OBJECTIVE

Type 2 diabetes is an established risk factor for dementia. However, the roles of glycemic control and diabetic complications in the development of dementia have been less well substantiated. This large-scale cohort study aims to examine associations of longitudinal HbA1c levels and diabetic complications with the risk of dementia incidence among patients with type 2 diabetes.

RESEARCH DESIGN AND METHODS

Data of eligible patients with diabetes, aged ≥50 years in the U.K. Clinical Practice Research Datalink from 1987 to 2018, were analyzed. Time-varying Cox regressions were used to estimate adjusted hazard ratios (HRs) and 95% CIs for dementia risk.

RESULTS

Among 457,902 patients with diabetes, 28,627 (6.3%) incident dementia cases were observed during a median of 6 years’ follow-up. Patients with recorded hypoglycemic events or microvascular complications were at higher risk of dementia incidence compared with those without such complications (HR 1.30 [95% CI 1.22–1.39] and 1.10 [1.06–1.14], respectively). The HbA1c level, modeled as a time-varying exposure, was associated with increased dementia risk (HR 1.08 [95% CI 1.07–1.09] per 1% HbA1c increment) among 372,287 patients with diabetes with postdiagnosis HbA1c records. Similarly, a higher coefficient of variation of HbA1c during the initial 3 years of follow-up was associated with higher subsequent dementia risk (HR 1.03 [95% CI 1.01–1.04] per 1-SD increment).

CONCLUSIONS

Higher or unstable HbA1c levels and the presence of diabetic complications in patients with type 2 diabetes are associated with increased dementia risk. Effective management of glycemia might have a significant role in maintaining cognitive health among older adults with diabetes.

The number of older adults living with Alzheimer disease (AD) and other forms of late-onset dementia (LOD) is increasing exponentially, in parallel with increases in life expectancy and population ageing, across the globe (1,2). Type 2 diabetes is another highly prevalent chronic disease in late life and is a well-established risk factor for dementia (3). Three meta-analyses of previous cohort studies showed a 53–73% higher risk of LOD or AD in patients with diabetes compared with subjects without diabetes (46). Furthermore, several reports have linked type 2 diabetes with the presence of specific AD biomarkers, such as cerebrospinal fluid phosphorylated tau and total tau (7), reduced brain volume, and fluorodeoxyglucose uptake (7,8).

Most previous epidemiological studies mainly focused on defining the increased risk of dementia associated with diagnosis or presence of diabetes. However, the biological mechanisms underlying this relationship and the role of glycemic control and diabetic complications in dementia development are less well understood (9,10). Reports on plasma glucose or glycosylated hemoglobin A1c (HbA1c) levels and cognitive outcomes had contradictory results (1114), probably limited by relatively small sample sizes, variable length of follow-up, or reverse causality bias. Of note, HbA1c variability, an important indicator of long-term control of diabetes, has recently been found to be positively associated with micro- and macrovascular complications and mortality in patients with diabetes, independently of HbA1c levels (15). The effect of HbA1c variability on dementia risk is less clear.

With regard to diabetic complications, extensive evidence suggests that hypoglycemia, a common acute complication of diabetes treatment, is associated with adverse cognitive outcomes among older adults (1618). In contrast, other prevalent diabetic complications resulting from microvascular lesions, such as diabetic retinopathy, nephropathy, and neuropathy, have drawn less attention with respect to dementia risk.

Our study aims to comprehensively evaluate the associations of longitudinal HbA1c levels and their long-term variability, as well as of diabetic complications, with dementia incidence in a large cohort of older patients with type 2 diabetes, leveraging electronic health record (EHR) data from the U.K. Clinical Practice Research Datalink (CPRD) (19).

Data Sources

The U.K. CPRD GOLD database is a primary care database that includes ongoing longitudinal collection of fully coded EHRs of >17 million individuals who are (or were) registered with >700 participating general practitioner (GP) practices in the U.K. (19). The available data include symptomatology, clinical diagnosis, results of investigations, prescriptions, secondary care referrals, and vaccinations. CPRD was linked to secondary care data, such as the Hospital Episode Statistics (HES), mortality data from the Office for National Statistics (ONS), and regional data on measures of social deprivation. The demographic profiles of the patient population in CPRD are similar to those of the general population of the U.K. (19).

