What’s in a name?

Diabetes rejoices in silly names. Maturity-onset diabetes of the young (MODY) is no doubt the silliest, but—for reasons that will emerge—latent autoimmune diabetes in adults (LADA) comes close. There is, however, one important difference between the two conditions: MODY can be defined at a molecular level and LADA is hard to define at all. LADA, as discussed in this issue of Diabetologia [1], exists somewhere in the borderland between type 1 and type 2 diabetes, and exemplifies the difficulties we have in telling the two conditions apart. From a historical perspective, it is easy to see how these difficulties arose. Idiopathic diabetes was long considered a single disease, despite mounting evidence of heterogeneity. Once formulated, however, the concept of two major disorders with distinct phenotypes, genetic determinants and aetiology won almost immediate acceptance [2] and now rules our thinking, just as the one-disease concept did in the past. For this reason it is rarely acknowledged that the methods we use are of limited value in telling the two conditions apart [3].

So how do we distinguish between them? Early-onset type 1 diabetes differs from type 2 diabetes in—among other things—age distribution, rapidly progressive insulin deficiency resulting in a need for insulin therapy, associations with the HLA system, and the presence of circulating islet autoantibodies. These distinguishing features become less helpful with increasing age, and our ignorance of the role of immune-mediated beta cell failure in people who develop diabetes over the age of 30 represents an embarrassing gap in our understanding of the condition. Type 2 diabetes, meanwhile, has no distinctive genetic or serological features, and tends to be defined by the absence of markers associated with type 1 diabetes. The relative prevalence of the two conditions (assuming there really are two conditions) in middle and later life therefore depends upon how you choose to draw a line between them. Let us consider some of the ways in which this line might be drawn.

Drawing lines in the sand

Age is the simplest means of classification, for ‘juvenile’ diabetes once implied an immediate need for insulin. A paper published in 1983 demonstrated a virtual absence of insulin secretion in all but one of 577 Danish patients diagnosed under the age of 30 [4]. This was in the days before the obesity epidemic, when type 2 diabetes was unknown in young people. The distinction was also circular (young people were put straight on to insulin because they were young) and was of course not helpful in older patients. The alternative to age, enshrined in the terms IDDM and NIDDM, was insulin therapy, although—as one early sceptic remarked—‘it is difficult to think of any condition in which the age of onset ... or the type of treatment required are a satisfactory basis for classification’ [5]. The limitations of classification by insulin treatment are self-evident, for clinicians vary in their readiness to introduce it, treatment thresholds continue to fall, and insulin is now widely used to treat asymptomatic individuals [1]. Attempts to get round this by defining IDDM in terms of early need for insulin merely underscore the unsatisfactory nature of the classification itself.

Measurement of residual insulin secretion, as judged by C-peptide, is a more objective criterion of insulin deficiency, but requires cautious interpretation. The key point here is that diabetes—regardless of type—develops when the body cannot produce as much insulin as it needs. An insulin-resistant individual will have high C-peptide at onset of hyperglycaemia and an insulin-sensitive individual will have lower levels, but the difference relates to insulin sensitivity and not to the underlying pathology of the beta cell [6]. Since insulin-sensitive people require less insulin to maintain normal blood glucose, they will have fewer functioning beta cells at the time diabetes is diagnosed (Fig. 1). This leaves little scope for therapies which stimulate insulin secretion or make the body more sensitive to its actions, effective though these are in insulin-resistant individuals with a larger functioning beta cell mass. The interval between diagnosis and insulin therapy will be shorter in those with lower C-peptide [7], but this is because they have fewer beta cells and does not necessarily mean they have a more aggressive form of disease.

Fig. 1
figure 1

People who are more insulin-resistant have more beta cells at diagnosis of diabetes. This cartoon depicts a beta cell mass that is declining over time. People with LADA are (as a group) less insulin-resistant than those with type 2 diabetes and therefore have fewer functioning beta cells when hyperglycaemia develops. They need insulin sooner because they have more advanced beta cell loss, not necessarily because they have a more rapid process of beta cell destruction

Full size image

The genetic bar code

Many hoped that genetic analysis might rescue the situation. This had been used with great success in classification of the lipid disorders, and geneticists dreamed of the day when clinicians would apply a ‘genetic bar code’ to each and every patient with diabetes [8]. This was not to be. Type 1 diabetes is a polygenic disorder, but the major contribution comes from the HLA region. More than 90% of children diagnosed under the age of 14 in the UK carry the susceptibility haplotypes HLA-DR3-DQ2 or HLA-DR4-DQ8, whereas protective haplotypes containing HLA-DQB1*0602 are virtually absent. These markers might seem to offer a royal road to identification of type 1 diabetes in later life, but there are two major drawbacks. The first is that haplotypes conferring susceptibility to type 1 diabetes are common in the background population, and therefore occur frequently in people with type 2 diabetes, irrespective of any aetiological significance. The second drawback is that the prevalence of these markers in type 1 diabetes is inversely related to age at onset: the excess of HLA-DR3-DQ2/DR4-DQ8 heterozygotes diminishes progressively, for example, while the prevalence of HLA-DQB1*0602 rises [9]. HLA analysis is therefore least helpful in classification of type 1 diabetes in the over-30s, precisely where it might be needed most.

