Waking up from the DREAM of preventing diabetes with drugs

The current epidemic of diabetes makes a drug to prevent it attractive. But despite promotion of recent research evidence, Victor Montori, William Isley, and Gordon Guyatt argue that we are not there yet

Diabetes affects about 4% of the world population and is associated with important costs, both in financial and human terms. The high prevalence, increasing incidence, and associated costs makes preventing diabetes a public health priority. The diabetes reduction assessment with ramipril and rosiglitazone medication (DREAM) trial recently showed that rosiglitazone reduced the risk of diabetes in people at risk. The results have prompted aggressive marketing of rosiglitazone as a preventive therapy; some clinicians are already responding to this initiative. We argue that the strategy will bring harms and additional costs while the benefits for patients remain questionable.

Preventing diabetes

Several randomised trials have shown that modest weight loss and physical activity can greatly reduce the risk of diabetes. The Diabetes Prevention Program documented a 58% relative risk reduction (confidence interval 48% to 66%) in high risk individuals; other trials have shown similar results.

Nevertheless, the possibility of preventing diabetes with drugs has caught the imagination of the drug industry. The medicalisation of pre-disease states and risk factors has become increasingly common, including targets of precursors of hypertension, osteopenia, and obesity. The prospect of marketing existing drugs to otherwise healthy people greatly expands the market for these drugs while increasing costs for society, increasing use of health care, and potentially reducing quality of life by converting healthy people into patients.

Effectiveness of drugs

Several trials have assessed the ability of drugs to prevent diabetes (box). Overall, except for metformin, the evidence is inconsistent and comes from trials of limited methodological quality. Two trials included drug discontinuation phases to determine if the drugs had changed the natural course of diabetes or was merely treating diabetes. Both discontinuation studies found that the proportion of diabetes diagnoses remained lower in the intervention arm; a third to half of the patients, however, were lost to follow-up and did not provide discontinuation data. Furthermore, the follow-up period after treatment was much shorter than the treatment time. None of the trials showed a reduction in the risk of diabetes complications.

Evidence for drug prevention of diabetes

Metformin

• Consistent evidence from 3 randomised trials
• The Diabetes Prevention Program (DPP) found metformin reduced the 3 year risk of diabetes (relative risk 0.69, 95% confidence interval 0.57 to 0.83), but lifestyle change was more effective

Troglitazone (no longer available)

Two trials found troglitazone was effective in preventing diabetes:

• Study in women with a history of gestational diabetes had large loss to follow-up11
• The DPP discontinued the trial arm because of fear of liver toxicity. Relative risk of diabetes diagnosis after 1 year of troglitazone was 0.25 (P<0.001), but the effect disappeared in the year after drug discontinuation12

Angiotensin converting enzyme inhibitors, angiotensin receptor blockers

• Systematic reviews of trials in hypertension, heart failure, and coronary disease that assessed diabetes as a secondary or post hoc outcome found large preventive effects13
• DREAM trial failed to confirm the effect

DREAM is a large randomised controlled trial that enrolled patients with impaired fasting glucose concentrations or impaired glucose tolerance and assigned them to high dose rosiglitazone or placebo.The trial effectively concealed allocation, adhered to the intention to treat principle, and achieved negligible loss to follow-up after a median follow-up of three years.

The trial's primary outcome was a composite end point of death and the diagnosis of diabetes. It was stopped early after almost 1000 primary end points had accumulated because of benefit in the treatment arm (table 1). The authors noted that for every 1000 people treated with rosiglitazone 8 mg/day for three years, about 144 people who would otherwise cross the glucose threshold we call diabetes will not do so; four to five patients without congestive heart failure will develop the condition.

The epidemic of pre-diabetes: the medicine and the politics

Aldous Huxley wrote that “Medical science has made such tremendous progress that there is hardly a healthy human left.” Changes to the American Diabetes Association (ADA) guidance on the diagnosis of pre-diabetes in 2010 make this statement even more true.1 If implemented globally the guidance could create a potential epidemic, with over half of Chinese adults,2 for example, having pre-diabetes, a national burden of around 493 million people.

Pre-diabetes is an umbrella term and the most widely used phrase to describe a blood concentration of glucose or glycated haemoglobin (HbA_1c) that lies above normal but below that defined for diabetes. We explore the evidence and value of pre-diabetes as a category or diagnosis (box 1) and argue that current definitions risk unnecessary medicalisation and create unsustainable burdens for healthcare systems.

Box 1: Definitions of “sub-diabetes” (impaired glucose metabolism)

Impaired glucose tolerance134

• Plasma glucose concentration 7.8-11.1 mmol/L (140-200 mg/dL) two hours after 75 g glucose load

Impaired fasting glucose

• WHO: fasting plasma glucose 6.1-6.9 mmol/L (110-125 mg/dL)4 5
• American Diabetes Association: 5.6-6.9 mmol/L (100-125 mg/dL)1

Pre-diabetes

• International Expert Committee (2009):
• “The categorical clinical states pre-diabetes, IFG, and IGT fail to capture the continuum of risk and will be phased out of use as A_1c measurements replace glucose measurements”
• Intervention for HbA_1c ≥6.0% (and maybe below this level if patient demonstrably at high risk6
• American Diabetes Association (2010): HbA_1c 5.7%-6.4%1

Impaired glucose tolerance was established in 1979,3 and its definition has not been altered since. People with impaired glucose tolerance are at increased risk of developing diabetes, with 10 year incidence as high as 60% in some studies.7 They are also at around 50% greater risk of coronary heart disease.7 8 9 Several studies show lifestyle intervention can prevent, or perhaps delay, the onset of diabetes but the role of other interventions is less clear. There is also important debate about how well the new and expanded definitions of pre-diabetes are associated with future diabetes and arterial disease, and responses to interventions to modify risk.

