The results of this prospective cohort study of patients with type-2 diabetes suggest that there is a substantial risk for hypoglycaemia in patients whose treatment is intensified after failure of oral mono or oral dual combination therapy. This risk is particularly high for episodes of hypoglycaemia that are symptomatic but where patients require no help and these episodes are repeatedly during the year. Multivariable adjusted predictors of hypoglycaemia were prior anamnestic hypoglycaemia (OR 4.05), microvascular disease such as retinopathy (OR 3.27), clinically relevant depression (OR 1.81) and, with respect to pharmacotherapy insulin use (OR 2.99). On the contrary, glitazones (OR 0.55), DPP-4 inhibitors (OR 0.57) and GLP-1 analogues (OR 0.48) were associated with a reduced risk. Although sulfonylureas were clearly linked with anamnestic hypoglycaemic events
, there was no significant association between SU use and the risk of hypoglycaemia during the 12 months FU.
Hypoglycaemia is a frequent adverse effect of antidiabetic drug treatment. Its definition is however, even today, still controversial. The practical definition from a clinical viewpoint of severe (requiring external help for recovery) and mild (self-treated) hypoglycaemia, as described in the DCCT
, has been widely adopted for epidemiological and clinical use. In 2005 the American Diabetes Association (ADA) defined hypoglycaemia as an event accompanied by a measured plasma glucose concentration ≤ 3.9 mmol/l. This plasma glucose threshold was chosen because in non-diabetic people glucose counter-regulation is activated at this level and antecedent glucose concentrations of ≤3.9 mmol/l reduce counter-regulatory responses to subsequent hypoglycaemia
. The DCCT definition does however not consider asymptomatic biochemical hypoglycaemia and thresholds for its diagnosis have been suggested to lie somewhere in between 3.5 and 3.9 mmol/l
, although the exposure to glucose levels of 3.5–4.0 mmol/l is likely to be of little clinical significance. Because of this controversy Swinnen et al. investigated how the level of blood glucose might impact prevalence rates reported
. They demonstrated that the ADA definition of ≤ 3.9 mmol/l would result in a high proportion of hypoglycaemic episodes without a particular clinical relevance.
Because of the aforementioned controversy on the correct definition of hypoglycaemia we have chosen to document the broadest range of possible phenotypes of hypoglycaemia in our registry including asymptomatic hypoglycaemia with blood glucose values <2.22 mmol/l or 40 mg/dl, symptomatic hypoglycaemia with a blood glucose <2.78 or 50 mg/dl but without a need for help, symptomatic hypoglycaemia with external lay help, medical help or requiring hospitalization (all irrespective of blood glucose values). This resulted in an incidence rate of 14.1%, the majority because of asymptomatic hypoglycaemia (3.3%) or symptomatic hypoglycaemia but without the need for help (9.7%). Further we had 0.7% of patients with moderate hypoglycaemia and 0.5% were classified as being severe. This corresponds to a rate of 1.2% based on the DCCT definition for severe hypoglycaemia requiring external help for recovery
. Reported rates of severe hypoglycaemic events (as per DCCT) in clinical studies vary between 0.4% (standard therapy group in the ADVANCE study
) and 3.1% per year (intensive therapy group in the ACCORD trial)
. In UKPDS the rates of major hypoglycaemic episodes per year were 0.7% in the conventional group (mean HbA1c 7.9%) vs. 1.4% with glibenclamide and 1.8% with insulin (mean HbA1c 7.0%)
Hypoglycaemia rates were strongly dependent on antidiabetic drug treatment as outlined in Figure
2. While insulin use independently conferred substantial added risk (OR 2.99; 95%CI 2.27-3.95), GLP-1 analogues, glitazones and DPP-4 inhibitors (ORs 0.48, 0.55 and 0.57 respectively) were associated with a reduced risk. This was also subject of a more detailed analysis in Table
2. The risk increase with insulin use is compatible with the results of a number of observational studies such as the UK Hypoglycaemia Study
, a retrospective questionnaire based study from Denmark
 and a study by Donnelly et al.