Positive thinking

Young people with type 1 diabetes are typically slim, although less so now than in the past. They present acutely, have low beta cell reserves at onset, are sensitive to the actions of insulin, and almost invariably possess HLA susceptibility haplotypes. Since these features do not discriminate reliably between type 1 and type 2 diabetes in later life, we must fall back upon detection of islet autoantibodies. Islet cell antibodies (ICA), first described in 1974, were identified in 5% to 8% of type 2 patients [10, 11], a subgroup with a ‘strong tendency’ to end up on insulin [11]. Irvine, borrowing from a now obsolete classification of diabetes, considered such patients to have a ‘latent’ form of juvenile diabetes. Groop et al. found ICA in 14% of 154 patients aged 35 to 75 years who had been treated with diet and oral agents for at least 1 year, and showed that this subgroup had lower basal and stimulated C-peptide levels and progressed more rapidly to insulin therapy [12]. The introduction of GAD antibodies plus a catchy new name—LADA—soon brought the condition to more general attention [13].

The value of autoantibody measurement was demonstrated in the UK Prospective Diabetes Study (UKPDS). This cohort was selected on the basis of a clinical diagnosis of type 2 diabetes, defined rather broadly as the absence of ketonuria. When tested for islet autoantibodies, 5.8% of participants had ICA, 9.8% had GAD antibodies, and 3.9% had both. Of 1,870 people not randomised to insulin from the outset, 4.5% of those without these antibodies were on insulin within 6 years, as against 56.3% of those with ICA, 50.8% of those with GAD antibodies, and 78.8% of those with both. The proportion of patients with antibodies fell with increasing age, as did the ability of these autoantibodies to predict progression to insulin treatment. The authors conceded that antibody testing ‘might not be important’ as a guide to insulin therapy if glucose control was adequately monitored [14].

Since LADA can only be diagnosed by antibody measurement, we must question whether this is a secure basis for classification. ICA were measured by washing serum from the person to be tested across a section of pancreas which was then treated with a fluorescent stain to mark sites of antibody binding. Antibodies to GAD and IA-2 contribute to ICA staining but do not fully explain it, and the missing ingredient has so far defied identification. The UKPDS showed that the combination of ICA and GAD antibodies predicts insulin requirement better than either measure on its own. This additive benefit does not necessarily derive from the ‘missing ingredient’ in ICA, however, since testing for two related markers improves specificity by eliminating false positives [15]. Useful though it has proved, ICA measurement has now largely fallen into disuse, for the test is labour-intensive and operator-dependent, and was never really mastered by more than a handful of laboratories. Since islet autoantibodies directed against insulin and IA-2 are rarely present in late-onset diabetes, detection of LADA has come to rely almost exclusively upon measurement of GAD antibodies.

We should therefore pause to consider the significance of a ‘positive’ test for GAD antibodies. This does not, to begin with, mean that these antibodies are either present or absent. They are in fact continuously distributed through the healthy population, and one might as well talk of positivity for plasma glucose or blood pressure. Since ‘positivity’ refers to a selected cut-off point, it is important to appreciate how this is defined. The ideal means of comparing the prevalence of a marker in a test population and in the background population from which this is drawn would be to match cases with controls of the same age and sex. Having determined the distribution of the marker in the two populations, you would retrospectively select a cut-off point that identified those who progressed to insulin treatment. This cut-off, expressed as a given percentile of the control population, would be chosen because it offered the optimal balance of sensitivity and specificity for the purpose intended. You might, for example, lower the threshold and thus increase sensitivity in order to identify as many progressors as possible. Conversely, you could increase the threshold to avoid false positives and maximise specificity if an intervention is planned. This approach has been used in type 1 diabetes [16] and has the advantage that it acknowledges the elastic nature of the self-imposed categories we use and makes us pause to think why we are using them. The idea that any such artificial boundary can make a clean division between two aetiological variants of diabetes will not stand up to serious scrutiny. Nor can we draw valid comparisons between one population and the next, for published estimates of the prevalence of ‘GAD positivity’ vary according to the cut-off that was selected, and the controls that were used to define it.

The importance of controls can be seen in the Botnia Study, in which ‘GAD positivity’ was said to be present in 4.4% of healthy individuals, as against 9.3% of those with diabetes. This guarantees a high rate of false positives, especially in those over 45 at diagnosis, in whom the prevalence in cases is only ∼50% greater than in controls. Not surprisingly, this study also showed that those in the lower two tertiles of GAD positivity differed little from those who were negative [17]. Nor do our difficulties end here, for GAD antibodies are of limited value in defining type 1 diabetes, and were, for example, absent in 24% of Belgian patients diagnosed with type 1 diabetes under the age of 40 [18]. We use GAD antibodies to identify LADA for the same reason that George Mallory climbed Everest—because they are there. We may have no other simple means of identifying immune-mediated diabetes in later life, but we should not confuse the disease with the marker we use to identify it.