Diagnostic change

Population measures of glycaemia are continuous, with no inflections to provide obvious cut-off points. Cut-offs for the diagnosis of diabetes are based on thresholds for risk of retinopathy.3 5 10 Lesser degrees of hyperglycaemia increase the risk of developing diabetes and maybe arterial disease. But in both cases the risk is graded, making any choice of cut-off point purely arbitrary.

Between 1979 and 1997, the intermediate category was called impaired glucose tolerance. The standard test was measurement of plasma glucose two hours after a 75 g glucose load. The US National Diabetes Data Group defined diabetes as concentrations >11.1 mmol/l (200 mg/dL) and impaired glucose tolerance as 7.8-11.1 mmol/L (140-200 mg/dL),3 and these definitions were ratified by the World Health Organization.

But glucose tolerance testing is laborious for the patient, who must fast, take the glucose load, and then have a blood test two hours later. It is also poorly reproducible—for example, a person with a test result of 8.0 mmol/L (just inside the definition for impaired tolerance) has a roughly 30% chance of a normal result on repeat testing.7 After recommendations from an ADA expert committee in 199710 and WHO in 1999,5 the criterion for diagnosis of diabetes was altered to a fasting plasma glucose concentration of ≥7.0 mmol/L (126 mg/dL), with the intermediate category termed impaired fasting glucose (6.1-6.9 mmol/L (110-125 mg/dL)).5 10 This avoided the need for a glucose challenge test.

In 2003 an ADA expert committee recommended reducing the threshold for impaired fasting glucose from 6.1 mmol/L (110 mg/dL) to 5.6 mmol/L (100 mg/dL).11 The committee said this expansion improved prediction of diabetes risk. But it may also have been influenced by concern that its 1997 fasting glucose criteria identified fewer people than the glucose tolerance test. WHO expressed concern at the public health implications of the change in threshold for impaired fasting glucose4; the expanded category would roughly double the prevalence of sub-diabetes and include people at lower risk of diabetes and cardiovascular disease, who were perhaps less likely to benefit from medical intervention.

More recently, the development of reference methods to standardise assays has allowed measurement of HbA_1c to enter as a third test to diagnose glucose intolerance.6 In 2009, there was reasonable consensus on using HbA_1c >6.5% to diagnose diabetes,1 6 12 although less around an intermediate category (box 1). But in 2010 the ADA reduced the threshold for this intermediate category from 6.0% to 5.7%,1 a decision not endorsed by any other group.

There has also been little support for the ADA’s proposal to label a category of pre-diabetes, into which is rolled all three definitions of sub-diabetes—impaired glucose tolerance, impaired fasting glucose, and borderline HbA_1c (box 2).4 6 12 13 14 This is partly because it has lowered the thresholds for impaired fasting glucose and HbA_1c, but it is also because the imperfect overlap between the three component definitions creates a large, poorly characterised, and heterogeneous category of glucose intolerance.

Box 2: Expert group recommendations on sub-diabetes

• World Health Organization/International Diabetes Federation (2006)4— Recommends using “intermediate hyperglycaemia” to describe glycaemic levels between normal glucose tolerance and diabetes. Use of pre-diabetes is discouraged to avoid any stigma associated with the word diabetes and the fact that many people do not progress to diabetes. In addition, this focus on diabetes may divert attention from the important and significantly increased cardiovascular risk
• International Expert Committee (2009)6—States that a continuum of risk for the development of diabetes across a wide range of sub-diabetic HbA_1c levels may make the classification of individuals into categories using HbA_1c problematic because it implies that we actually know where risk begins or becomes clinically important. The continuum of risk in the sub-diabetic glycaemic range argues for the elimination of dichotomous sub-diabetic classifications, such as pre-diabetes, impaired fasting glucose, and impaired glucose tolerance
• World Health Organization (2011)12—Levels of HbA_1c just below 6.5% may indicate the presence of intermediate hyperglycaemia, but the precise lower cut-off point for this has yet to be defined . While recognising the continuum of risk that may be captured by the HbA_1c assay, the International Expert Committee recommended that people with an HbA_1c level of 6.0-6.5% were at particularly high risk and might be considered for interventions to prevent diabetes
• National Institute for Health and Care Excellence (2012)14—Recommends using a validated computer based risk assessment tool to identify people who may be at high risk of type 2 diabetes. A fasting plasma glucose of 5.5–6.9 mmol/L or an HbA_1c level of 6.0–6.4% indicates high risk

Effect of ADA criteria on prevalence

A recent study in 98 658 Chinese adults2 found a prevalence of impaired glucose tolerance of 8.3%, but over three times as many people (27.2%) satisfied the expanded ADA criteria for impaired fasting glucose and even more (35.4%) met the glycated haemoglobin criteria. Furthermore, the imperfect overlap of the populations that the tests identify provided a total population of 50.1% with ADA defined pre-diabetes.2 These numbers represent 493.4 million Chinese adults.

