 1.2 (95% CI 1.1 to 1.3), suggesting that 2-hour PG may be the best predictor of all-cause mortality after 7 years' follow-up.

However, there are some curious features in this paper. The authors report that FPG, 2-hour PG and HbA_1c levels were classified into quintiles, but the numbers of participants in each quintile were very different. The ‘quintiles’ just appear to be bands, and unequal ones, by level of PG or HbA_1c.

Cardiovascular mortality was higher in the higher bands for all three measures. In most cases there were parallel increases in risk factors for CVD, such as hypertension, but the differences between lowest and highest band were different, being less for HbA_1c than some other variables. For example, the difference in SBP between highest and lowest bands was 16 mmHg for FPG, 15 mmHg for 2-hour PG but only 6 mmHg for HbA_1c. The differences in BMI between highest and lowest bands were 3.4 kg/m² for FPG, 3.4 kg/m² for 2-hour PG and 1.7 kg/m² for HbA_1c. The highest band was split in FPG and 2-hour PG so these figures use the lower half of the highest bands, which contained most of the participants.

Cross-sectional

The Danish Inter99 Study83 compared the cardiovascular risk profile of 6258 Danish participants with results of different measures of blood glucose. They used the PRECARD program to assess IHD risk. PRECARD is based on a suite of CVD risk factors. There was no significant difference between median PRECARD score for those identified as diabetic by the HbA_1c test or the OGTT, but there was a trend towards higher risk score for those classified as diabetic by both tests, implying that they are associated with different but overlapping risk factors (Table 10). Those diagnosed as diabetic by OGTT only had higher BMI scores (33% with BMI of ≥ 30 kg/m² compared with 27% diagnosed by HbA_1c only), more hypertension (74% vs 48%), a greater proportion with raised triglyceride levels (29% vs 21%) and more central obesity. Conversely, those diagnosed by HbA_1c only had a much higher proportion of smokers (64% vs 32%). So where there were differences in risk factors, in most cases they were worse in those diagnosed by OGTT only, with smoking being the exception.

Receiver operating characteristic analysis indicates that HbA_1c testing performs better than FPG and 2-hour PG tests at discriminating between individuals at high and low risk of developing IHD over a 10-year period using the PRECARD tool (Centre for Preventive Medicine, Medical Department M, Glostrup University Hospital, Glostrup, Denmark), particularly at specificity of between 50% and 90%.

Smoking has been linked with diabetes before Sargeant et al. (2001)105 from the EPIC-Norfolk study reported that mean HbA_1c level was higher in smokers and hypothesised that smoking affected insulin sensitivity. One question is whether or not smoking affects HbA_1c level rather than diabetes.

The ADDITION study81 compared blood tests and cardiovascular risk profile in 8696 individuals. As in the Danish Inter99 Study,83 individuals identified as diabetic by an OGTT, but not HbA_1c testing, had significantly higher SBP, mean triglyceride level, total cholesterol level and mean Framingham 10-year cardiovascular risk profile than those identified as diabetic by HbA_1c testing but not OGTT (Table 11). A screening programme first using HbA_1c testing and subsequent oral glucose tolerance testing only in screen-positives may miss these high-risk individuals. There is clearly something about those identified by the OGTT that increases risk, possibly because the 2-hour PG correlates with insulin resistance.

The MET SIM (METabolic Syndrome In Men) study of 5836 Finnish men106 also found the least favourable lipid profiles in men with diabetes as defined by WHO OGTT 1999 criteria and HbA_1c level of > 6.5%, with mean triglycerides (2.03 mmol/l) and high-density lipoprotein (HDL) cholesterol (1.21 mmol/l). Diabetes defined by HbA_1c testing only was found to be associated with a trend towards lower triglycerides (1.55 mmol/l) but also lower HDL cholesterol (1.30 mmol/l) than diabetes, defined by WHO 1999 criteria triglycerides (1.67 mmol/l) and HDL cholesterol (1.37 mmol/l).

It appears that there may be a hierarchy of CVD risk prediction: OGTT stronger than HbA_1c testing and HbA_1c testing stronger than FPG testing. In the absence of details, this implies that OGTT positivity is driven by elevated 2-hour levels.

In a cross-sectional study of 1219 non-diabetic individuals,107 carotid intima media thickness was found to be correlated with HbA_1c level and 1-hour glucose level but not with fasting glucose or 2-hour glucose, for the subgroup with normal PG. When adjusted for confounders of age, sex, waist, SBP, HDL and total cholesterol, only HbA_1c testing was correlated to intima media thickness. Similarly analysis with all 1219 individuals (558 with NGT, 409 with IFG, 78 with IGT, and 174 with IFG and IGT) found that only HbA_1c testing was significantly correlated to intima media thickness after adjustment for confounders. Hence, the predictive power of the 2-hour PG may be due to correlations with other aspects of the metabolic syndrome, such as hypertension and lipids, i.e. it is a correlate not a cause.

In the Rancho Bernardo study,34 the risk of cardiovascular mortality was increased in women with isolated postchallenge (2-hour) hyperglycaemia (age-adjusted HR 2.6; 95% CI 1.5 to 4.8) but not in men (age-adjusted HR 0.7; 95% CI 0.3 to 1.6). 34

The point estimates for Edinburgh Artery Study32 participants were similar with ORs for cardiovascular mortality of 2.7 (95% CI 0.6 to 11.6) for women and 0.8 (95% CI 0.09 to 6.7) for men.

However, a Paris study34 found that the heart disease mortality rate in men with normal fasting glucose, but IGT was three times that of those with NGT.

Ethnicity: Afro-Caribbean

A joint analysis of data from the Screening for Impaired Glucose Tolerance (SIGT) study and Third National Health and Nutrition Examination Survey (NHANES III)108 found that black people have higher HbA_1c levels than white people, even after adjustment for covariates, including age, sex, BMI, BP, FPG and 2-hour PG. The increase in HbA_1c level for black people with NGT was 0.13% and 0.21% (SIGT and NHANES, respectively), increasing to 0.26% and 0.30% for those with pre-diabetes, and 0.47% and 0.47% in the diabetic range. Furthermore,108 the HbA_1c test identifies proportionally more black people than white people as both diabetic and pre-diabetic than the OGTT, using both International Expert Committee (IEC)70 and ADA criteria.