Study Population

Individual-level data between 1987 and 2018 were extracted for this study. Patients were included if they were aged ≥50 years at any point during their CPRD registration period and had a diagnosis of diabetes, based on relevant CPRD Medcode or a prescription of antidiabetes drugs (oral hypoglycemic agents or insulin) (Supplementary Table 1). In addition, eligible participants were required to have been registered in CPRD for at least 1 year prior to diabetes onset to ensure that the date of newly diagnosed diabetes was captured and to allow time for baseline information to be recorded. Patients with a diagnosis of type 1 diabetes or those who had a diagnosis of diabetes or initiation of treatment prior to 30 years of age were excluded. Patients were also excluded if they had a diagnosis of dementia before cohort entry. To account for reverse causality bias (i.e., that prodromal cognitive/functional impairment prior to dementia diagnosis could result in poorer management of diabetes), those who developed dementia or died during the first 2 years after cohort entry or the onset of diabetic complication were excluded from the analyses of diabetic complications and dementia risk; those who developed dementia or died during the first 2 years following the first post–diabetes diagnosis HbA1c record after 50 years of age were excluded from the analyses of HbA1c levels and dementia risk.

A total of 457,902 individuals fulfilled the inclusion and exclusion criteria and were included in the analysis on diabetic complications and dementia risk. Among these participants, 372,287 individuals had at least one HbA1c record at ≥50 years of age and post–diabetes diagnosis and were included in the analysis of longitudinal HbA1c levels and dementia risk.

Exposure Assessment

Episode of hypoglycemia, microvascular diabetic complications such as nephropathy, retinopathy, and neuropathy (including diabetic foot), and other complication events (such as coma, ulcer, and unspecified records of diabetic complications) were extracted using the corresponding CPRD Medcode (Supplementary Table 2). To comprehensively identify hypoglycemia episodes, we used codes of severe hypoglycemia, hospital-treated hypoglycemia, hypoglycemia without coma, and unspecified hypoglycemia (17). The date of onset of a specific type of complication was defined according to its first relevant health record. The date of onset of overall microvascular complications was defined as the earliest date of developing nephropathy, retinopathy, or neuropathy.

Longitudinal HbA1c concentrations were recorded as test results and extracted using CPRD Medcode and Enttype code (Supplementary Table 3). The HbA1c value, measured as a continuous variable, was also stratified into different clinically established categories (<6%, 6–7%, 7–8%, 8–9%, 9–10%, and ≥10%). For patients with at least three HbA1c records during the first 3 years of follow-up, mean and coefficient of variation (CV; the ratio of the SD to the mean) of the 3-year HbA1c measurements were estimated and assessed as additional exposure variables, reflecting the average level and variability of long-term HbA1c concentrations.

In addition, information on the following covariates was extracted: age at cohort entry, sex, calendar year of cohort entry, region in U.K., index of multiple deprivation (IMD; a proxy of socioeconomic status linked to CPRD), BMI (latest record up to 10 years before cohort entry to reduce missing values), smoking status (latest record up to 5 years before cohort entry), duration of diabetes at cohort entry (based on the first clinical record of diabetes diagnosis), history of antidiabetes treatment, and history of major comorbidities, including chronic heart disease, stroke, hypertension, chronic kidney disease, chronic obstructive pulmonary disease, and cancer.

Outcome Ascertainment

The outcome event was dementia incidence. We did not distinguish by specific LOD type, as such granular level of data are variably registered in CPRD, and the precision of health data recording varies over time. Moreover, it is acknowledged that most cases of LOD involve mixed brain pathologies (20,21). Patients were considered to have dementia if they: 1) had a dementia diagnosis based on Medcode in CPRD; 2) had a dementia diagnosis based on ICD codes in linked HES or ONS databases; or 3) had at least one dementia-specific drug prescription (donepezil, galantamine, rivastigmine, or memantine [22]) (Supplementary Table 4). Patients with dementia with diagnoses of unrelated etiologies, such as following HIV infection, Creutzfeldt-Jakob disease, or alcohol- and drug-induced, were excluded. Among the extracted subjects with dementia, 96% were based on diagnosis code, and 4% were based on dementia-specific drug prescription. The outcome event date was defined as the first dementia diagnosis date or the first prescription date of dementia-specific drugs, whichever occurred earlier.