In the US a study using nationally representative data of 3627 people aged over 18 showed that the age adjusted prevalence of impaired glucose tolerance was 13.5%.15 This compared with a prevalence of 6.8% for impaired fasting glucose by WHO criteria, 25.5% for impaired fasting glucose by expanded ADA criteria, and 13.7% for borderline raised glycated haemoglobin. Another study using a similar dataset found that the lower thresholds for fasting glucose and glycated haemoglobin increased the prevalence by factors of 3 and 4 respectively, these extra numbers being at lower levels of risk.16

The convenience of measuring glycated haemoglobin is likely to influence diagnostic patterns. Glucose tolerance testing is uncommon and testing fasting glucose is inconvenient. Glycated haemoglobin can be measured regardless of time of day, making the process of screening and case finding simpler. But this will result in the highest prevalence of pre-diabetes.

Overdiagnosis and underdiagnosis

Using the oral glucose tolerance test, fasting glucose, and HbA_1c to diagnose glucose intolerance is harder and more error prone than diagnosing diabetes. This is because intolerance is created between two cut-off points (rather than one for diabetes) for measures that have substantial biological and assay variability.

Another challenge is that even were the three tests to diagnose a similar prevalence of the population as being glucose intolerant, they do not identify the same people.7 13 For example, the prevalence of borderline HbA_1c concentrations in non-Hispanic black people is twice as high as in non-Hispanic white people, while the converse is true for impaired glucose tolerance. People of black African heritage also have higher concentrations of glycated haemoglobin and other markers of glycaemia than other ethnic groups.17 18 Care is therefore needed when thresholds for glucose intolerance derived from one population are applied to other demographic groups.

Furthermore, glucose tolerance by all criteria deteriorates with ageing13 so prevention of diabetes may represent little more than delaying its eventual development. Because impaired glucose tolerance, fasting glucose concentrations, and HbA_1c reflect different metabolic phenomena, any relation with complications such as arterial disease may also differ.

Questions over value of pre-diabetes

The logic of creating a diagnostic category of pre-diabetes is that it can provide benefit by precisely identifying those who will develop diabetes, allowing for effective interventions targeting both the disease and its complications. However, the evidence does not necessarily support this logic.

Is a test of glycaemia necessary for prediction?

A recent paper reviewed 94 risk prediction models for diabetes, less than half of which included a measure of glycaemia.19 There was almost complete overlap of the discrimination and calibration characteristics of those with and without such measures.

Does diagnosis of pre-diabetes guarantee future diabetes?

The term pre-diabetes implies inevitable progression and risks stigmatisation. Yet a meta-analysis of the progression rates of pre-diabetes defined according to different glycaemic measures found that even with the best predictor, impaired glucose tolerance, more than half of people identified will be free of diabetes 10 years later.20 The same meta-analysis suggests that around two thirds of people with impaired fasting glucose will not have diabetes after 10 years. To date, studies have suggested that rates of progression in people with borderline glycated haemoglobin are similar to those with impaired fasting glucose,21 22 23 but none has assessed the new lower ADA glycated haemoglobin threshold.

Does lifestyle intervention prevent diabetes and its complications?

There have been three major trials of diabetes prevention with intensive lifestyle counselling—in China (n=577),24 Finland (n=522),25 and the US (the Diabetes Prevention Program, n=3234).26 All were in people with impaired glucose tolerance and a mean age around 50 years. Each reported a 40%-60% relative risk reduction in the incidence of diabetes, with one case of diabetes being “averted” by treating around seven people with impaired glucose tolerance for three years.27 28 29 But the rates of diabetes during follow-up after the trials imply that the lifestyle interventions delayed the onset of diabetes by around two to four years, rather than prevented it altogether.28 29

The Chinese study had three intervention groups: healthy diet, exercise, or both. It reported that the combination of diet and exercise intervention reduced the 20 year incidence of severe diabetic retinopathy from 16.2% to 9.2%.30 The 23 year cardiovascular and all cause mortality was reduced by 20% to 12% and by 38% to 28% respectively, these differences being seen only in women.31 These findings seem surprising for interventions that delayed diabetes onset by only 3.6 years.29 The Finnish study found no effect on cardiovascular risk,32 and this was confirmed in a meta-analysis.33 There are no data on the effect of similar interventions among people labelled as pre-diabetic using impaired fasting glucose or HbA_1c.

The interventions in these studies were based on individual attention and advice. Rolling out intensive lifestyle interventions like these to populations with pre-diabetes (comprising an estimated 86 million people in the US34 or 493 million in China2) would be challenging. Indeed a recent meta-analysis of 22 studies of lifestyle interventions through routine healthcare programmes for diabetes prevention found a mean weight loss of 2.1 kg35—less than half the 5.6 kg reported in the US Diabetes Prevention Program,26 with commentators concluding that “the absence of any persuasive evidence for the effectiveness of community programs calls into question whether the use of public funds or national prevention initiatives should be supported at this time.”16

What about drugs?

The concept of pharmacological prevention is attractive for both the busy clinician and the drug industry. The Diabetes Prevention Program included a randomised controlled trial of metformin and troglitazone in people with impaired glucose tolerance. The troglitazone arm was discontinued because of toxicity. Metformin reduced the 2.8 year incidence of diabetes by 31% compared with placebo,26 but the final oral glucose tolerance test was done while participants were still taking metformin—the first line treatment for type 2 diabetes. Most of this effect remained after 1-2 weeks of drug washout.36 Longer follow-up showed that metformin did not prevent diabetes but delayed diabetes by around two years, even though over half these people were taking metformin during the follow-up.28

Two studies of thiazolidinediones have also been published, both in people with impaired glucose tolerance. The three year DREAM trial37 of rosiglitazone studied 5269 people with impaired glucose tolerance or with impaired fasting glucose by WHO criteria (box 1) and the ACT NOW trial38 of pioglitazone followed 602 people with impaired glucose tolerance for around 2.4 years. In both trials, the incidence of diabetes was reduced (relative risk reduction 62% in DREAM and 72% in ACT NOW). However, testing was done without drug washout, raising the question of whether diabetes had been prevented or merely disguised by treatment.

