We investigated the risk of dementia in patients with type 2 diabetes with or without prior hypoglycaemic episodes.
We investigated the risk of dementia in patients with type 2 diabetes with or without prior hypoglycaemic episodes.
One million subjects randomly selected from the National Health Insurance Research Database, Taiwan.
A total of 15 404 diabetic subjects without prior dementia and a mean age of 64.2 years were enrolled in the study. About 2% (n = 289) of participants had at least one episode of hypoglycaemia in a 3-year period; these subjects were older and more likely to be women and also had higher rates of insulin use and comorbidities compared to those without hypoglycaemia. During a total of 7 years of follow-up (mean and median follow-up, 3.8 and 4.8 years, respectively), 1106 patients with diabetes (7.2%) developed dementia. The incidence rate of dementia was higher in diabetic subjects with [29.9 per 1000 person-years (95% CI 22.1–39.2)] compared to those without [11.1 per 1000 person-years (95% CI 10.3–11.8)] hypoglycaemic episodes. The crude rate ratio (RR) and age- and gender-adjusted RR values for dementia were 2.76 (95% CI 2.06–3.70, P < 0.001) and 1.60 (95% CI 1.19–2.14, P = 0.002), respectively, in diabetic subjects with hypoglycaemia compared to those without hypoglycaemia. Results of Cox proportional hazards analysis revealed that hypoglycaemia, older age, female gender and insulin use were independent predictors of dementia.
Adult diabetic patients with prior hypoglycaemia had a significantly increased risk of dementia. The influence of hypoglycaemic episodes on brain function warrants further investigation.
Hypoglycaemia is one of the most serious complications during treatment of diabetes and has long been recognized as a major barrier to achieving ideal glycaemic control. It has recently been noted that an increased number of hypoglycaemic episodes might account for higher mortality in those patients with diabetes who receive either more or less intensive glycaemic therapy [1, 2].
Type 2 diabetes mellitus is associated with a 1.5- to 2.5-fold increased risk of dementia [3, 4], and cognitive dysfunction is an independent predictor of clinical outcomes in patients with type 2 diabetes . Although the aetiology of dementia and cognitive impairment in people with type 2 diabetes is probably multifactorial, poor glycaemic control with or without other associated risk factors, perhaps by promoting the development of cerebral macrovascular or microvascular lesions, has been shown to increase the risk of dementia in the diabetic population [3, 6]. The influence of hypoglycaemic episodes on subsequent risk of impaired cognitive function or vice versa has been the focus of recent studies [5, 7]. In fact, severe hypoglycaemia, which is generally reversible, has been shown to induce focal neurological deficit and transient ischaemic attacks . In subjects with type 1 diabetes, an association between hypoglycaemia and impaired cognitive function has been shown in some [9, 10], but not in all studies . In subjects with type 2 diabetes, this association is often complicated by the ageing process and the presence of several comorbidities. In a longitudinal cohort study of 16 667 elderly patients with type 2 diabetes from a claims database, Whitmer et al.  showed that severe hypoglycaemic episodes were associated with a greater risk of dementia. In the Fremantle Diabetes Study, Bruce et al.  demonstrated an association between hypoglycaemia and both cognitive impairment and dementia. However, there were no significant associations between historical hypoglycaemia and cognitive decline. In the Edinburgh Type 2 Diabetes Study, severe hypoglycaemia was found to be associated with poor cognitive ability late in life . These studies were all limited either by the use of a single healthcare delivery system , a cross-sectional design  or a self-reported questionnaire . To further explore this issue, we investigated the association between in-hospital diagnosed hypoglycaemic episodes and subsequent dementia by reviewing the medical data of a sample from a claims database that was representative of the general population.
The National Health Research Institutes, Taiwan, maintains the National Health Insurance (NHI) Research Database (NHIRD) and authorizes its use for research purposes. We obtained a subset of the NHIRD with one million random subjects, accounting for about 5% of all subjects enrolled in the NHI programme. There were no statistically significant differences in age, gender or healthcare costs between the sample group and all enrollees (data not show). The database contains medical claims information regarding ambulatory care, inpatient care, dental services and prescription drugs as well – insurance data from all subjects between January 1996 and December 2009 . The International Classification of Diseases ninth revision (ICD-9) coding system was incorporated into the data from the beginning of 2000 and was used in this study.