. The UK Hypoglycaemia Study
 found similar incidence rates of hypoglycaemia in those recently started on insulin and those treated with sulfonylureas but reported increased rates (25%) in those with a longer duration of insulin treatment. Rates reported from Denmark were 16.5% in insulin treated type-2 diabetic patients (44 episodes/100 patient years)
 and 35 episodes/100 patient years of severe hypoglycaemia in the study by Donnelly et al.
. On the other hand, insulin sensitisers (glitazones) and incretin based therapies (DPP-4 inhibitors, GLP-1 analogues) have been associated with a low risk of hypoglycaemia. In the ADOPT study rates of severe hypoglycaemia requiring help (0.1%) in patients on rosiglitazone were comparable to those receiving metformin (0.1%) but substantially lower than those receiving the SU glibenclamide. Rates of self-reported symptomatic hypoglycaemia were 10% for metformin and rosiglitazone and 38.6% for glibenclamide
. Newer agents based on the incretin system such as DPP-4 inhibitors and GLP-1 analogues have been associated with low risk of hypoglycaemia except when combined with sulfonylureas or insulin
Sulfonylureas, which have been associated with a considerable risk of hypoglycaemia were associated with no added risk in the prospective follow-up, contradicting analyses we reported for anamnestic hypoglycaemia prior to treatment escalation at baseline (multivariable adjusted OR of 2.58; 95%CI 2.03-3.29)
. The notion of an increased risk with sulfonylureas has previously been reported from the UK Prospective Diabetes Study (UKPDS)
 where 31% of subjects reporting mild hypoglycaemic symptoms with insulin secretagogues in the first year of use and from the ADOPT study
. A population based study from Germany looking at over 30 000 patients over 4 years found less episodes of severe hypoglycaemia with the newer generation sulfonylurea glimepiride (0.86 vs. 5.6 events with glibenclamide per 1000 patient years)
. Against this background our results are somewhat surprising especially in comparison to analyses predicting anamnestic hypoglycaemia in DiaRegis but have to be weighed against a number of changes that were introduced over the course of 12 months: 1) sulfonylurea use was reduced between baseline and the 12 months follow-up from 29.5 to 24.2%, at least partially also because of hypoglycaemia; 2) oral monotherapy went down from 68.4% at baseline to 17.8% at the 1 year FU in favour of combined OAD treatment regimens and insulin / GLP-1 analogue use. This may have masked hypoglycaemia rates seen with sulfonylureas as has been shown to be the case in combination with for example DPP-4 inhibitors
Finally rates of hypoglycaemia were lower for anamnestic or recalled episodes than those prospectively collected in the patient’s diary. There are a number of potential explanations for this finding beyond treatment intensification. It appears possible, that is merely reflects a recall bias and retrospective recording of hypoglycaemia may have underestimated the background rate. This is exemplified in individuals with type 1 diabetes who can reliably remember episodes of severe hypoglycaemia after an interval of one year, but recall of mild hypoglycaemia becomes unreliable within a week
[28, 29]. In people with insulin-treated type 2 diabetes, recall of severe hypoglycaemia is similarly robust over a period of one year
 but the reliability of recall of mild hypoglycaemia has not been examined and is unlikely to be preserved. In our study, the increase in hypoglycaemia was most marked for episodes which were symptomatic but where no help was required, which are precisely those which may not have been recalled at entry to the Diabetes Registry.