Harms and risks of overdiagnosis

But even if drugs can delay diabetes in some or all types of pre-diabetes, should people receive these drugs in order to slow the incidence of diabetes? The concept, perhaps combined with epidemic levels of pre-diabetes in “emerging markets,” is exciting the pharmaceutical industry. A search on the ClinicalTrials.gov registry using the search terms “pre-diabetes” and “drugs” shows 422 such trials (21 April 2014). However, there is a hazard in creating a pre-disease associated with a disease such as type 2 diabetes, which is itself little more than a risk factor. The biochemical diagnosis of type 2 diabetes is based on a surrogate endpoint.39 The downsides of being diagnosed with diabetes include the need for medical care and treatment, with its costs and risks, challenges with insurance and employment, anxiety about future complications, and self image. Pre-diabetes could be defined as a risk factor for developing a risk factor. With this label comes much of the same baggage as for diabetes, without evidence of long term benefit (box 3).

Box 3: The balance sheet of “preventing” diabetes40

• The DREAM study37 reported that 14 in 100 people were prevented (or postponed) from developing diabetes by taking rosiglitazone for 3 years. This means that 86 in 100 healthy people who weren’t going to develop diabetes in three years were put on a drug that causes heart failure and fractures and has been under suspicion of increasing cardiovascular risk
• The US Diabetes Prevention Program results imply that you can give an at-risk person with pre-diabetes a 100% chance of using metformin with the goal of reducing by 31% their risk of developing a condition that might require them to use metformin26

Individual or population approach?

Only a year before the ADA produced its latest guidelines, it partnered the European and international diabetes associations to appoint an expert committee.6 The committee recommended abandoning the term pre-diabetes and suggested an HbA_1c level of ≥6.0% as a threshold for preventive interventions. Nevertheless it is the ADA’s 2010 criteria, and the label of pre-diabetes, that dominate the scientific literature, despite the reservations of many organisations, including WHO (box 2). The marked contrast in approach may represent the dominance of a medical model over a public health approach, predicating individual lifestyle advice and perhaps drugs, to prevent or delay increasing glycaemia. This “glucocentric” approach41 is perhaps influenced by the dominance in committees of clinical endocrinologists, rather than by any ties to industry, as has been suggested for other conditions.42 And there are risks of authoritative US based guidelines being extrapolated to other populations, with their prestige potentially influencing global treatment.43

The implementation of the new ADA criteria for pre-diabetes1 is unfeasible. Providing everyone identified by these criteria with personalised lifestyle advice, with or without metformin or other medication, will place unmanageable demand on health services. This strategy also risks distracting attention from those who actually have diabetes and are at higher risk, and in arguably greater need of personalised medical attention.

The dramatic increase in the numbers of people developing diabetes is a global public health problem and needs population and ecological strategies to tackle it. Interventions to improve diet and increase physical activity are less likely to succeed when they seem to be aimed at just a subset of the population which is being encouraged to swim against the tide—although when, as in China, over 50% of adults have pre-diabetes the tide may be turning.

Population strategies to “prevent diabetes” and to treat diabetes are identical. The dividing line is, in this sense, largely irrelevant: pre-diabetes represents little more than a downward shift of the criteria for diagnosing a single disease, so embracing people who may or may not develop the condition.

Fortuitously, first line “treatment” for pre-diabetes by whatever definition is lifestyle advice. And because the risk factors overlap with those of other non-communicable diseases, the question is why focus attention on a specific group of people with a diagnosis of pre-diabetes while ignoring the remainder of the healthy population who would benefit from the same advice. For countries with a high prevalence, such as China, the case for a whole population public health approach is compelling. The real question is whether it is “worth” having the category of pre-diabetes at all.

The ADA should collaborate with the International Diabetes Federation (which regularly collates data on global prevalence of diabetes and impaired glucose tolerance for its Diabetes Atlas44) and with WHO. Together these bodies should seek to define the characteristics of glycated haemoglobin as a predictor of future risks of both diabetes and arterial disease in different populations—ages, ethnicity, and geography. This should be compared with fasting and two hour post-load glucose concentrations (table⇓).

Evidence on value of various definitions of sub-diabetes

• View popup
• View inline

The effect of preventive interventions needs exploring at both public health and individual level. Biochemical measures are of greater importance to physicians than to patients, whose main concerns are the long term complications of the condition, and these outcomes must be the prime considerations when designing future studies. Because the effect of glucose lowering on such outcomes may take decades to become apparent, modelling approaches may be needed. Until then, the recommendations of the 2009 International Expert Committee regarding the continuum of risk6 should be accepted and the term pre-diabetes put in cold storage.

We need a shift in perspective. It is critically important to slow the epidemic of obesity and diabetes. Rather than turning healthy people into patients with pre-diabetes, we should use available resources to change the food, education, health, and economic policies that have driven this epidemic.

What to discuss with patients

• A diagnosis of pre-diabetes does not mean that you will develop diabetes. In fact, of 100 people like you, fewer than 50 are likely to develop diabetes in the next 10 years
• There are ways of reducing your risk of developing diabetes that involve changing your diet and being active. These can result from efforts you make as well as changes in your environment (food supply, workplace conditions, education, and other social determinants of health)
• There are drugs to delay diabetes, but these are the same drugs you will need if you do develop diabetes, and the value of starting them before you have developed diabetes is unknown

What is the next diabetes investigation?