Similarly, in the ARIC study,71 black people accounted for just 15.5% of study participants with HbA_1c levels of < 5%, 11.9% of participants with HbA_1c levels of 5.0–5.5%, rising to 27.0% of those with HbA_1c levels of 5.5–6.0%, 49.1% of those with HbA_1c 6.0–6.5%, and the majority of participants (52.2%) with HbA_1c levels of ≥ 6.5% were black. Ethnicity for all participants in this study was either black or white. There was no significant interaction between race and HbA_1c value regarding the risk of CHD, ischaemic stroke or death from any cause (p > 0.80), suggesting that the raised HbA_1c levels in the black population do not represent a higher risk in that population and, therefore, there need be no adjustment in recall thresholds for black people based on cardiovascular risk. There was an interaction between race and risk of diagnosed diabetes after 15 years (p = 0.007), based on data from follow-up telephone calls. There was no interaction between race and risk of diabetes at 6 years (p = 0.8) which was based on data from visits. Therefore, HbA_1c score may be associated with different risks of diabetes according to race, or this effect may simply be driven by ethnic differences in reporting and health-care-seeking behaviour affecting rates of self-reported diabetes. For self-reported diabetes at 15 years of follow-up there was also a significant interaction between race and FPG level (p = 0.01). The NHANES99 in the USA also found that a greater proportion of black people than white people had a FPG level of < 5.5 mmol/l alongside higher HbA_1c levels.

The UKPDS study109 found that in newly diagnosed people with diabetes between 1977 and 1991, the HR (adjusted for confounders) associated with Afro-Caribbean ethnicity for myocardial infarction (MI) was 0.3 (95% CI 0.2 to 0.6). Diagnosis of diabetes was by two FPG measurements of > 6 mmol/l, which is lower than currently accepted criteria.

The IRAS study84 found that African Americans made up 39% of those with HbA_1c levels of ≥ 6.5% and only 24% of those with HbA_1c levels of < 5.7%. In contrast, white Americans made up 25% of those with HbA_1c levels of ≥ 6.5% and 45% of those with HbA_1c levels of < 5.7% (the remainder were Hispanics). Taking all of those at increased risk of diabetes (IGT, IFG or HbA_1c levels between 5.7% and 6.4%) as the gold standard, the sensitivity of HbA_1c (5.7–6.4%) was higher in African Americans (31%) than White Americans (10%). The reverse was true for IGT (45% and 60% in African Americans and white Americans, respectively). IFG performance did not differ by ethnicity.

The Health Aging and Body Composition study110 of 70- to 79-year-olds in the USA found similar trends in HbA_1c levels, detecting a greater proportion of black than white people. A HbA_1c score of > 6.5% was observed in 5.7% of black and 1.8% of white participants, and a score of 5.7–6.5% was found in 37% of black people in comparison with 15% of white people; 3.5% of black people and 2.4% of white people had FPG levels in excess of 126 mg/dl, and 19% of black people and 22% of white people had FPG levels of between 100 and 125 mg/dl.

Ethnicity: South Asian

The term ‘South Asian’ is used to refer to people with ancestry in India, Pakistan, Nepal, Sri Lanka and Bangladesh, even if that ancestry is quite distant. For example, many South Asians in the UK moved here from Uganda but had ancestors in places such as Gujarat. In studies from the USA, ‘Asian’ often refers to people of Chinese or Korean ancestry.

The LEADER (Leicester Ethnic Atherosclerosis and Diabetes Risk) study81 found that using the HbA_1c cut-off level of 6.5% rather than the 1999 WHO criteria resulted in a 1.4- and 2.2-fold increase in detection rates for white Europeans and South Asians, respectively, indicating that the HbA_1c test may be relatively more sensitive in the South Asian population.

A small cross-sectional study111 in 139 subjects – 26% South Asian – found that South Asian origin was associated with higher HbA_1c levels than in white subjects (6.1% vs 5.9%, p = 0.02), even although the South Asians were younger and weighed less. There were no significant differences in FPG and 2-hour PG levels between the two groups.

Similarly, in 3819 US participants with IGT,112 after correcting for confounders, ethnicity affected HbA_1c levels, with means of 5.8% for white people, 6.0% for Asian people and 6.2% for black people.

In the INTERHEART (A Study Of Risk Factors For First Myocardial Infarction In 52 Countries And Over 27,000 Subjects) study,113 the increase in OR of having an MI for every 1% increase in HbA_1c level differed significantly by self-identified ethnicity (p < 0.0001) and region (p < 0.0001). A 1% increase in HbA_1c level corresponded with an OR increase for MI, which was higher in western Europe (OR 1.85, 95% CI 1.52 to 2.26) in comparison with China (OR 1.22, 95% CI 1.14 to 1.30), suggesting that the predictive value for MI is greater in western Europe. Note, however, that T2DM in China is less associated with overweight than in Europeans, having more of an insulin deficiency pattern than an insulin resistance one.

Similarly, a 1% increase in HbA_1c level had differing effects on the odds of MI for different ethnicities, the greatest effect was in the ‘other Asian’ subjects (OR 1.80 95%, CI 1.54 to 2.09) alongside European (OR 1.50 95%, CI 1.37 to 1.66), South Asian (OR 1.47 95%, CI 1.34 to 1.62) and Arab subjects (OR 1.54, 95% CI 1.41 to 1.67) in comparison with Chinese (OR 1.22, 95% CI 1.14 to 1.30) and Latin American subjects (OR 1.28, 95% CI 1.17 to 1.41). The effect of a 1% increase in HbA_1c level on the odds of MI in black subjects is unclear, with those classified as black African (OR 1.44, 95% CI 1.12 to1.84) being higher than for coloured African subjects (OR 0.86, 95% CI 0.61 to 1.22) but with very large overlapping CIs.