Patients with diabetes were included in the study if they fulfilled the following three criteria: (i) diagnosis of diabetes in the year 2000, including prevalent and new cases (ii); older than 45 years of age as of 1 January 2003; and (iii) lack of diagnosis of dementia before 1 January 2003. The follow-up period was from 1 January 2003 until 31 December 2009 (Fig. 1).
Diabetes was defined according to ICD-9-CM code 250, and the use of at least one prescription treatment (oral hypoglycaemic agents or insulin) as an outpatient or inpatient was verified. Hypoglycaemia was defined as ICD-9-CM codes 251.0 (hypoglycaemic coma), 251.1 (other specified hypoglycaemia) and 251.2 (hypoglycaemia, unspecified) from the outpatient or inpatient databases between 1 January 2000 and 31 December 2002. The following ICD-9-CM codes were excluded: 250.3 (diabetes mellitus with coma, because hypoglycaemia cannot be distinguished from diabetic ketoacidosis); 250.8 (diabetes mellitus with other specified manifestations, because hypoglycaemia is not specified); 270.3 (leucine-induced hypoglycaemia); 775.6 (neonatal hypoglycaemia); and 775.0 (hypoglycaemia in an infant born to a diabetic mother).
Incident cases of dementia were identified from both inpatient and outpatient databases based on ICD-9-CM diagnostic codes 290.0 (uncomplicated senile dementia), 331.0 (Alzheimer's disease), 290.4 (vascular dementia) and 290.1 (dementia not otherwise specified) from 1 January 2003 to 31 December 2009. Comorbidities, including hypertension (ICD-9-CM 401–405), cardiovascular disease (CVD; ICD-9-CM 390–438), ischaemic heart disease (ICD-9-CM 410–414), chronic kidney disease (CKD; ICD-9-CM 582–583) and hyperlipidaemia (ICD-9-CM 272), were recorded. Data for a total of 15 404 diabetic subjects were used for statistical analysis; amongst these, 4248 subjects were newly diagnosed with diabetes in the year 2000.
Distributions of subjects with and without hypoglycaemic episodes according to age, gender and clinical comorbidities were examined using chi-squared tests for categorical variables and Student's t-tests for continuous variables. The Kaplan–Meier curve and log-rank test were used to examine the difference in the occurrence of dementia between subjects with and without hypoglycaemic episodes. The crude and age- and gender-adjusted hazard ratio in subjects with hypoglycaemic episodes were calculated. The rate ratio (RR) and 95% confidence interval (CI) were also calculated. Multivariate Cox proportional hazards models were used to explore the relation between hypoglycaemic episodes and occurrence of dementia, adjusted for age, gender and comorbidities. The proportional hazards assumption was tested graphically and by including the interaction of time with each covariate. All statistical tests were two sided, and a P value of 0.05 was considered significant. All analyses were performed using sas software, version 9.2 (SAS Institute, Cary, NC, USA).
The mean age of the diabetes cohort was 64.2 years at the time of the survey, and 289 patients (about 2%) had at least one episode of hypoglycaemia between 1 January 2000 and 31 December 2002 (Fig. 1). Table 1 shows the demographic characteristics and comorbidities of the diabetic subjects with and without hypoglycaemic episodes. Comparing these two groups, those with hypoglycaemia were significantly older (P < 0.001), predominantly women (P < 0.001) and had a higher frequency of insulin use (P < 0.001). The rates of comorbidities, including hypertension, CVD, ischaemic heart disease and CKD, were all higher (all P < 0.001) in diabetic subjects with hypoglycaemic episodes; however, the rate of hyperlipidaemia was lower (P < 0.001).