Learning points

• Haemoglobin A_1c (HbA_1c) can now be used as an alternative test to glucose concentration for diagnosing type 2 diabetes or identifying people at high risk of developing the disease
• Be aware of the conditions in which use of HbA_1c would be inappropriate, including suspected type 1 diabetes, pregnancy, acute medical illness, and kidney failure
• Also be mindful of conditions that might affect HbA_1c, such as abnormal haemoglobins and anaemia
• Do not routinely test both glucose and HbA_1c in the same patient

A 48 year old man presented to his general practitioner with a 12 month history of fatigue (which he put down to long office hours) and with urinary frequency. He had no previous health problems, his blood pressure was 145/85 mm Hg, and his body mass index was 29. His father had developed type 2 diabetes at the age of 65 years, and his paternal grandmother had been found to have diabetes at the age of about 60 following the development of a gangrenous toe. The patient’s dipstick urine test showed no glycosuria, ketonuria, proteinuria, blood, leucocytes, or nitrites.

All the possible causes of fatigue should be considered,1 but given the patient’s symptoms and his risk factors for developing type 2 diabetes, including family history and being overweight, a diagnosis of diabetes certainly needs to be excluded. Tests for diabetes are used to evaluate both patients with symptoms (as in this case) and asymptomatic patients who have been identified by a validated risk assessment tool as being at high risk of developing type 2 diabetes.2

Using glucose to diagnose diabetes

Since the early 20th century, the diagnosis of diabetes has been based on the measurement of glucose concentrations in the blood. This usually takes the form of laboratory measured fasting plasma glucose concentration and, when indicated, a glucose concentration two hours after an oral glucose load. However, “random” (post-prandial) measurement can suffice if it is unequivocally raised, especially in a patient with symptoms. The diagnostic threshold concentrations for glucose in use by the World Health Organization are defined as those above which it is known that a person will be at high risk of developing, if they are not already present, the microvascular complications of diabetes, particularly retinopathy.3 In non-pregnant adults, the main indication for an oral glucose tolerance test is when the fasting plasma glucose concentration lies between the values suggestive of normality and overt diabetes—namely, in the impaired fasting glucose range of 6.1-6.9 mmol/L inclusive. The two hour post-glucose load measurement can then help to distinguish patients who have solely impaired fasting glucose from those who have both impaired fasting glucose and impaired glucose tolerance (plasma glucose concentration 7.8 to <11.1 mmol/L) and from those who can be diagnosed as having diabetes purely on the basis of their two hour glucose result being 11.1 mmol/L or above (box 1).

Box 1 Venous plasma glucose thresholds3

Diabetes mellitus

• Fasting glucose ≥7.0 mmol/L or
• Two hour post-glucose load ≥11.1 mmol/L or
• Random glucose ≥11.1 mmol/L

Impaired glucose tolerance

• Fasting (if measured) <7.0 mmol/L and
• Two hour post-glucose load ≥7.8 to <11.1 mmol/L

Impaired fasting glucose

• Fasting glucose ≥6.1 to <7.0 mmol/L and
• (If measured) two hour post-glucose load <7.8 mmol/L

• For asymptomatic patients, at least one additional glucose test result with a value in diabetic range is essential for diagnosis. Impaired glucose regulation refers to a patient who has either impaired fasting glucose or impaired glucose tolerance

Using haemoglobin A_1c to diagnose type 2 diabetes

As can be seen, measuring glucose in the blood to diagnose diabetes can be inconvenient for patients, as they are usually required to fast overnight; if an oral glucose tolerance test is needed, the procedure is laborious, time consuming, and costly. For this reason, in recent years, more consideration has been given to whether measurement of glycated haemoglobin—haemoglobin A_1c (HbA_1c)—might be a valid alternative to glucose as a diagnostic test for diabetes, although this concept has led to controversy.4 Quite apart from not requiring a patient to fast overnight, HbA_1c measurement has several other potential advantages over glucose (box 2), including its property of giving an indication of glycaemia over several preceding weeks rather than at a single time point and, partly as a consequence, reduced day to day variation within an individual compared with glucose.5

Box 2 Advantages of HbA_1c over glucose in diagnosing type 2 diabetes

• Does not require patients to fast, take a glucose solution (which can sometimes cause nausea), or return for second blood test after two hours
• Assesses glycaemia over previous weeks or months
• Lower biological variability than fasting glucose or two hour post-glucose load concentration
• Fewer pre-analytical concerns, including time to analysis
• Already used to guide management of diabetes
• Standardisation of HbA_1c measurement should help with harmonising results between laboratories

Advances in the global standardisation of HbA_1c measurement culminated in WHO publishing advice in 2011 that recommends an HbA_1c threshold of 48 mmol/mol (6.5%) or above for the diagnosis of type 2 diabetes but does not give specific guidance below this single value.6 Since then, an expert committee in the United Kingdom, which included seven clinical professional bodies and National Health Service organisations, came to a consensus recommending that a diagnosis of diabetes should be made only after a confirmed raised HbA_1c value. The committee also introduced a new category of patients who are judged as being at high risk of developing diabetes solely on the basis of an HbA_1c value of 42-47 mmol/mol (6.0-6.4%) (figure⇓).7

Using haemoglobin A_1c (HbA_1c) to diagnose type 2 diabetes in non-urgent situations. *HbA_1c values >120 mmol/mol (13.1%) are likely to indicate marked hyperglycaemia that may need urgent assessment