In the ADDITION study,114 South Asians with T2DM (FPG of ≥ 7.0 mmol/l or 2-hour PG of ≥ 11.1 mmol/l) were at higher estimated risk of CVD over 10 years using the ETHRISK score* in comparison with Western Europeans [risk score = 27.7% (95% CI 25.2 to 30.2); risk score = 18.3% (95% CI 16.5 to 20.1), respectively]. (*ETHRISK is a web-based calculator for assessing risk in ethnics groups in the UK. 115) A significantly higher ETHRISK score was also observed in those with IFG exclusively [FPG level of 6.1–6.9 mmol/l and (if measured) 2-hour PG of < 7.8 mmol/l or IGT exclusively (FPG of < 7.0 mmol/l and 2-hour PG of ≥ 7.8 and < 11.1 mmol/l). ETHRISK score assesses risk according to smoking, sex, age, SBP, HDL and total cholesterol levels, adjusted for ethnicity. South Asians were also more likely to test positive for IGT exclusively (OR 1.66, 95% CI 1.33 to 2.06) or T2DM (OR 2.18, 95% CI 1.56 to 3.06), but not IFG exclusively (OR 1.03, 95% CI 0.67 to 1.60), suggesting that the FPG test at this threshold may be a less sensitive test in the South Asian population with regard to predicting risk of CVD. Changing the criteria for detecting diabetes from FPG level of ≥ 7.0 mmol/l or 2-hour PG of ≥ 11.1 mmol/l to HbA_1c level of ≥ 6.5% would double the prevalence of diabetes in this data set and the increase would be greater in South Asians than in white Europeans (2.1-fold vs 1.4-fold increase, respectively), suggesting that HbA_1c testing may be more sensitive than the OGTT in the South Asian population. This result is similar to the LEADER study. 81

The reasons for the much higher prevalence of T2DM are complex and have been reviewed recently by Bhopal. 116 He hypothesises a four-stage model starting at birth, with small, relatively fat babies with low lean body mass and fewer beta cells, continuing through childhood and early adult life with energy intake excessive compared with low physical activity levels, followed by the development of insulin resistance and fatty livers, and ending in beta cell failure.

Others have noted the difference between BMI and risk in South Asians, with ‘overweight’ being defined as BMI of > 23 kg/m² and obesity as BMI of > 25 kg/m², and the need for lifestyle intervention across the entire life course. For a review, see Szczepura (2011). 117

How often should screening be done?

There is a shortage of evidence to answer this question. Takahashi et al. (2010)118 carried out annual OGTTs in Japan and found that the cumulative incidence of diabetes after 3 years by band of HbA_1c was:

• Baseline < 5.0% CI 0.05%

• 5.0–5.4% CI 0.05%

• 5.5–5.9% CI 1.2%

• 5.5–5.9% CI 1.2%

• 6.0–6.4% CI 20%.

This not only suggests that the screening interval should not be < 3 years for those with initial HbA_1c level of < 6.0%, but also supports the case for using the cut-off level of 6.0% HbA_1c.

Davies and Day (1994)119 suggested a repeat interval of 30 months, based on postprandial glycosuria screening.

Background and aims

The aims of the ADDITION study were to assess whether screening for undiagnosed T2DM is feasible, and to determine whether intensive treatment of people with screen-detected T2DM can reduce cardiovascular mortality and prevent or delay the macrovascular and microvascular complications of diabetes.

The trial was carried out in Denmark, the UK (Cambridge/East Anglia and Leicester) and the Netherlands. It was conducted in two phases: screening and intervention.

Design

Screening study

The first phase screened people for diabetes, using a variety of different approaches in different centres. (Note: the trial was not set up to compare the effectiveness of different approaches.)

The screening methods were reported in the paper by van Donk et al. (2011)122 Screening was conducted in four steps:

1. identification of those at high risk using risk scores

2. capillary random blood glucose (RBG) testing with or without HbA_1c

3. capillary FBG testing

4. OGTT.

Not all study centres followed these four steps (Figure 2). In the Leicester centre, the screening moved directly to step 4 from step 1. One centre in the Netherlands and those using one of the two types of opportunistic strategies in Denmark skipped step 2.

FIGURE 2.

Screening stages used in the ADDITION study.

Results of screening study

The results of the screening phase of the ADDITION trial were reported by van den Donk et al. (2011). 122

Results of intervention study

Overall, 3233 individuals had screen-detected diabetes by the end of the screening phase of the ADDITION study,123,124 and 3057 individuals (male 1771, female 1286) were recruited to the second phase (the intervention study) of the study: 1533 in Denmark, 867 in Cambridge, 498 in Netherlands and 159 in Leicester. 123 The patients were followed up for a mean period of 5.3 years. 123

The baseline characteristics of the individuals participating in this phase of the study are given in Table 12. 124

The baseline characteristics differed significantly among centres. 124 In the Leicester centre, patients were younger than in other centres (mean age of 57.2 years vs 59.9 years in other centres, p < 0.001). This centre also had a larger proportion of non-white patients (41.3% vs 3.9% in other centres, p < 0.001) and approximately 8% of the patients were unemployed (2.2% in other centres, p < 0.001). The proportion of patients smoking cigarettes was highest in Denmark (35%), followed by the Netherlands (26%), Cambridge (18%) and Leicester (16%). Similarly, the number of patients drinking a large amount of alcohol per week was higher in Denmark than in other countries. However, they had the lowest mean HbA_1c level compared with patients from other centres (6.8% vs 7.3%, p < 0.001). 124

Of the 73% patients who had BP of > 140/90 mmHg at baseline, 58% had not been prescribed any antihypertensive drug. The mean SBP and diastolic blood pressure (DBP) in these patients was 152 mmHg (SD 23 mmHg) and 88 mmHg (SD 12 mmHg), respectively. In those patients already on antihypertensive agents, the mean SBP was 151 mmHg (SD 22 mmHg) and the DBP was 86 mmHg (SD 12 mmHg). 124

Approximately 86% of patients were not on lipid-lowering drugs and in these patients the mean total cholesterol level was 5.6 mmol/l (SD 1.1 mmol/l). In those patients receiving lipid-lowering drugs, the mean total cholesterol level was 4.9 mmol/l (SD 1.0 mmol/l). In the whole ADDITION trial, approximately 64% of patients had a total cholesterol level of > 5 mmol/l. 124

Although the practices were randomised to two groups, some imbalances in allocation of participants in Denmark were seen. More patients were assigned to intensive treatment than to routine care (n = 910 vs n = 623). The group assigned to intensive therapy had by chance more IHD. The authors suggested that the imbalance between the two groups could have occurred because of awareness of the staff in the intervention arm about the potential benefits of early detection and treatment, in those at high risk of CVD, might have led to the staff testing only those at risk. The authors noted that, at diagnosis, patients overall showed high levels of untreated cardiovascular risk factors. 123