Predictors of hypoglycaemia
Out of a number of variables considered to be potentially predictive of subsequent hypoglycaemia such as anamnestic hypoglycaemia, age, HbA1c, heart failure, non-proliferative or proliferative retinopathy, coronary artery disease, blood glucose self-measurement, evidence of clinically relevant depression and antidiabetic drug treatment, prior anamnestic hypoglycaemia (OR 4.05), microvascular disease such as retinopathy (OR 3.27) and depression (OR 1.81) were identified to be associated with an increased risk beyond antidiabetic drug treatment as outlined above. On the other hand, the predictive value of a number of variables (except for microvascular disease) identified in a multivariable analysis of the ACCORD dataset
 was not confirmed which may be related to the selected patient cohort of the ACCORD study with intensified treatment.
A particular role of anamnestic hypoglycaemia has previously been reported from intervention studies such as the Diabetes Control and Complications Trial (DCCT) and a study reporting on the effects of a teaching programme for intensification of insulin therapy where a history of severe hypoglycaemia was one of the main predictors for an increased risk of future severe hypoglycaemia
[32, 33]. The finding that depression is associated with subsequent episodes of (severe) hypoglycaemia confirms prior data from Finland where depression was also a significant independent risk factor for hypoglycaemia
. Furthermore, Williams et al. found in OAD treated type-2 diabetes patients that hypoglycaemia correlated to significantly lower health related quality of life that included more anxiety / depression on a subscale
Blood glucose self-measurement was also predictive for subsequent hypoglycaemic events, most likely because of the increased awareness in cases where episodes were asymptomatic. This sounds like a self-fulfilling prophecy but is important not only because asymptomatic biochemical hypoglycaemia may result in neurological impairment but also because repeated hypoglycaemia blunts symptomatic and hormonal responses to subsequent episodes leading to impaired awareness of hypoglycaemia, also called hypoglycaemia associated autonomic failure (HAAF)
. These patients often experience glucose concentrations below 2.0 mmol/l without becoming symptomatic. Furthermore, a number of variables such as glycaemic control, alcohol, exercise, and age affects and reduces symptomatic and hormonal responses to subsequent hypoglycaemia
[37–41]. Elderly patients also report different symptoms and responses to hypoglycaemia with less autonomic and more prominent neuroglycopenic symptoms
. In this group, hypoglycaemia can be misdiagnosed as dementia or neurological events
. We therefore believe that blood glucose self-measurement not only helps to detect “mild” hypoglycaemia but also to detect “asymptomatic but severe hypoglycaemia”.
Among the strength of the dataset the prospective collection of data on hypoglycaemic events after an adjustment of treatment at baseline has to be noted. After the switch patients were followed for one year and data on hypoglycaemia rates, co-morbidity, treatment patterns and outcomes collected over the follow-up supplementing prior, mostly retrospective database analyses. On the other hand there are some limitations to the current analysis deserving consideration: 1) The representativeness for the subset of the German population with type-2 diabetes cannot be assessed. This is because we chose to recruit patients in primary and specialized care but not the general population. This is unlikely to affect the overall conclusion however because it can be assumed that, after an initial period in which type-2 diabetes may be unknown (up to 5 years) most patients in Germany regularly attend physicians for diabetes care. 2) Hypoglycaemia rates, especially those based on laboratory values, are still a matter of debate. This is despite the ADA giving recommendations for its diagnosis based on a threshold of ≤ 3.9%, but others have challenged that the consequence of blood glucose values ≤ 3.9% are clinically irrelevant
[13, 14]. We have chosen rather low threshold < 2.22 mmol/ especially for those without symptoms in the attempt not to overrate the frequency of these asymptomatic episodes. It may however be challenged based on the aforementioned. 3) Although data of the most important known risk marker of severe hypoglycaemia, impaired hypoglycaemia awareness, was scored in the DiaRegis
, the completeness of data was insufficient to include this risk marker in the analyses. 4) Finally one might question the selection of variables considered to identify independent predictors of hypoglycaemia. While generally the majority of baseline variables with a significant difference between groups was considered, those with a high likelihood of interference were not selected. On the other hand antidiabetic pharmacotherapy, with only minor differences at baseline were included into the model because of their likely impact on hypoglycaemia rates.