When not to use HbA_1c to diagnose diabetes

One of the main advantages of HbA_1c—that it can give an indication of previous glycaemia—is also a disadvantage when hyperglycaemia could have developed rapidly, as rises in HbA_1c will lag behind those of glucose. This is why the test is unsuitable in clinical situations such as suspected type 1 diabetes, as well as many of the others described in box 3. Also, most laboratories are able to analyse glucose much more rapidly than HbA_1c, so requesting HbA_1c could introduce delay in an acute situation. In kidney failure (chronic kidney disease stage 5), the picture is complicated by patients often having a combination of haemolytic, iron deficiency, and chronic inflammation anaemias as well as forming urea derived carbamylated HbA_1c, which can also affect some HbA_1c analyses. Several treatments for HIV are also known to influence the HbA_1c value independently of glycaemia. Measurement of HbA_1c is not recommended when determining whether a pregnant woman has gestational diabetes, as it seems to be a poorer predictor of adverse fetal outcome than is glucose.8

Box 3 When not to use HbA_1c for diagnosis and when to be cautious

Do not use HbA_1c

• All children and young people
• Pregnancy—current or recent (<2 months)
• Suspected type 1 diabetes, at any age
• Short duration of symptoms of diabetes (<2 months)
• Patients at high risk of diabetes who are acutely ill
• Patients newly taking drug that may cause rapid rise in glucose, such as corticosteroids, antipsychotic drugs
• Acute pancreatic damage or pancreatic surgery
• Kidney failure
• Patients being treated for HIV infection

Be cautious in requesting or interpreting HbA_1c

• Patient has or may have abnormal haemoglobin
• Patient is anaemic (any cause)
• Patient is likely to have altered red cell lifespan (for example, post-splenectomy)
• Patient has had recent blood transfusion

Other cautions with using HbA_1c

Although HbA_1c should not be used in the situations already described, caution must also be exercised when using HbA_1c in the presence of an abnormal haemoglobin or in conditions that may affect red cell survival (box 3).7 For example, haemoglobin E will form HbE_1c instead of HbA_1c, which may lead to an incorrect assessment of HbA_1c depending on the particular measurement method used by the local laboratory. Haemolytic anaemia can cause low HbA_1c values compared with glucose measurements, and iron deficiency anaemia can cause a raised HbA_1c, although how much influence iron deficiency might have at the diagnostic threshold is not yet clear. After a splenectomy, the lifespan of red blood cells is often increased and so could lead to HbA_1c values that are higher than would be anticipated for the level of glycaemia.

HbA_1c increases with age beyond what can be explained by any changes in fasting glucose or two hour post-glucose load concentrations, and people with Afro-Caribbean or Asian heritage have higher HbA_1c values than do those from Europid descent, which also cannot be accounted for by differences in oral glucose tolerance test results. However, the relevance of these observations to the use of HbA_1c as a diagnostic test remains uncertain.7

Glucose or HbA_1c for diagnosis?

The diagnosis of type 2 diabetes can be made on the basis of either HbA_1c or blood glucose criteria being met. However, these will not identify an identical population of people, as they are not completely concordant with one another.4 For this reason, UK recommendations advise that only one or other test is used to follow the same patient and not a mixture of the two. So if HbA_1c shows a patient to be at high risk of diabetes, he or she should be followed up using the same test rather than blood glucose also being measured at the same time or later. The exception is if HbA_1c measurement is initially or subsequently identified as being inappropriate for that person, in which case a change to glucose measurement is warranted.

Laboratory or point of care measurement?

Several instruments for rapid point of care testing of HbA_1c are available for the monitoring of patients known to have diabetes, but most of these analysers do not perform sufficiently well to be used for diagnostic purposes.9 If they are used, the analytical quality needs to be able to match that of clinical laboratories.6

Outcome

This patient had his HbA_1c measured and found to be 44 mmol/mol (6.2%). As this placed him into the category of being at increased risk of diabetes, he was given lifestyle and dietetic advice and had an assessment of other cardiovascular risk factors. He was asked to report any worsening in his symptoms of diabetes should this happen before the annual HbA_1c measurements now planned.

Patients with ketosis prone type 2 diabetes

This article discusses how to diagnose and manage patients with ketosis prone type 2 diabetes

Summary points

• Patients presenting with diabetic ketoacidosis may have type 1 or type 2 diabetes
• Diabetic ketoacidosis should be treated with insulin in accordance with nationally agreed guidance
• After treatment of diabetic ketoacidosis, patients found to have type 2 diabetes may not require lifelong insulin treatment
• Consider ketosis prone type 2 diabetes in older, overweight, non-white patients who present with diabetic ketoacidosis at their first presentation of diabetes; this diagnosis is also a possibility in patients with any features that are atypical for type 1 diabetes
• Discharge all patients on insulin and arrange for specialist follow-up
• Under specialist supervision consider whether insulin can be down-titrated on the basis of clinical progress and, where possible, C peptide and antibody measurements

Who gets diabetic ketoacidosis?