Data on the primary composite outcome were available for 3055 out of 3057 patients (99.9%) at 5 years. The primary end point included cardiovascular death, MI, stroke, revascularisation and amputation. The proportions of patients suffering from first cardiovascular events were similar – approximately 8.5% (n = 117) and 7.2% (n = 121) patients in the routine care and intensive treatment, respectively. Although there was a reduction in the occurrence of first cardiovascular event by 17% in the intensive treatment group (HR 0.83, 95% CI 0.65 to 1.05), the difference between the two groups was not significant (p = 0.12). Out of the 117 patients in the routine care arm in whom first cardiovascular events occurred, 32 had a MI, 19 had a stroke, 22 died and 44 patients had revascularisation. In the intensive treatment group, 26 died due to CVD, 29 had a non-fatal MI, 22 had a stroke and 44 had revascularisation. In the intensive treatment group, there was reduction of cardiovascular death, non-fatal MIs and revascularisation by 12% (HR 0.88, 95% CI 0.51 to 1.51), 30% (HR 0.70, 95% CI 0.41 to 1.21) and 21% (HR 0.79, 95% CI 0.53 to 1.18), respectively. However, there was no difference between the two groups for stroke (HR 0.98, 95% CI 0.57 to 1.71) and all-cause mortality (HR 0.91, 95% CI 0.69 to 1.21). None of the patients had amputation as first cardiovascular event. 123

There were 196 deaths in total: 92 in the routine care and 104 in the intensive treatment group. Sixty deaths were due to cardiovascular events, 97 due to cancer and 39 to other causes. 123

After 5 years, 84% of the 2859 patients still alive returned for follow-up assessment. 123 The clinical and biochemical findings of another 328 patients were collected from the general practices. No information could be found on remaining 131 patients. When the patients who had no information were compared with patients that had follow-up, the authors found that the patients with no information were more likely to come from an ethnic minority group, and their baseline, total and LDL cholesterols were high. 123

The changes seen during mean follow-up of 5.3 years are given in Table 13. 123

There were similar changes in SBP, DBP, and in total and LDL cholesterol levels. Use of glucose-lowering drugs increased in both groups, but more in the intensive treatment group. More patients in the intensive treatment group received aspirin, angiotensin-converting enzyme (ACE) inhibitors and lipid-lowering drugs. The proportion of patients smoking decreased by the end of the study. 123

The proportion of patients achieving the target HbA_1c level of < 7%, BP of < 135/85 mmHg, and cholesterol of < 5 mmol/l (if no CVD) or < 4.5 mmol/l (if CVD) was higher in the intensive treatment than in the routine care group (Table 14: read from the published figure no. 4). 123

The proportions of patients reporting hypoglycaemia as assessed by the diabetes treatment satisfaction questionnaire were similar between the groups, although the proportions are not given (χ² = 4.44, p = 0.62). 123

The 10-year results were disappointing. 102 After a median follow-up of 9.6 years [interquartile range (IQR) 8.9 to 9.9 years; 184,057 person-years], there was no significant difference in all-cause mortality between the two groups (HR 1.06, 95% CI 0.90 to 1.25; p = 0.46). Cancer was the commonest cause of death. Intervention caused no significant reduction in cardiovascular mortality (HR 1.02, 95% CI 0.75 to 1.38) or diabetes-related mortality (HR 1.26, 95% CI 0.75 to 2.10). It should be noted that in the 10-year results, only 466 patients from 33 practices were included and, of these, five practices were randomised to no screening.

Barakat et al. (2013)125 reported that recruits to ADDITION in the Cambridge centre, made little change in their physical activity. This paper was based on 867 participants from 49 general practices. It is not clear whether or not the results are from both intervention and control arms, but the numbers suggest both. The advice given was to undertake 35 minutes of brisk walking a day. There was some variation among participants, but even those who reported increased recreational activity had little change in HbA_1c: reductions of 0.093% in men and 0.022% in women. The reduction in men was statistically significant (p = 0.02) but of doubtful clinical significance. So even volunteers in a trial, who were told that they have newly diagnosed T2DM, made little change in physical activity.

Clinical effectiveness

The review of clinical effectiveness was based primarily on RCTs. Nine RCTs comparing lifestyle interventions (predominantly diet and physical activity advice, with regular reinforcement and frequent follow-up) with standard lifestyle advice or placebo were identified. They included 5875 people randomised to receive lifestyle advice, exercise programmes or combinations thereof. The trials varied in design and quality. The primary outcome for the trials was progression to T2DM.

Participants had IGT. In most of the trials, lifestyle interventions reduced progression to diabetes (RR range 0.33 to 0.96).

The Diabetes Prevention Program (DPP) from North America (which had higher risk recruits than most other trials) reported that the prevalence of diabetes at 3 years was 29% in the control group compared with 14% in the lifestyle intervention arm. 128

The DPS15 had the longest follow-up, to 7 years, which included the 4 years of intervention and then 3 years of postintervention follow-up. After 4 years, 4% of the lifestyle group and 7.4% of the control group had developed diabetes, roughly a halving of risk. At 7 years, the difference had diminished slightly but the intervention group retained most of the benefit, suggesting that 4 years of the lifestyle intervention had resulted in a sustained change in lifestyle habits. 16

The benefits of the lifestyle intervention were greatest in those with the highest compliance and who achieved more of the targets (such as weight loss and dietary change). For example, in the Finnish study, those who achieved four or five of the five targets had a risk of developing diabetes which was only 23% of those who achieved none. However, even among the volunteers in the trials, many did not succeed, and others succeeded in the short term (such as the first 6 months) but not in the longer term. The key to success is sustained lifestyle change, especially weight loss.

The DPS involved quite intensive lifestyle intervention. A subsequent study from Finland used low-intensity intervention (six sessions of counselling by public health nurses) over an 8-week period, and found that the effects persisted for 3 years. 129 However, the effects were much less than in DPS: 0.8 kg weight loss compared with 4.5 kg in the DPS. The most successful prevention studies had more intensive interventions with more frequent contacts. 130

Perry et al. (2005)131 from Cork in Ireland identified factors from previous studies, which protected against diabetes: BMI < 25 kg/m²; waist–hip ratio of < 0.85 for women and 0.90 for men; never smoking; medium- to high-level physical activity; light drinking (three to five, to seven units per week); and a prudent diet. In their sample of middle-aged Irish men and women drawn from general practice populations, 7.5% had none of these protective factors. Insulin resistance was calculated using the homeostasis model analysis (HOMA) score (based on fasting levels of both insulin and glucose). Taking the 7.5% with no protective factors as the reference group, multivariate analysis gave ORs for insulin resistance of 0.59 with one protective factor, 0.48 with two, 0.14 with three, and 0.04 with four or more. About 13% had four or more protective factors.

Hence, there is little doubt that lifestyle measures could prevent most cases of T2DM. Weight loss would also benefit those who already have diabetes, or hypertension, and improvements can follow even modest weight loss. Goldstein reviewed studies in which large and small amounts of weight were lost, and concluded that even modest weight reductions of 10% or less, resulted in significant benefit in a substantial subset. 132 Even loss of a few kilograms can provide benefit.