Laboratory-based scientists naturally pride themselves upon having a ‘good assay’, as judged by workshop comparisons, but what really matters is the extent to which a test fulfils the purpose for which it is intended. At present any competent assay will identify individuals with very high levels of GAD antibodies, and will therefore predict early insulin requirement in a proportion of those tested, but—in the absence of adequate standardisation and controls—current estimates of the prevalence of ‘GAD positivity’ in different populations have little objective value.

Does LADA exist?

Is LADA a distinct disease entity or one end of a spectrum of immune-mediated diabetes? Given that it is logically impossible to distinguish between two conditions when one (LADA) is defined solely by the characteristics it shares with the other (type 1 diabetes), this easily becomes an exercise in semantics. There is currently no reason to assume that the underlying aetiology is any different. A simple example may help. Jack Spratt is a lean middle-aged jogger and Humphrey Dumpty is his indolent overweight neighbour. Both suffer from progressive beta cell failure and have high levels of GAD antibodies. Jack Spratt is lean and insulin-sensitive, so his blood glucose does not begin to rise until he approaches end-stage beta cell failure. He needs insulin straight away. Humphrey Dumpty, in contrast, is relatively insulin-resistant. He can still make useful amounts of insulin when diabetes is detected. He goes on a diet, enrols at a gym, and starts treatment with an insulin secretagogue, an insulin sensitiser, or both. This corrects the short-term imbalance between insulin secretion and sensitivity and a year or more will pass before he too requires insulin. Jack is considered to have late-onset type 1 diabetes, whereas Humphrey will be said to have LADA, but they differ only in sensitivity to insulin, and the underlying process of beta cell failure is exactly the same. At our present level of understanding, it makes more sense to regard autoimmune diabetes as a continuum, and to devote less time and energy to drawing artificial distinctions between its various manifestations.

Antibody testing, for all its limitations, suggests that immune-mediated diabetes is widely prevalent in the adult-onset population and this has enormous implications for our understanding of the condition. But is it of any practical value? To answer this, we need to consider three further questions.

First, we must ask how well GAD antibodies perform in predicting early insulin requirement. At first glance this looks impressive, for they had a specificity of 94.6% in the UKPDS cohort. High specificity means that the test is good at identifying people who will not need insulin within 6 years. The sensitivity, however, was 37.9%, which means that nearly two-thirds of those needing insulin lacked GAD antibodies. Furthermore, the positive predictive value was 50.8%, meaning that half of those who test positive will not go on to insulin (Fig. 2). The test is least useful in those diagnosed over the age of 45—the age group that contains the overwhelming majority of patients—and only 41/116 (35%) of GAD-positive individuals in this category required insulin within 6 years [12]. Clinicians are not always good at interpreting predictive tests [19] and one wonders how many older patients with GAD antibodies leave the clinic believing that they have a 95% chance of early insulin therapy when their real risk is only one in three?

Fig. 2
figure 2

GAD antibodies are of limited value in predicting insulin requirement. In the UKPDS cohort [14], 1,870 patients were not assigned to insulin at diagnosis of type 2 diabetes. After 6 years, 237 required insulin and 90 (38%) of these had GAD antibodies. Of all those with GAD antibodies, 90/177 (51%) went on to insulin, falling to 41/116 (35%) in ‘GAD-positive’ patients diagnosed over the age of 45 (see text). Presentation in terms of natural frequencies, as here, is easier to follow than the conventional method of calculating probability [19]

Full size image

Second, do GAD antibodies help us decide when to start insulin? There is no answer to this, for the question has not been tested by prospective study. Many clinicians might, however, agree that insulin therapy is much less of a therapeutic landmark than it once was, and that careful monitoring of glycaemic control is all that is needed to make the decision.

Third and most important, does a diagnosis of LADA imply a different approach to clinical management? Speculation apart—and this has not been lacking—there is currently no evidence to support this view. One tiny pilot study suggested that early insulin therapy might be of value [20], but this hypothesis has already been disproved. The UKPDS, it will be remembered, randomised patients to insulin or sulfonylurea at diagnosis, and in so doing set up the conditions for a controlled trial of early therapy for LADA. The study confirmed that patients with a diagnosis of LADA are more likely to progress to insulin, but it also showed that after 10 years, or indeed at any point in between, those initially randomised to diet or sulfonylurea therapy did not differ in any respect from those initially randomised to insulin [21]. On present evidence, therefore, two out of three older patients with a diagnosis of LADA will not need insulin within 6 years, the test is not helpful in deciding when to start it, and no special treatment is required. The clinical utility of a diagnosis of LADA is therefore in doubt.

In summary, the grounds for designating LADA as a distinct aetiological entity are insubstantial, its epidemiology is plagued by methodological problems, and the clinical value of diagnosing it has not been demonstrated. Fourlanos and colleagues map out the route by which we might escape from some of these limitations [1]. Meanwhile, we should note that the tests we use to distinguish between aetiological variants of diabetes are for the most part too imprecise to support the weight of inference that has been placed upon them. The mental categories we derive from such measures are useful as a guide to thought, but are positively harmful as a substitute for it. When pointing to the moon, as Bruce Lee once said, you should not focus attention upon your finger.