Risk of Incident Dementia According to Glycemic Status and Comorbidities of Hyperglycemia: A Nationwide Population-Based Cohort Study

The rapid aging of populations makes it more important to prevent and manage dementia. There are no curative treatments for dementia, and the contribution of modifiable risk factors (e.g., obesity, hypertension, diabetes, dyslipidemia) to dementia risk is becoming a concern for its prevention (1–3). Among these risk factors, diabetes has been consistently shown to be associated with an increased dementia risk, applicable to vascular dementia (VD) and Alzheimer dementia (AD) (4,5). For example, the abnormal protein processing that causes AD is accelerated by preceding diabetes (6,7).

Prediabetes is an intermediate hyperglycemic state with blood glucose levels higher than normal but lower than the criterion for diabetes. Study results suggest that the brains of patients with prediabetes are exposed to abnormally high levels of insulin due to insulin resistance or impaired glucose tolerance, or both (8). Thus, it can be assumed that prediabetes and diabetes similarly contribute to incident dementia. There has been some research on prediabetes and cognitive function. However, only a few studies have investigated the association of prediabetes with the incidence of dementia (9–11). Several studies found no significant differences in cognitive function between cases of prediabetes and normoglycemia (12–16).

Some vascular diseases increase the risk for dementia occurrence (3). Therefore, the incident risk of dementia is likely to be higher in individuals with comorbidities than in those with diabetes or hyperglycemia alone. However, despite its importance, there is a lack of sufficient studies on the coexistence of comorbidities that are also highly related to diabetes (e.g., ischemic cardio-/cerebrovascular disease and kidney disease) with hyperglycemia and the effect of this relationship on dementia incidence (17,18).

In addition, the contribution of sex to the association of vascular risk factors with the development of dementia is still unclear (19,20). Although the lifetime prevalence of AD is higher in women, sex difference in the risk of AD according to age group and region shows mixed results (21). Whereas men appear to have a slightly higher risk for VD, some risk factors for VD have a greater effect on women (20). It seems important to examine the relationship between sex and other risk factors.

This study was based on a large representative cohort of the general population in South Korea. We investigated whether the risk of incident dementia was associated with hyperglycemia in consideration of the presence of some comorbidities, including differences that may appear depending on age and sex.

We used health insurance claims data and results of health examinations from the National Health Insurance Service (NHIS) of South Korea (22,23). The NHIS administers a mandatory health insurance system that covers the entire population of 50 million individuals in South Korea. The claims database provided by the NHIS consisted of information about sociodemographic characteristics (i.e., age, sex, and residential area), health care utilization history, and diagnostic codes based on the ICD-10 (22). The NHIS also manages the national health examination program (biennial examinations) for people aged ≥40 years. The data provided by the NHIS consisted of laboratory tests and physical measurements (e.g., height, weight, and blood pressure) and self-administered questionnaire responses to medical history and health-related lifestyle questions (23). We performed this study by linking these two data sets, the claims and the health examination databases, through anonymized patient identifiers. The study protocol was approved by the Yongin Severance Hospital Institutional Review Board (IRB No. 9-2020-0103).

For this study, we initially enrolled all participants aged ≥40 years who participated in the biennial health examination program from 2009 to 2010 (n = 15,168,021) (Fig. 1). Subjects with missing variables (n = 5,377,995) were excluded. Then, we discarded the 2010 data for subjects who had health examinations in both 2009 and 2010 (n = 1,300,306). Subjects with a history of any dementia (ICD-10 codes: F00.x–F03.x, F10.7, G30.x, G23.1, G31.0, G31.1, or G31.8) (n = 44,082) and those diagnosed with dementia within 1 year of study entry (n = 44,688) were then excluded. Among the final study population (n = 8,400,950), we defined the index date for each subject as the first (baseline) health examination date in 2009–2010. The follow-up duration for each subject was from the index date to the earliest occurrence of the outcome event or the end of the study period (31 December 2016).