One of the key factors in cost-effectiveness analysis is adherence to lifestyle changes. Even among the volunteers in the trials, a large proportion did not adhere. It was also noticeable that in many trials, initial gains were lost after the intervention ceased.

A more radical option, bariatric surgery for obesity, reduces progression from IGT to diabetes (and also reverses diabetes in many cases).

Cost-effectiveness

The modelling in the HTA review examined the cost-effectiveness of preventing or delaying T2DM in people with IGT, including the effects of interventions on CVD.

Modelling based on data from the trials may not reflect what would happen in routine care. Trials are protocol driven and patients are supposed to stay on the treatments to which they are randomised. In normal care, if an intervention is not working, it should be stopped. The modelling assumed that people with IGT would initially be treated with a structured lifestyle intervention similar to that in the Finnish trial, but that those who did not comply would be switched to metformin after 12 months. Metformin is now a very cheap drug, and reduces the risk of progression to diabetes, although not by as much as adherence to lifestyle measures does. Applying an early switch to metformin in the non-adherers means that the adherers remaining on diet and physical activity will do better than seen in the lifestyle arms of the trials. It was assumed that the non-adherers to lifestyle modifications will have better adherence to metformin, so that they will also do better than if left on the lifestyle interventions.

Using the switching assumption, intervention is highly cost-effective and, in certain scenarios, cost-saving.

The main conclusion of the review was that in people with IGT, lifestyle change (diet and physical activity) is clinically effective and cost-effective in reducing progression to diabetes.

Study selection

The databases searched were MEDLINE, EMBASE, The NHS Economic Evolution Database (NHS EED), HTA, Cost-effectiveness Analysis (CEA) Registry, Web of Science, EASD and ADA for studies published since 2008. Titles and abstracts were reviewed and if considered potentially relevant; full papers were reviewed. The sifting was carried out independently by two reviewers. Any discrepancies were resolved by discussion. The flow chart (Figure 3) illustrates the search results.

Inclusion criteria Studies were included if they assessed the costs and/or the outcomes (long or short term) of screening strategies for either undiagnosed T2DM or IGT/IFG.

FIGURE 3.

Flow chart of economic search and filtering.

Exclusion criteria – studies were excluded if they:

• did not assess the costs and or the outcomes of screening strategies for either undiagnosed diabetes or IGT/IFG

• reported in languages other than English

• were assessing screening tests for complications of diabetes

• were assessing screening tests for gestational diabetes

• were targeting an incorrect age group of the population (either teenagers or very old individuals)

• were based on data recorded in non-westernised countries.

The costs [US dollars (US$), Australian dollars (AU$) and euros (€)] in the texts have been converted into UK pounds sterling (£) on the basis of purchasing power parity. 133

Screening for diabetes could produce benefit in two ways. First, identification of those with pre-diabetes and lifestyle intervention, could prevent or delay the onset of diabetes. Second, intervention in those found to have diabetes could reduce the morbidity and the costs owing to the complications, primarily cardiovascular, associated with diabetes.

It is known that diabetes reduces average survival, possibly by as much as 15 years. 134 However, the excess mortality among people with diabetes has been falling over recent decades135 and may be less in those diagnosed more recently.

Most previous studies have examined screening for diabetes alone, but incorporating it into a broader vascular risk assessment programme, as being introduced in England, would reduce costs. The issue then becomes whether adding a test of blood glucose to all or selected individuals is cost-effective.

Universal screening compared with selective screening

Screening for diabetes or for impaired glucose tolerance as well

Gillies et al. (2008)143 compared three different screening interventions: (1) screening for diabetes; (2) screening for diabetes and IGT followed by lifestyle interventions; and (3) screening for diabetes and IGT followed by pharmacological interventions against no screening.

The base-case scenario was one-off screening was at the age of 45 years, in people with above-average risk of diabetes. Risk was based on known history of IHD, hypertension, dyslipidaemia, family history of T2DM and a BMI of > 25 kg/m². They based their effectiveness estimates for the prevention of diabetes on a previous review of pharmacological and non-pharmacological interventions of their own. They assumed that lifestyle intervention would cost about £400 in the first year, and £280 per year thereafter. These costs were derived from a review of interventions in obesity rather than from the diabetes trials.

The results confirmed that all interventions are cost-effective compared with a no-screening strategy. The estimated costs per QALY were (1) £14,150; (2) £6242; and (3) £7023 for the aforementioned strategies. So screening for, and intervention in, both diabetes and IGT was more cost-effective than screening for diabetes alone.

A fuller account of this excellent study is included in Gillett et al. (2012). 19

In Germany, Schaufler et al. (2010)144 also concluded that screening for both diabetes and IGT was worthwhile. They examined the cost-effectiveness of two screening and intervention strategies, for T2DM and pre-diabetes, and with lifestyle and metformin interventions. The target population covered individuals aged 35–75 years. Screening was yearly with the OGTT.

A Markov Monte Carlo/TreeAge Pro Suite 2007 (TreeAge Software, Inc., Williamstown, MA, USA) microsimulation model was used. Sensitivity analysis was also carried out to assess the robustness of the results. The outcomes reported included quality of life, lifetime costs, incidence and age at occurrence of diabetes-related complications and age at diabetes diagnosis. National databases were used for retrieving the population data. A lifelong horizon period was used for the modelling purpose. Other information pertaining to modelling was collected from relevant articles and previous studies. This evaluation was carried out from the perspective of German Statutory Health Insurance's (SHI) perspectives. Sensitivity analysis was also carried out for validating the different scenarios, keeping in mind the private health-care infrastructure of the German settings.

Screening for diabetes allowed diagnosis 3.8 years earlier and a gain of 0.8 life-years. Intervention proved cost-effective and reduced the average cost of illness by £605 (€807) per detected T2DM case in the screening programme when compared with no screening at all. Intervening in pre-diabetic individuals gave average savings per detected case of £4861 (€6481). Costs per QALY showed that screening was cost-effective for T2DM (€563 per QALY with lifestyle) and could even be cost-saving in pre-diabetes.

Webb et al. (2011)114 suggested that there is a CVD risk reduction of 6% and 13% in diabetic and IGR screen-detected patients, respectively.

Should the approach to screening vary with ethnicity?

The question here is whether people belonging to South Asian ethnicity (India, Pakistan, Nepal, Bangladesh and Sri Lanka) be screened at younger ages or at lower BMIs.