Diabetic ketoacidosis (DKA) is not just the hallmark of absolute insulin deficiency in type 1 diabetes—it is increasingly being seen in people presenting with type 2 diabetes.1 2 This is at odds with traditional physiological teaching—that clinically significant ketosis does not occur in the presence of insulin concentrations associated with type 2 diabetes because there will always be sufficient insulin to suppress lipolysis (fig 1⇓).3 Current knowledge suggests that some people with type 2 diabetes may develop acute reductions in insulin production, which, coupled with insulin resistance, can cause DKA, usually without a precipitant.4 This is particularly so in African-Caribbean and other non-white ethnic groups.5 6 This potentially life threatening presentation of type 2 diabetes is referred to as ketosis prone type 2 diabetes (also Flatbush or type 1b diabetes). Clinicians should be aware of this variant of type 2 diabetes because observational studies in African-Caribbean people presenting with ketoacidosis indicate that 20-50% have type 2 diabetes.2

Fig 1 Physiological effects of circulating insulin on ketone production. Lipolysis is the process by which triglycerides are hydrolysed to fatty acids. This is controlled by hormone sensitive lipase, which in turn is inhibited by insulin. Fatty acids are oxidised to acetyl CoA, which enters the Krebs cycle to produce cellular energy. In type 1 diabetes, absolute insulin deficiency causes acetyl CoA production to exceed the oxidative capacity of the Krebs cycle, causing the formation of ketone bodies. In type 2 diabetes, endogenous insulin is sufficient to suppress uncontrolled lipolysis and ketone formation. However, in ketosis prone type 2 diabetes, insulin secretion can be acutely reduced, which, on a background of insulin resistance, leads to uncontrolled lipolysis and ketone formation

• Download figure
• Open in new tab
• Download powerpoint

What is known about the pathophysiology of ketosis prone type 2 diabetes?

It is unclear why some people with type 2 diabetes are susceptible to DKA. Polymorphisms in key transcription factors involved in islet cell development are common in ethnic groups that are prone to this condition.7 Other studies have implicated glucose-6-phosphate dehydrogenase deficiency, which may lead to reduced protection of β cell function in the presence of oxidative stress caused by acute hyperglycaemia.8

At presentation of DKA, people with ketosis prone type 2 diabetes fulfil the same biochemical criteria for ketoacidosis as those with type 1 diabetes. However, unlike people with type 1 diabetes, after initial insulin treatment and improvement in glycaemic control, endogenous insulin production recovers over a relatively short time.9 This recovery in insulin secretion is usually sufficient to allow these patients to be managed with oral agents alone for many years.9 10 In between episodes of DKA, β cell function is preserved but suboptimal, and patients remain insulin resistant.5 9 11

Why is it important to recognise ketosis prone type 2 diabetes?

It is important to consider whether patients presenting with ketoacidosis have ketosis prone type 2 diabetes or type 1 diabetes because the diagnosis may never subsequently be questioned. Incorrectly diagnosing ketosis prone type 2 diabetes as type 1 diabetes at presentation may lead to unnecessary long term insulin treatment with potential weight gain, hypoglycaemia, and implications for employment and quality of life. Correct recognition of ketosis prone type 2 diabetes enables most cases to be treated successfully with oral agents and insulin to be safely down-titrated and stopped over a period of months.2 9 12

Patients with ketosis prone type 2 diabetes will also need different education and follow-up from those with typical type 2 diabetes. Despite effective treatment with oral hypoglycaemic agents, patients with ketosis prone type 2 diabetes are at risk of further hyperglycaemic episodes or DKA.9 As with type 1 diabetes, education should focus on capillary blood glucose testing, home ketone testing, and the recognition and avoidance of DKA.2 12 Current guidelines advocate testing for urine ketones only in self management of type 1 diabetes,13 and guidance on self management of type 2 diabetes does not mention ketosis prone type 2 diabetes. Testing for both capillary blood glucose and urine ketones may ensure early self management of hyperglycaemia associated ketosis, allowing for appropriate early management and avoidance of admission, as is seen for type 1 diabetes.

How do we recognise ketosis prone type 2 diabetes?

Clinical features

Owing to the phenotypic heterogeneity of people with ketosis prone type 2 diabetes, type 1 diabetes, and type 2 diabetes, no reliable specific features can clearly distinguish ketosis prone type 2 diabetes (table 1⇓).

Table 1

 Clinical and biochemical differences between adult onset type 1 diabetes, type 2 diabetes, and ketosis prone type 2 diabetes¹²

• View popup
• View inline

However, ketosis prone type 2 diabetes needs to be considered in all non-white patients presenting with DKA, especially those from African-Caribbean, west African, and Hispanic backgrounds, although it has also been reported in white and other minority populations.6 12 14

In the absence of reliable discriminatory features, patients with ketosis prone type 2 diabetes are generally older, more obese, and more likely to have a family history of type 2 diabetes.5 9 12 Age is a poor discriminator because 20-30% of new diagnoses of type 1 diabetes occur above the age of 20 years and ketosis prone type 2 diabetes has been reported in children.12 15

More that half of all emergency admissions to hospital for DKA in patients with ketosis prone type 2 diabetes occur at the time of initial diagnosis of diabetes, after a relatively short history of polyuria, polydipsia, and weight loss with no obvious precipitating causes.2 9 14 The remaining presentations occur in patients with established type 2 diabetes.

Biochemical features

Laboratory tests routinely carried out in emergency departments to establish the diagnosis of DKA (glucose >11 mmol/L (1 mmol/L=18.02 mg/dL), bicarbonate <15 mmol/L (1 mmol/L=1 mEq/L) or pH <7.3, and ketosis with ketonuria or ketonaemia >3 mmol/L) do not distinguish between ketosis prone type 2 diabetes and type 1 diabetes.16 However, patients with ketosis prone type 2 diabetes tend to have higher plasma glucose and glycated haemoglobin (HbA_1c) values than those with type 1 diabetes.2 17

Thus ketosis prone type 2 diabetes can be firmly diagnosed only in retrospect, because specialised laboratory testing and the passage of time are needed to show insulin independence. However, the atypical features described should prompt clinicians to consider the diagnosis. All patients with DKA should be managed with insulin as per national DKA protocols and be discharged on insulin, with an early appointment at the diabetes clinic to undertake tests, review the results, and assess insulin requirements. Biochemical tests such as pancreatic autoantibodies and C peptide measurement may help specialists to make the diagnosis (see below).