Subjects with diabetes were defined as having a fasting plasma glucose (FPG) value of ≥126 mg/dL or ICD-10 codes for diabetes (E11.x–E14.x) and prescribed antidiabetes medication, or both, at each index date (24). Glycemic status was subcategorized into five groups as follows: 1) normoglycemia (FPG <100 mg/dL), 2) impaired fasting glucose (IFG) (FPG 100–125 mg/dL), 3) new-onset diabetes (FPG ≥126 mg/dL at the index date and no claims for the ICD-10 code of diabetes or antidiabetes medication before the index date), 4) diabetes with a duration of <5 years (or known diabetes <5 years), and 5) diabetes with a duration ≥5 years (or known diabetes ≥5 years). Prediabetes was defined by criterion for IFG, based on the three criteria for prediabetes proposed by the American Diabetes Association (25). The duration of diabetes was defined as the period elapsed from the date of acquisition of ICD-10 codes (E11.x–E14.x) with prescription of antidiabetes medications before the index date in 2009–2010. This approach to categorization of subjects according to glycemic status was similar to that used in a previous study (26).

Based on previous studies related to modifiable risk factors of dementia (1,2), we examined the following variables as covariates, which can be obtained through our data sources. Body weight (kg), height (cm), waist circumference (cm), and blood pressure (mmHg) were measured as a part of the health examinations. BMI was calculated as weight divided by height squared (kg/m²). All blood samples were taken after overnight fasting. Information about health-related lifestyles was acquired using self-administered questionnaires and categorized as current smoker or not, heavy drinker (alcohol consumption ≥30 g/day) or not, and regular exercise (exercise with high intensity of >20 min/day and ≥3 days/week or with a moderate intensity of >30 min/day and ≥5 days/week) or not.

Baseline comorbidities were identified as hypertension (ICD-10 codes I10.x–I13.x, I15.x and treatment with antihypertensive medications, or systolic/diastolic blood pressure ≥140/90 mmHg, or both), dyslipidemia (ICD-10 code E78.x with lipid-lowering agents, or serum total cholesterol ≥240 mg/dL, or both), chronic kidney disease (CKD) (estimated glomerular filtration rate <60 mL/min/1.73 m² [27]). Ischemic heart disease (IHD) (I21.x–I25.x) and ischemic stroke (IS) (I63.x–I64.x) were defined according to ICD-10 code descriptions. CKD and two ischemic diseases (i.e., IHD and IS) were assessed as covariates also in accordance with previous studies (28–31).

The primary outcome was incidence of all-cause dementia (ICD-10 codes F00.x–F03.x, F10.7, G30.x, G23.1, G31.0, G31.1, or G31.8) and events were defined using the registered ICD-10 codes. We also verified the diagnostic codes specifying AD (F00.x and G30.x) versus VD (F01.x).

Descriptive statistics were used to analyze baseline characteristics of study subjects. Dementia incidence rate was calculated by dividing the number of events in each group by 1,000 person-years. For study outcomes, adjusted hazard ratios (aHRs) with 95% CIs were estimated using Cox proportional hazards regression analysis. The regression models were sequentially adjusted for age and sex (model 1), the covariates of model 1 together with the health behaviors-related factors (i.e., smoking, drinking, and exercise) (model 2), and all confounders put into model 2 with the factors related to medical condition (i.e., BMI, hypertension, and dyslipidemia) (model 3). We used the same methods to analyze study outcomes for the AD and VD groups. In addition, we further analyzed aHRs according to change in glycemic status between the baseline and second health examinations. We conducted other hazards regression analyses to investigate the risk of dementia occurrence according to presence of diabetes and glycemic state, with and without comorbidities (i.e., IHD, IS, and CKD). As age and sex are important risk factors for dementia, we performed subgroup analyses with study subjects stratified by age (<65 or ≥65 years) and sex (male or female), respectively. We performed all statistical analyses using SAS 9.2 software (SAS Institute, Cary, NC).

Among a total of 8,400,950 people aged ≥40 years, the percentages of those in the groups with normoglycemia, IFG, and diabetes were 58.3% (n = 4,898,760), 24.0% (n = 2,014,820), and 17.7% (n = 1,487,370), respectively (Table 1). The proportion of patients with new-onset diabetes was 9.8% (n = 821,739), which was higher than the 7.9% (n = 665,631) for those with known diabetes. Age increased according to worsening glycemic status and diabetes duration. Men accounted for the highest proportion of new-onset diabetes (60.3%). BMI, waist circumference, blood pressure, and FPG gradually increased according to worsening glycemic status; the highest levels were present in patients with new-onset diabetes. Current smokers and heavy drinkers were most prevalent in the group with new-onset diabetes. Prevalence of hypertension, dyslipidemia, IHD, IS, and CKD increased according to glycemic status and duration of diabetes; the highest levels were in patients with diabetes duration >5 years. The values for prevalence of hypertension and dyslipidemia were higher in the groups with known diabetes than in the group with new-onset diabetes. Blood pressure and cholesterol levels were higher in patients with new-onset diabetes than in those with known diabetes.