The European guideline by Paulweber et al. (2010)43 suggests that prevalence rates vary two- to fivefold among people from different ethnicities. This difference is explained by the different lifestyle or disposition to a particular disease owing to genetic reasons.

The STAR (Screening Those At Risk) study145 reports that the rate of progression from IGT to diabetes is higher for South Asians living in the UK. 114 It suggests that screening for diabetes in South Asian people in the UK should use different threshold levels. Partly due to their sedentary lifestyle, many people from South Asian backgrounds have high BMI at a very young age. And, as reported earlier, at a given HbA_1c, South Asians had higher BP and triglyceride levels, and higher incidence of CVD.

Davis et al. (2008)146 report that the Afro-Caribbean population also has high prevalence of T2DM.

There is, therefore, a case for starting screening from an earlier age in some ethnic groups.

Which test of blood glucose should be used?

Icks et al. (2004)147 compared the effectiveness for the combinations of either of HbA_1c testing, fasting glucose testing or the OGTT. The most effective was the combination of the HbA_1c testing followed by the OGTT, with a very high detection rate (54%) and cost per detected case, ranging from £16 to £26 (€21–34.5).

Chatterjee et al. (2010)96 used data from the SIGT study, in which 1259 adults without known diabetes had all of random PG, 50-g GCT, HbA_1c testing and the OGTT. They modelled costs of screening and subsequent treatment of diabetes and IGT, and the costs of false-negatives and true-positives. Their focus was on screening for both pre-diabetes and undiagnosed diabetes; results are not given separately, and are for only a 3-year period. They provide costs but not cost-effectiveness data. Taking all costs into account, they conclude that the 50-g GCT provides the best strategy in terms of total health system costs.

Quality assessment

Table 15 shows the description of the various cost-effective studies that were included. The assessment of the selected studies was carried out using the British Medical Journal checklist modified form of the Drummond and Jefferson (1996)148 guidelines that are used for the evaluation of the economic studies. The studies have been assessed on three main aspects, namely study design, data collection (Table 16), and analysis and interpretation of results (Table 17).

All studies met most of the quality criteria. Most studies fulfilled the study design criteria. All the studies are well structured and address clear questions. The research question here relates to the feasibility of screening for T2DM. The studies differed from each other in terms of the screening interventions compared, lifetime horizon, evaluation perspective, country settings, target population and incremental cost-effectiveness ratios (ICERs). Only two studies138,139 had clear justification for the economic approach used.

Most of the studies followed systematic method of data collection, each study providing the source of effectiveness estimates and different primary outcome measure for the economic evaluations. However, these studies also differed in the outcomes used for quality evaluation. Some studies143,144 used QALY's as their primary outcomes, others47 used count of event incidents as primary outcomes. Two Australian studies139,149 used cost-per-DALY as outcome for comparison. All the studies have given adequate detail of design.

Costs have been discounted for all. Sensitivity analysis is carried out in all the studies by varying parameters for the validation of robustness of the economic model.

The study population remains a concern for few studies. The study of Hoerger et al. (2007)150 concerning cost-effectiveness of selective screening only looked at individuals with hypertension. Furthermore, the costs associated with selecting the hypertensive individuals were not accounted for. Similarly, the Gillies et al. (2008)143 analysis included only individuals with IGT.

A moderate participation rate was assumed in the Schaufler and Wolff study,144 although various participation rates were tested in the sensitivity analysis to verify the robustness of the model. The sensitivity analysis indicates that higher participation rates would lead to more cost-effective results. It is unrealistic to expect very high participation of individuals for screening when confirmation is by the OGTT. This study used a health-care insurance perspective.

A limitation of the Kahn study47 for our purposes is the way in which risk is assessed. First, there is no preliminary selection by risk factors. Second, the risk assessment is carried out using the FPG, which, as noted, has poor sensitivity.

Few studies included ethnicity. Gillies et al. (2008)143 included a sensitivity analysis. Not all of the studies considered all of the relevant and health outcomes. The Gillies et al. (2008)143 study did not consider both microvascular and macrovascular outcomes, which could actually show better results in terms of cost-effectiveness.

Does screening for type 2 diabetes and impaired glucose tolerance meet the criteria of the National Screening Committee?

The UK NSC criteria for evaluating screening programmes were adapted from the WHO criteria published in 1966. The criteria are published by the NSC on their website (www.screening.nhs.uk/).

The HTA report on screening considered the case for it against the NSC criteria. Most criteria were met, but not all.

National Institute for Health and Care Excellence guidance

Guidance issued by NICE on risk identification and interventions to prevent T2DM in adults at high risk, summarised in the British Medical Journal of 28 July 2012. 151

It should be noted that the introductory statement on the NICE website said:

The guidance is not advocating a national screening programme for type 2 diabetes.

Much of the guidance was about how to prevent T2DM, but there were some recommendations on screening. A two-stage approach was recommended, as in this report. The first stage would use either a self-administered questionnaire or a computerised system based on data in GP records to identify those at high risk. No particular tools were recommended.

The second stage was a measure of blood glucose, using either HbA_1c or FPG testing, with cut-offs of 6.5% and 7.0 mmol/l, respectively. Abnormal levels should be confirmed by FPG testing, HbA_1c testing or the OGTT.

Those with FPG levels of 5.5–6.9 mmol/l or HbA_1c 42–47 mmol/mol (6.0–6.4%) should be offered an intensive lifestyle change programme, but no repeat blood glucose test is suggested at this stage. Repeat blood testing should be offered a year later.

It is recommended by NICE that those at high risk, but with HbA_1c level of < 6.0%, should have risk assessed every 3 years.

One concern with this approach is that, although those with HbA_1c levels of 6.0–6.4% are at higher risk, most of those who develop diabetes may come from the large proportion with HbA_1c levels of < 6.0%. In the EPIC-Norfolk study, two-thirds of those who developed diabetes had baseline HbA_1c levels of < 6.0%. 152

An alternative: opportunistic screening

The emphasis in this review has been on systematic population-based screening. Another option would be opportunistic screening in general practice. Such a system has been described by Pereira Gray et al. (2012)153 from Exeter. They report the results from 3 years of opportunistic screening of patients who attend for other reasons, but have one or more risk factors for diabetes, including CVD, hypertension, hyperlipidaemia, obesity, recurrent skin infections or a family history of diabetes. Screening was done using mainly FPG testing (two FPG values of ≥ 7.0 mmol/l) but some OGTTs were carried out, using a 2-hour threshold of ≥ 11.1 mmol/l for diagnosis. In a practice with around 6400 adults, 5720 screening tests were carried out over a 3-year period, of which 190 were OGTTs. A total of 2763 patients were screened at least once – 43% of the practice population. Of the 86 new cases of T2DM diagnosed during the period, 54 (63%) were detected by screening. There were 333 screening tests reported as abnormal but most (279) did not turn out to be in people with diabetes.