|Variable||Total (n = 15 404)||Without hypoglycaemia (n = 15 115)||With hypoglycaemia (n = 289)||P-value|
|n (%)||n (%)||n (%)|
|Age, years||64.2 ± 9.9||64.0 ± 9.9||70.3 ± 9.7||<0.001a|
|45–54||3412 (22.2)||3390 (22.4)||22 (7.6)||<0.001|
|55–64||4611 (29.9)||4559 (30.2)||52 (18.0)|
|65–74||5053 (32.8)||4935 (32.7)||118 (40.8)|
|≥75||2328 (15.1)||2231 (14.8)||97 (33.6)|
|Male||6947 (45.1)||6850 (45.3)||97 (33.6)||<0.001|
|Female||8457 (54.9)||8265 (54.7)||192 (66.4)|
|No||12 529 (81.3)||12 383 (81.9)||146 (50.5)||<0.001|
|Yes||2875 (18.7)||2732 (18.1)||143 (49.5)|
|No||5603 (36.4)||5554 (36.8)||49 (17.0)||<0.001|
|Yes||9801 (63.6)||9561 (63.3)||240 (83.1)|
|No||6670 (43.3)||6589 (43.6)||81 (28.0)||<0.001|
|Yes||8734 (56.7)||8526 (56.4)||208 (72.0)|
|Ischaemic heart disease|
|No||12 992 (84.3)||12 782 (84.6)||210 (72.7)||<0.001|
|Yes||2412 (15.7)||2333 (15.4)||79 (27.3)|
|No||14 755 (95.8)||14 512 (96.0)||243 (84.1)||<0.001|
|Yes||649 (4.2)||603 (4.0)||46 (15.9)|
|No||10 313 (67.0)||10 091 (66.8)||222 (76.8)||<0.001|
|Yes||5091 (33.1)||5024 (33.2)||67 (23.2)|
During a mean follow-up period (from 1 January 2003) of 3.8 years and median period of 4.8 years, a total of 1106 patients with diabetes (7.2%) were diagnosed with dementia (Table 2). The incidence rates of dementia were 29.9 per 1000 person-years (95% CI 22.1–39.2) and 11.1 per 1000 person-years (95% CI 10.3–11.8) for diabetic patients with and without hypoglycaemic episodes, respectively. The crude RR and age- and gender-adjusted RR values for dementia were 2.76 (95% CI 2.06–3.70, P < 0.001) and 1.60 (95% CI 1.19–2.14, P = 0.003), respectively, for all diabetic subjects either with or without hypoglycaemia. Further analysis of the number of hypoglycaemic episodes indicated that the RR of having dementia increased as the number of hypoglycaemic episodes increased (P for trend < 0.001; Table 2). Adjustment for age and gender, however, only slightly decreased the statistical significance (P for trend = 0.004).
|Hypoglycaemic episodes||Without dementia (n = 14 298)||With dementia (n = 1106)||Incidence rate of dementia (per 1000 person-years)||Crude hazard ratio||Age- and gender-adjusted hazard ratio|
|n (%)||n (%)||Rate (95% CI)||RR (95% CI)||P-value||RR (95% CI)||P-value|
|Yes||242 (83.7)||47 (16.3)||29.9 (22.1–39.2)||2.76 (2.06–3.70)||<0.001||1.60 (1.19–2.14)||0.003|
|1||178 (84.0)||34 (16.0)||29.3 (20.5–40.3)||2.49 (1.77–3.50)||<0.001||1.47 (1.04–2.07)||0.003|
|2||51 (85.0)||9 (15.0)||31.3 (15.0–56.3)||2.54 (1.32–4.89)||0.005||1.41 (0.73–2.71)||0.311|
|≧3||13 (76.5)||4 (23.5)||44.9 (14.0–104.4)||3.74 (1.40–9.97)||0.009||2.60 (0.97–6.92)||0.056|
|Without diabetes||238 997 (97.4)||6485 (2.6)||3.9 (3.8–4.0)||0.35 (0.33–0.37)||<0.001||0.50 (0.47–0.53)||<0.001|
Kaplan–Meier survival analysis demonstrated a greater probability of dementia episodes between subjects with and without hypoglycaemic episodes, and the dementia probability curves of the two groups were significantly different after controlling for age and gender (P < 0.001, log-rank test; Fig. 2a). The increased number of hypoglycaemic episodes enhanced the risk of developing dementia (P < 0.001, log-rank test; Fig. 2b).