During a median 5.94 years of follow-up, all-cause dementia, AD, and VD occurred in 4.2% (n = 353,392), 3.2% (n = 265,842), and 0.5% (n = 43,942) of the study population, respectively. Overall, aHR values for all-cause dementia, AD, and VD were greater in those with diabetes compared with those without diabetes (Fig. 2), after adjustment for confounding factors. Patients with diabetes had ∼1.4- to 1.5-times higher risk for development of dementia. The risk of incident dementia increased according to glycemic status. For example, the aHR (95% CI) for AD was 1.01 (1.00–1.02) in those with IFG, 1.43 (1.41–1.44) in those with new-onset diabetes, and 1.44 (1.43–1.46) in those with known diabetes (Supplementary Table 1).

The aHRs for all-cause dementia and AD in the group with new-onset diabetes were similar or lower than in the groups with known diabetes (overall); the aHR for VD in the group with new-onset diabetes was generally higher than in the groups with known diabetes (Fig. 2). Interestingly, when the duration of known diabetes was divided by a duration of 5 years, the aHR for any dementia in new-onset diabetes was lower than that of known diabetes ≥5 years, but higher than that of known diabetes <5 years. For example, the aHR (95% CI) for VD was 1.48 (1.44–1.52) in new-onset diabetes, 1.32 (1.28–1.37) in known diabetes <5 years, and 1.66 (1.60–1.72) in known diabetes ≥5 years.

In the models for dementia risk according to change in glycemic status between baseline and the second health examinations (Supplementary Table 2), regardless of the type of dementia, the aHRs (95% CI) were the highest in the group with persistent diabetes (1.47 [1.45–1.49] for all-cause dementia). The aHR in the group that changed from normoglycemia to diabetes was 1.39 (1.35–1.42), which was higher than the group with IFG that proceeded to diabetes (1.23 [1.19–1.26]). Also, the aHR in the group that proceeded from normoglycemia to IFG was 1.03 (1.01–1.05), while aHRs were not significant in the steady IFG group or the normalized group.

Dementia risk increased according to glycemic status and presence of comorbidities (Fig. 3). Except in patients with IS, the aHRs for dementia in the known diabetes groups without comorbidities were greater than those in the normoglycemic group with IHD or CKD. For example, the aHR (95% CI) for all-cause dementia in known diabetes without IHD was 1.47 (1.45–1.48) but was 1.20 (1.18–1.22) in the normoglycemia and IHD groups. Similar results were found in the CKD groups. In the case of IS, the aHR for dementia was higher in normoglycemia with IS than that in known diabetes without IS (aHRs [95% CI] for dementia, 1.90 [1.86–1.95] vs. 1.46 [1.44–1.47]). The risk of dementia showed a synergistic effect when hyperglycemia and comorbidities coexisted; this effect was greatest in the presence of both diabetes and IS.

In the absence of comorbidities, aHRs for all-cause dementia in new-onset diabetes were similar to those of known diabetes but greater than those of known diabetes in the presence of comorbidities (Fig. 3). For example, aHRs (95% CI) for dementia in new-onset diabetes and known diabetes were 1.44 (1.42–1.45) and 1.45 (1.44–1.47), respectively, in the absence of CKD. In the presence of CKD, the aHRs (95% CI) for dementia in new-onset diabetes and known diabetes were 1.67 (1.64–1.70) and 1.61 (1.58–1.64), respectively.

Subgroup analyses were performed according to age and sex (Supplementary Tables 3 and 4). Except for nonsignificant interactions between sex and risk of VD according to the presence of diabetes and glycemic status, all other interactions of age or sex on the risk of dementia were significant. The aHRs for all-cause dementia, AD, and VD in subjects with diabetes were higher in those <65 years than in those ≥65 years. On the contrary, whereas aHRs of the IFG group in those <65 years were not significant compared with the normoglycemia group, aHRs of the IFG group in those ≥65 years were greater than the normoglycemia group for all-cause dementia and AD. In all-cause dementia and AD, aHRs among women in the IFG groups were increased compared with those having normoglycemia, whereas aHRs among men in the IFG groups did not show a significant difference. Meanwhile, aHRs among men with diabetes were greater than those of women with diabetes in all-cause dementia and AD. Regardless of age and sex, coexistence of diabetes and comorbidities resulted in a synergistic effect on dementia risk.