The cost per screen-detected case was £377 and 51 patients had to be screened to detect one case of diabetes.

The screening was undertaken as patients attended for other purposes. The authors do not report how many patients with one or more risk factors did not attend during the period, or how many attended but were not screened. It would be useful if they would check the practice computer records for all people who would be invited for systematic screening to see how many were not screened opportunistically. It would also be useful to retest a sample of the FPG-negatives, to assess sensitivity.

Bertram et al. (2010)149 produced a cost-effectiveness model based on opportunistic screening of patients attending Australian general practices for other reasons, using as selection criteria age > 55 years or age > 45 years plus one or more of high BMI, hypertension, and family history of T2DM; ethnicity; previous gestational diabetes. They concluded that a combined package of opportunistic screening and lifestyle intervention would be cost-effective.

Another option?

Maynard et al. (2007)154 reported an entirely different method of screening – spectroscopic measurement of advanced glycation end-products (AGEs) in the skin using a fluorescent technique. The authors report that within a minute, skin AGEs (SAGEs) can be measured. Fasting is not required.

They studied 351 subjects comparing FPG, HbA_1c and OGTT data with SAGE results. The comparison was mainly with FPG and HbA_1c testing, with OGTT as the reference standard.

The OGTT found IGT in 55 subjects and T2DM in 29. Sensitivities are reported as 75% for SAGE, 58% for FPG at a threshold of 100 mg/dl and 64% for HbA_1c (at 5.8%); these thresholds being chosen to all provide 77% specificity. In ROC analysis, SAGEs had an AUC of 79.7% compared with HbA_1c 79.2% and FPG 72%.

This technology does not appear to have come into use. A literature search (September 2012) found that the original paper had been cited 18 times, mainly in reviews, with only a few studies producing new data. It does sound as if further research would be worthwhile.

Other views and contributions to the debate about screening

Those in favour of screening for T2DM point to rising prevalence, the proportion undiagnosed, the significant presence of complications at diagnosis and the range of effective treatments for T2DM. 155 They also point to the effective and cost-effective methods of reducing or delaying progression to diabetes from IGT.

Those against note that the additional information provided by knowledge of glucose status may not affect advice or management:

Changes to diet or physical activity levels will always be advisable for people who are overweight or sedentary whatever their overall diabetes or cardiovascular risk score and whatever their glucose result. 156

They also note that changes in lifestyle and overweight render most of the population at risk and, therefore, argue that a population-level approach is required, than that targeting individuals. 156

The 2008 statement from the US Preventive Services Task Force recommended screening for diabetes in people with sustained BP of ≥ 135/80 mmHg, but made no recommendation for other people on the basis of lack of evidence.

Table 18 gives details of a few guidelines, including those from NICE, and shows a fair amount of variation around the world.

Recent evidence from the NHS Health Check programme

A study from Birmingham by Smith et al. (2013)162 has examined the performance of the diabetes filter in the Health Check programme. The aim of the filter is to identify those people in whom a measure of PG is recommended. The filter is based on ethnicity, BMI and BP. The measure of PG can be FPG or HbA_1c. The Heart of Birmingham PCT has a high-risk population for T2DM, largely because of the ethnic mix, and so it was decided to test all those attending with HBA_1c, without selection by the filter. The performance of the filter in identifying those at risk of diabetes could then be checked.

The results showed that the sensitivity of the filter was only 67%. One-third of those with diabetes or at high risk of it (as determined by HbA_1c of ≥ 42 mmol/mmol or greater) would have been missed. Conversely, of those deemed positive by the filter, over half would have been false-positives (positive predictive value 41%).

Further refinement of the filter seems to be indicated.

Research needs

These include:

• How to motivate people to adopt healthier lifestyles.

• Cost-effectiveness modelling of selection for screening at different risk scores.

• The use of the non-fasting 50-g GCT in screening.

• If screening were to be approved in principle, modelling of different screening scenarios.

• Determining the incidence of T2DM in 10-year age bands and, hence, the extent of earlier onset.

The highest priority is how to motivate people to adopt healthier lifestyles. The rising prevalence of T2DM could be largely prevented if people led healthier lifestyles and, in particular, avoided obesity. One of our referees suggested that public health interventions to reduce the incidence of T2DM should be compared with screening to detect it.

Further work is needed on modelling the cost-effectiveness of selection for screening by risk factors and scores, including refining the Health Check diabetes filter, or replacing it with QDiabetes Risk Score or FINDRISC.

The non-fasting 50-g GCT might be the best way of screening, but the evidence base is sparse. Trials are needed.

If screening were to be approved in principle, there should be modelling of different screening scenarios to how to implement screening and intervention in a way with which the NHS could cope. That might involve screening only those at highest risk in the first year or two of the programme, and then gradually extending it.

We need to monitor the age at onset of T2DM to determine the extent of earlier onset.

The age at which screening for diabetes and IGT should start in different ethnic groups needs further study. It may be worthwhile starting at a younger age in some populations.

Further research is needed into opportunistic screening in general practice, and the implications for population screening. In practices such as in Exeter, which have had systematic opportunistic screening, there may be little to gain from further screening. This needs investigation.

Conclusions

The case for population screening for T2DM remains unproven. It does not meet all of the NSC criteria.

Contributions of authors

Beth Hall undertook literature searches.

Sian Taylor-Phillips drafted Chapter 2, with help from Norman Waugh and Gaurav Suri.

Deepson Shyangdan wrote Chapter 4 on the ADDITION trial, checked and formatted the report, and created the bibliography.

Gaurav Suri drafted Chapter 6.

Norman Waugh wrote the remaining chapters, edited the whole document and is guarantor.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.

Previous review of screening for type 2 diabetes

Waugh N, Scotland G, McNamee P, Gillett M, Brennan A, Goyder E, et al. Screening for type 2 diabetes: literature review and economic modelling. Health Technol Assess 2007;11(17).