Because the distributions of age, gender and comorbidities were significantly different between diabetic subjects with and without hypoglycaemic episodes, we constructed a Cox proportional hazards model to adjust for possible confounding variables (Table 3). After controlling for demographic characteristics and comorbidities, diabetic subjects with hypoglycaemic episodes had an increased risk of dementia compared to those without such episodes (HR = 1.45, 95% CI 1.07–1.95). Age was an independent predictor of dementia, and the independent HR increased by 2.70 (95% CI 2.52–2.90) for every 10-year increment in age. The risk of dementia was increased by 32% in women compared with men (HR = 1.32, 95% CI 1.17–1.50). In addition, insulin users had a 32% increased risk of developing dementia compared with nonusers (HR = 1.32, 95% CI 1.14–1.52). After controlling for demographic characteristics, there was no longer a significant difference in the effect of diabetic complications, such as CVD, hypertension, ischaemic heart disease, CKD and hyperlipidaemia, on dementia risk between those with or without hypoglycaemic episodes.
|Variable||Hazard ratio||95% CI||P-value|
|Ischaemic heart disease|
Results from present study clearly demonstrate that adult diabetic subjects with previous in-hospital diagnosis of hypoglycaemia had an almost 3-fold increased risk of developing dementia in the subsequent 7-year follow-up period compared to those without hypoglycaemia (RR 2.76). This difference in risk was reduced but remained significant after adjustment for age and gender. The incidence rate of dementia with or without hypoglycaemic episodes was slightly lower than that previously reported , probably due to the lower age criterion in the present study (45 vs. 55 years) as well as the fact that hypoglycaemic episodes after the first 3 years were not included in the analysis. Consistent with previous findings , our data showed that patients with multiple episodes of hypoglycaemia had a graded increase in dementia risk. This was further supported by recent evidence that recurrent hypoglycaemia exacerbated cerebral ischaemic damage in streptozotocin-induced diabetic rats for increased free radical release from mitochondria .
There are several possible explanations for the link between hypoglycaemia and subsequent development of dementia. Severe hypoglycaemia has been shown to induce focal neurological deficit and transient ischaemic attacks, which might result in neuronal cell death and accelerate the process of dementia . Using functional magnetic resonance imaging in subjects with type 1 diabetes, Bolo et al.  reported reduced cerebral efficiency during hypoglycaemia manifested by deterioration of blood oxygen level–dependent activation and deactivation. Hypoglycaemia induces abnormalities in platelet function and activation of the fibrinolytic system [18, 19]. Increased adrenaline levels during episodes of hypoglycaemia in turn lead to an increase in platelet activation, leucocyte mobilization and blood coagulability . It was also recently shown that endothelial function may be compromised during acute hypoglycaemia . Vessel wall stiffness was found to be greater during hypoglycaemia in patients with type 1 diabetes of longer duration than in those with diabetes of shorter duration . In patients with type 2 diabetes, hypoglycaemia increased the risk of both macrovascular and microvascular events found in a subset of patients with long-term diabetes .
Hypoglycaemic episodes could be considered to reflect the effects of comorbidities or other unmeasured confounding variables rather than a direct cause of development of dementia . In the present study, the attenuation of the risk estimates remained statistically significant after adjustment for age and gender, suggesting that the effects of these confounding variables are not substantial. Taking into account potential comorbidities, the results of Cox proportional hazards analysis demonstrated that only age, female gender and insulin use remained independently associated with risk of dementia. It is well known that hypoglycaemia can affect cognitive function particularly in brains that are vulnerable due to old age . The reasons why diabetic women were more susceptible to development of dementia are not clear at present. Further analysis in the present study also showed that there was no interaction between gender and hypoglycaemia with regard to the future development of dementia (data not show). Indeed, gender differences in incidence of cognitive dysfunction or dementia have not been identified to date in patients with diabetes . It is also unclear whether premature death after hypoglycaemia would be more likely to occur in men and therefore mask the dementia risk. In epidemiological studies of adults without diabetes, hyperinsulinaemia has been independently associated with poorer cognitive performance [24, 25]. Insulin has also been shown to promote secretion and inhibit extracellular degradation, of amyloid-beta protein . We also found that there was no interaction between use of insulin injection and hypoglycaemia with regard to the future development of dementia (data not show). Our observations indicate that even considering the associated comorbidities, insulin exerts its effect either directly or indirectly via causing hypoglycaemia, thus increasing the risk of dementia.