In this population-based study using a nationwide cohort, we found that all degrees and durations of “hyperglycemia” were associated with an increased risk of incident all-cause dementia. Also, some comorbidities, such as IHD, IS, and CKD, were associated with an increased risk of occurrence of all-cause dementia. This finding included AD and VD, which were synergistic with hyperglycemia. To the best of our knowledge, this study is the largest to find that dementia incidence was proportional to severity or duration of exposure to hyperglycemia and comorbidities rather than simply examining the presence of diabetes.

A number of epidemiological studies found that diabetes increases the risk of cognitive impairment and dementia, especially the most common dementia type, AD (2,4,5,7,32–34). The association of diabetes with AD could be explained by the insulin resistance related to hyperglycemia, which dysregulates brain insulin signaling and stimulates amyloid-β deposition in the brain (7,35,36). Hyperglycemia also causes small vessel damage in the brain via disturbance of the blood supply; this change can result in AD and VD (3,7).

We found that even in the prediabetes stage, the extent and exposure period of hyperglycemia was associated with an increased incidence of dementia, after adjusting for potential confounders. Despite the possibility that prediabetes or IFG could progress over time, previous studies found that dementia risk increases within a 10-year period in subjects with prediabetes at a specific time point. A 9-year follow-up study of 1,173 subjects (47 subjects with prediabetes at study entry) found an association between prediabetes and incident dementia (aHR 1.67 [95% CI 1.04–2.67]) (9). Likewise, another study found that subjects with an average glucose level corresponding to prediabetes have poor memory performance compared with normoglycemic subjects (10). In that study, during the 7-year follow-up period of 2,067 subjects, the aHR for incident dementia was 1.18 (95% CI 1.04–1.33) in subjects with prediabetes. Although these two studies reported a somewhat higher risk than we saw in our study (aHR 1.01 [95% CI 1.01–1.02]), the results of both studies correspond in context to results of other relevant studies (3,8,30–32). Recently, a study of 449,973 subjects aged 40–69 years that used the UK Biobank database found that prediabetes is not associated with increased risk of all-cause dementia or AD but is associated with the risk of VD (11). The results were similar between our subgroup analysis stratified by 65 years of age and the results of the U.K. study for those <65 years of age.

When we divided the known diabetes group by the 5-year prevalence period, we found that the risk of incident dementia in new-onset diabetes was higher than that in known diabetes <5 years and lower than that in known diabetes ≥5 years. This result indicated that the risk of incident dementia in new-onset diabetes, which is still prediagnosed and premedicated, outweighs the risk in the population who are treated for diabetes and manage the risk factors for complications after diagnosis. Another interpretation is that even if diabetes is managed, the risk of incident dementia unavoidably increases with prolonged exposure to hyperglycemia.

This epidemiology study of a large population found that the degree of hyperglycemia was related to dementia occurrence. The results of this study suggested that the association of hyperglycemia with dementia occurrence was revealed using a one-time FPG measurement. This assessment is routinely performed in health examinations and is much easier to apply than other tests of diabetes (e.g., oral glucose tolerance test or HbA_1c). On the other hand, a study based of a community sample in Eastern Asia found that associations with dementia occurrence were different depending on the operational definition of hyperglycemic status (37). Therefore, like our study, glycemic status should be separated into categories when regular health examination data are used in analyses of factors that contribute to dementia occurrence.

The analysis of incident dementia according to change in glycemic status showed that the association was greater when IFG is assumed to be a transitional period that progresses to diabetes. When metabolic deterioration progressed more rapidly, the association with incident dementia increased. Additional study is necessary to elucidate this association between glycemic change and risk of incident dementia.

To better understand the effect of hyperglycemia on dementia development, hyperglycemia and comorbidities of hyperglycemia should be included in studies. Previous studies have been limited mainly to studies on the coexistence of diabetes and hypertension (9,38,39). We examined the risk of incident dementia according to glycemic status, with and without comorbidities. Our results indicated that IHD, IS, and CKD could be independent risk factors for incident dementia. The association of IS with dementia was greater than that of hyperglycemia. But, for the other comorbidities, we found that the presence or absence of hyperglycemia was a more important factor contributing to incident dementia.