How does ketosis prone type 2 diabetes differ from hyperosmolar hyperglycaemic syndrome?

Hyperosmolar hyperglycaemic syndrome is another life threatening metabolic complication of type 2 diabetes, characterised by hyperglycaemia (plasma glucose usually >30 mmol/L), hyperosmolarity (serum osmolality >320 mOsm/kg of water), and hypovolaemia.18 This syndrome is usually easy to distinguish from DKA. Because it is not associated with acidosis or ketosis, hyperglycaemia develops more insidiously and concentrations of glucose are often higher at presentation. See table 2⇓ for key differences between hyperosmolar hyperglycaemic syndrome and DKA.

Table 2

 Comparison of diabetic ketoacidosis and hyperosmolar hyperglycaemic syndrome^{16 18}

• View popup
• View inline

There are also important differences in the acute management of hyperosmolar hyperglycaemic syndrome and DKA. Clinical guidelines recommend fixed rate insulin infusions in hyperosmolar hyperglycaemic syndrome only in the presence of severe ketosis, specifying that this is given at half the rate recommended for DKA to minimise the risk of cerebral oedema.18 Patients with ketosis prone type 2 diabetes, however, should be managed as per the national guidance for DKA, which states a fixed rate insulin infusion.16

What is the natural course of ketosis prone type 2 diabetes?

In these patients, ketoacidosis is caused by an acute reduction in insulin secretion and action, on the background of severe insulin resistance.11 As with type 1 diabetes, exogenous insulin is needed to treat the ketoacidosis. However, once the acute metabolic derangement of hyperglycaemia and accelerated lipolysis (the cause of the ketosis) is reversed with insulin, both β cell function and insulin sensitivity improve. In most cases, good glycaemic control can be maintained with oral agents alone within three to six months.9

Data from follow-up studies of patients with ketosis prone type 2 diabetes show that 70% of patients have at least one repeat episode of acute hyperglycaemia or DKA within two years if treated with diet and lifestyle changes alone. These patients also showed a progressive requirement for insulin with time.9

Data from randomised controlled trials on the treatment of ketosis prone type 2 diabetes are limited. Recurrence of serious hyperglycaemia was lower after treatment with sulfonylureas than diet alone in one study (20% v 72%).19 In addition, pioglitazone significantly reduced the risk of further hyperglycaemia in 68% of cases compared with 32% for lifestyle modifications alone.20 However, neither drug mitigated the risk completely. Metformin, dipeptidyl peptidase-4 inhibitors, and incretin mimetics have not been evaluated, although studies are ongoing.

How should we monitor and follow up patients with suspected ketosis prone type 2 diabetes?

The management challenge in this type of diabetes is not at presentation but at follow-up, when, in addition to considering the diagnosis, the correct distinction between type 1 diabetes and type 2 diabetes also needs to be made.

Consensus from specialist centres suggests that, after an acute admission, all patients should be treated with and discharged on insulin.

Biochemical testing

Autoimmunity and β cell function (using fasting or glucagon stimulated C peptide) should be assessed one to three weeks after resolution of ketoacidosis in a specialist diabetes clinic.2 12 Such tests are not routinely available at all hospitals but are readily accessible at specialised clinical laboratories. Pancreatic autoimmune markers such as glutamic acid decarboxylase (GAD65) or islet antigen 2 (IA2) antibodies are not present in ketosis prone type 2 diabetes, so their absence distinguishes the condition from type 1 diabetes.4

Although the concentration of C peptide, a marker of β cell function, is low at the time of diagnosis of DKA (and therefore of no use at admission), it increases within a few weeks to months, when β cell function recovers.12 This is the hallmark of ketosis prone type 2 diabetes.

The measurement of glucagon stimulated C peptide is currently the best predictor of long term insulin independence, although fasting serum C peptide values also correlate well. Classification of ketosis prone type 2 diabetes according to C peptide values and autoantibody results had 99% sensitivity and 96% specificity for predicting absence or presence of β cell function 12 months after the initial DKA episode. This was significantly better than criteria relying on body mass index, clinical features, and insulin dependence.21

If autoantibodies are negative, C peptide concentrations are sufficient, and glycaemic control is maintained, insulin doses can safely be down-titrated, as long as the patient can perform home blood glucose monitoring and ketone testing.12 Such an approach requires specialist supervision. Once insulin treatment has been stopped and oral agents prescribed, frequent assessment of β cell function reserve, preferably with C peptide measurement, is advised, unlike in the routine follow-up for type 2 diabetes.2 12

The measurement of C peptide will establish whether the patient has recovered sufficient endogenous insulin production to allow insulin treatment to be down-titrated. Follow-up measurements will also predict which patients are likely to require insulin treatment. Conventionally, these decisions have been made clinically—using symptoms, body weight, and glycaemia. However, C peptide measurements are now more widely available and have an emerging evidence base for use in a variety of contexts in the management of people with diabetes.22 Further studies are needed before robust guidelines for its routine use in assessing β cell function and insulin independency in people with ketosis prone type 2 diabetes can be produced.