By analysing a relatively similar claim data set, Hsu et al.  reported that the risk of developing Alzheimer's disease in people with type 2 diabetes was increased 2-fold compared with nondiabetic subjects, and use of sulphonylurea and metformin partially reduced this increased risk. The authors found that the incidence of dementia was similar to or slightly lower than the present findings in diabetic subjects without hypoglycaemia (0.778–1.19 vs. 1.11), although they used different ICD coding to define hypoglycaemia.
Some potential limitations of this study should be considered. The incidence of dementia in patients with diabetes may have been underestimated as some of these subjects might have died before the appearance of dementia . Collection of data on hypoglycaemic episodes only in the first 3 years could also be considered a limitation. However, the present study design was selected because people with cognitive impairment might be prone to more episodes of hypoglycaemia , which might further complicate the cause–sequence analysis. In this regard, Whitmer and co-workers confirmed a similar trend after introducing a longer ‘backward’ lag, so that only episodes of hypoglycaemia during the first 5 years of follow-up were considered . Another potential concern is that, due to the nature of the claims database, we could not obtain all laboratory data. It was shown that uncontrolled glycaemia increased the risk of Alzheimer's disease in a diabetic population with a follow-up period of 9 years . However, in at least one previous study, it was found that the association between hypoglycaemic episodes and dementia was independent of glycaemic control as assessed by the level of glycated haemoglobin .
Furthermore, we might have underestimated the frequency of hypoglycaemia. Individuals with asymptomatic or mild to moderate symptoms of hypoglycaemia may treat themselves or receive care from relatives. Indeed, minor or self-reported symptoms of hypoglycaemia were quite common and accounted for most of the hypoglycaemic episodes , even in non-insulin-user type 2 diabetic subjects in Asia . However, in the ADVANCE study, severe, but not mild, cognitive dysfunction increased the risk of subsequent severe hypoglycaemia . Very recently, analysis from the ACCORD study also confirmed that poor baseline cognitive function increased the risk of severe hypoglycaemia in patients with type 2 diabetes both in intensive and standard glycaemic groups . Although the ICD-9-CM code 250 could not distinguish between patients with type 1 or type 2 diabetes, the likelihood of having type 1 diabetes in our study population was minimal as age at onset of type 1 diabetes is usually below 10 years , whereas we enrolled only diabetic subjects older than 45 years at the beginning of 2003. A further limitation is that physicians might unintentionally code nonspecific hypoglycaemia as 250.3 (diabetes mellitus with coma) or 250.8 (diabetes mellitus with other specified manifestations) or indeed not file any claims for hypoglycaemia. It is also possible that, before the occurrence of hypoglycaemia, dementia or cognitive impairment might have been present but undiagnosed and thus not recorded.
In conclusion, our study represents one of the largest cohorts to examine the association between severe hypoglycaemia and development of dementia in adults with diabetes. We found that adult diabetic subjects who had prior in-hospital diagnosis of hypoglycaemia were almost three times more likely to develop dementia (HR 2.76). Multiple episodes of hypoglycaemia present a graded increase in risk. The influence of hypoglycaemic episodes on brain function warrants further investigation.
WH-HS and C-HL collated the data and reviewed/edited the manuscript. WH-HS and C-HL wrote manuscript, reviewed/edited manuscript.
No conflict of interest was declared.
This study was supported in part by grants from Taichung Veterans General Hospital, Taiwan (TCVGH-1013001C, TCVGH-1013002D and TCVGH-1017303B) and the National Science Council, Taiwan (NSC-98-2314-B075A-002-MY3). The authors would like to thank the Biostatistics Taskforce of Taichung Veterans General Hospital for statistical support. This study is based in part on data from the NHIRD provided by the Bureau of National Health Insurance, Department of Health, and managed by the National Health Research Institutes (Registered number 99278). The interpretation and conclusions contained herein do not represent the views of the Bureau of National Health Insurance, Department of Health or National Health Research Institutes.