It is well-known that ischemic conditions of the heart and brain, such as myocardial infarction and cerebral infarction, reduce cerebral blood flow. This change results in small vessel disease, sustained hypoperfusion of the brain, and negative outcomes such as neurodegeneration (30,31). Because IS directly affects the brain structurally and functionally, it is understandable that IS had a greater effect on the occurrence of dementia than the presence or absence of hyperglycemia. CKD is also a risk factor for cognitive decline and dementia (28,29). The results of our study indicated that the presence of CKD was associated with an increased risk of dementia, although not as much as the presence of IHD or IS.

Treatment of diabetes, which is a chronic disease, aims to prevent various long-term complications. Assuming the risk of diabetes-associated complications slightly increases during the prediabetes period, the results of our study suggest that active blood glucose management should start then. At least in terms of dementia prevention, prediabetes should no longer be considered an entirely benign condition.

We also found that there was an interaction between age and diabetes or glycemic status in analyses for risk of incident dementia, whereas the interaction between sex and diabetes or glycemic status was significant for all-cause and AD but not for VD. Even though older age and female sex are known risk factors of AD, their relationships with glycemic status were not simply linear. We suggest that future studies should also collect clinical data in hospitals from outside a claims database and then link the two. Such a linkage of multiple data sources would allow evaluation of the complex contributions to incident dementia according to interactions with age/sex and various glycemic conditions.

This study had some limitations. First, because the study was based on analysis of a claims database, the possibility of registering inaccurate diagnostic codes cannot be ruled out. However, a validation study in South Korea compared diagnoses in this claims database with the actual diagnoses in hospital records. The study found an overall positive predictive value for accuracy of diagnoses of 83.4% (40). Hence, the claims database could be generally accepted as accurate for most diseases. In particular when serious diseases such as dementia are diagnosed, the diagnostic codes have likely been carefully entered.

Second, in this nationwide study, we defined glycemic status using only FPG levels. To include as many subjects as possible, we inevitably had to define hyperglycemia according to results of routine health examinations that did not contain oral glucose tolerance tests or HbA_1c measurement. It is also necessary to consider using multiple definitions of hyperglycemia, including HbA_1c, in future studies.

Third, although we considered a lag time for diagnosis of dementia, still there may have been reverse causality, especially in the case of AD due to its long prodromal period.

Lastly, there is a potential for residual confounding because various characteristics (e.g., cognitive function, educational level, and family history of dementia) were unmeasured. We tried to overcome the limitations of claims data studies by using the health examination database with anthropometric and laboratory data. Notwithstanding these limitations, this study was based on a cohort representative of the entire population of South Korea. Until now, large-scale cohort studies such as ours are still rare.

Observational data cannot establish causality. However, our results were consistent with the hypothesis that hyperglycemia and other comorbidities are associated with acceleration of the brain aging process, leading to an increased risk of dementia. More long-term follow-up studies are needed to determine how long-term exposure to hyperglycemia increases the risk of developing dementia. Because the use of medical services can vary depending on individual health behaviors and/or underlying disease, there may be differences between the results of our study and results of studies using community- or hospital sample-based cohorts. Even though we found that degrees of hyperglycemia and its comorbidities were associated with increased risks of incident dementia, associations between prediabetes and specific cognitive abilities remain unclear.

In conclusion, the results of this population-based study suggest that even a mild degree of hyperglycemia and diabetes and other comorbidities could be important risk factors contributing to the incidence of dementia, including AD and VD. These findings strongly support the American Diabetes Association’s recommendations that screening for prediabetes using an informal assessment of risk factors should be considered for asymptomatic adults; screening should also be performed for dementia prevention. It is also important for clinicians to know that prediabetes and diabetes are not discrete conditions; they are part of a hyperglycemia continuum that ends with diabetes. More active intervention in patients with prediabetes and other comorbidities can help improve population-level health outcomes in an aging society.

Acknowledgments. Editorial services were provided by Caron Modeas, Evolved Editing, LLC.

Funding. This study was supported by a faculty research grant of Yonsei University College of Medicine (6-2020-0148).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. W.J.K. wrote the manuscript and researched data. S.J.L. wrote the draft of the manuscript. E.L. reviewed the manuscript. E.Y.L. contributed to the discussion and reviewed the manuscript. K.H. researched data and contributed to the results. K.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.