The objectives of this review were as follows:

• to consider the aims of screening for undiagnosed diabetes, and whether screening should be for other abnormalities of glucose metabolism, such as IGT, or the ‘metabolic syndrome’

• to update the previous review for the NSC on screening for diabetes, including reviewing choice of screening test

• to consider what measures would be taken if IGT and IFG were identified by screening and, in particular, to examine evidence on treatment to prevent progression to diabetes in these groups

• to examine the cost-effectiveness of screening, by a review of previous economic models, and by new modelling to take account of recent developments in treatment, such as the use of statins

• as part of the economic analysis, to consider groups at higher risk at which screening might be targeted.

Methods

The literature searches (carried out up to the end of June 2005) and review concentrated on evidence published since the last review of screening for NSC in 2001. Both reviews and primary studies were included. The 2001 review was:

Wareham NJ, Griffin SJ. Should we screen for type 2 diabetes? Evaluation against the National Screening Committee criteria. BMJ 2001;322:986–8.

Previous review of prevention of diabetes

Gillett M, Royle P, Snaith A, Scotland G, Poobalan A, Imamura M, et al. Non-pharmacological interventions to reduce the risk of diabetes in people with impaired glucose regulation; systematic review and economic evaluation. Health Technol Assess 2012;16(33).

Screening for type 2 diabetes: updating review for National Screening Committee

HTA project number 2011/18.

Project lead:

Professor Norman Waugh

Warwick Evidence

Health Sciences Research Institute

Gibbett Hill Campus

University of Warwick

Coventry CV4 7AL

Background

The prevalence of type 2 diabetes has been increasing, linked to trends in overweight and obesity, to other lifestyle changes such as reduced physical activity, and to an ageing population. About 4% of the UK population is known to have diabetes.

However, it is estimated that around another 0.5% or more have undiagnosed diabetes (YHPHO model). Type 2 diabetes (T2DM) can have an insidious onset with few or no symptoms. By the time of diagnosis, people with T2DM may already have damage to eyes (retinopathy) or to kidneys (nephropathy). More seriously, they may have accelerated arterial disease, most notably coronary artery disease. Diabetes increases the risk of arterial disease. They may have myocardial infarction (heart attacks) with no prior warning, and some of these will be fatal.

In addition to diabetes, there are two other conditions which are characterised by hyperglycaemia;

• Impaired glucose tolerance (IGT) wherein blood glucose levels are raised, though not to diabetic levels, after meals. IGT is usually diagnosed by means of a oral glucose tolerance test (OGTT)

• Impaired fasting glucose

Both may progress to diabetes. IGT is also important because it increases the risk of arterial disease.

There is therefore a case for screening for type 2 diabetes. The last HTA review of screening (Waugh et al.) noted that not all the NSC criteria were met, but that the case for screening was becoming stronger.

Current and planned screening services for diabetes

One effect of devolution is that the four UK territories have different policies for screening. Early in the review, we will contact each of the health departments to ascertain their current plans.

In England, screening for diabetes would be done, in those over 40, as part of the vascular risk programme. This will involve face to face contact. It appears that screening would be universal.

In Scotland, there are no plans to introduce a vascular risk screening programme. Instead, the Scottish Executive is considering using questionnaires on lifestyle and health, possibly via NHS 24. This provides an opportunity for selective screening based on questionnaire data, possibly akin to the Finnish system.

Wales – plans to be ascertained.

Northern Ireland – plans to be ascertained.

Previous reports

These include:

• HTA report on screening for type 2 diabetes, published 2007, written 2006.

That report noted that if screening for T2DM were to be introduced, there would, depending on test used and thresholds chosen, be more people detected with IGT or IFG than with diabetes. The HTA Programme therefore commissioned a complementary review of non-pharmacological prevention of type 2 diabetes in people with IGT or IFG. (In editorial process, due for submission of revised report in August 2011).

More recently, the SPHN has produced a report on screening for type 2 diabetes (not yet published but will be on SPHN website in due course).

An economic assessment of screening options by Gillies and colleagues was published in 2008 (BMJ). It concluded that screening for both diabetes and IGT would be cost-effective, but that “The cost-effectiveness of a policy of screening for diabetes alone, which offered no intervention for those with impaired glucose tolerance, is still uncertain.”

Current questions

The issues to be addressed in this review are:

• Should screening for type 2 diabetes be universal at age, say, 40 years and over, or should only people at higher risk be screened?

• If screening were to be selective, how should risk be assessed? There are various risk scores, all of which involve age and BMI. Some can be derived from data held on GP computer systems, but others require questionnaires to be sent or administered to patients. The question here is how much risk scores improve detection over a simple age and BMI approach.

• Should the approach to screening vary with ethnicity? The prevalence of T2DM is higher those with South Asian ethnicity (meaning India, Pakistan, Bangladesh, Sri Lanka, Nepal). Should they be screened at younger ages or at lower BMIs?

• Which test of blood glucose should be used? Glycated haemoglobin, or HbA_1c, is now being used for diagnosing diabetes. One issue here is whether uptake of screening varies amongst tests. The most sensitive test may not be the best option if uptake is low. Another important issue is that different screening tests detect over-lapping but different groups, with different risk of cardiovascular disease. Since the main aim of screening is to reduce CVD, the choice of test is important. It may be that detection of cardiovascular risk should take precedence over detection of diabetes.

• Should screening be just for diabetes, or for IGT as well?

Structure of report

Chapter 1 will review the background and summarise past reports.

Chapter 2 will review methods for selection based on risk scoring, including a systematic review of different scoring systems.

Chapter 3 will review the tests for blood glucose, including sensitivity and specificity. It will also review the evidence on acceptability and uptake.

Chapter 4 will consider the issues around vascular risk and their implications for the screening programme.

Chapter 5 will review recent studies of the economics of screening, and may include de novo modelling.

Chapter 6 will summarise the options for NSC, and will identify research needs.

Methods

The review will be done using the usual methods:

• Systematic searches of MEDLINE, EMBASE and all sections of the Cochrane Library for reviews and primary studies of both clinical and cost-effectiveness, and of current trial databases for ongoing research.

• Existing guidelines of screening from the UK and other relevant countries will be summarised. If recommendations differ, reasons for differences will be explored.

• Lists of retrieved abstracts will be checked by two reviewers with full papers obtained in cases in doubt.

• Studies will be quality assessed by appropriate methods as per CRD 4 or NICE.

• Data will be extracted by one researcher and checked by another using predefined data extraction forms.

• Data from screening studies will be used to produce sensitivity, specificity, PPV and NPV, if sufficient data are given to construct 2 × 2 tables. This will apply to the first stage of screening (risk scoring) and to the second stage (blood glucose).

• The case for screening will be assessed against the NSC criteria.

8th August 2011.