Statistical Analysis

Distributions of baseline characteristics were summarized and compared between subcohort patients with baseline HbA1c levels <7% (53 mmol/mol) and ≥7%. Time-varying Cox proportional hazards models, with age as the underlying time-scale, were used to estimate hazard ratios (HRs) and 95% CIs of dementia associated with diabetic complications or longitudinal HbA1c levels in separate analyses. Exposures were treated as time-varying variables.

In analyses of diabetic complications, the presence of hypoglycemic episodes and microvascular complications in aggregate and by type (nephropathy, retinopathy, and neuropathy) in association with dementia incidence were examined in separate Cox models. Patients who developed a relevant diabetic complication during follow-up contributed person-years to the no-complication group up until the complication diagnosis date and then contributed person-years to the complication group. Time of cohort entry for each patient was defined as date of diabetes onset, aged 50 or 1 January 1987, whichever was the latest. The end of follow-up was defined as date of dementia incidence, death, transfer-out date, last data collection date of GP practice, or 1 May 2018, whichever occurred first. The patient-level transfer-out date recorded in CPRD refers to the date the patient transferred out of the practice. The practice-level last data collection date refers to the date of the latest data upload from each GP practice (19). To examine independent effects of different types of microvascular complications, an additional analysis mutually adjusting for nephropathy, retinopathy, and neuropathy was conducted.

For the analysis of longitudinal HbA1c level, given that glycemic levels change over time for each patient with diabetes, the time-varying Cox model was used to estimate the HR and 95% CI of dementia incidence per 1% increment (absolute value) of HbA1c among patients with at least one HbA1c record post–diabetes diagnosis after 50 years of age. In a separate Cox model, time-varying HbA1c category (<6%, 6–7%, 7–8%, 8–9%, 9–10%, and ≥10%) was assessed as the exposure variable to explore the potential nonlinear relationship with dementia risk, with 6–7% as reference group. The beginning of follow-up for this cohort was defined as date of the first postdiagnosis HbA1c record after 50 years of age (i.e., baseline HbA1c). The end of follow-up was defined as above. To account for reverse causality bias, HbA1c concentrations recorded within 2 years prior to dementia incidence or death were excluded.

For the analysis of long-term average level of HbA1c and its variability, the 3-year mean and CV of HbA1c were simultaneously entered into a conventional Cox model to estimate their independent associations with dementia incidence. Follow-up time for this analysis was calculated from 3 years after the baseline HbA1c record (to avoid concurrent bias), until the date of dementia incidence or censoring time. Patients who developed dementia or died during the first 2 years of follow-up were excluded. In addition, mean and CV of HbA1c were both modeled as categorical variables (<6%, 6–7%, 7–8%, 8–9%, 9–10%, and ≥10% for mean value; quartiles for CV) in a separate Cox model.

To account for potential confounding factors, three sequential models with increasing levels of adjustment for covariates were created for all analyses: model 1 adjusted for age, sex, calendar year, and region; model 2 further adjusted for IMD (in quintile), smoking status (nonsmoker, current smoker, ex-smoker, or missing), BMI category (<25, 25–30, and ≥30 kg/m2 or missing), and history of comorbidities; and model 3 additionally adjusted for diabetes-related factors, including duration of diabetes, presence of diabetic complications (only for HbA1c analysis), baseline HbA1c level (only for diabetic complications analysis), and prescription of antidiabetes drugs (no drug, only oral hypoglycemic drug, or insulin). Covariates were also modeled as time-varying variables and updated at complication diagnosis date or each HbA1c record date during follow-up. Missing values in smoking status and BMI category during follow-up were imputed with last observation carried forward.

We further repeated the main analyses in males and females separately and tested the effect modification by sex. Several sensitivity analyses were conducted to assess the robustness of our findings: 1) restricting to participants who were at least 60 years old at cohort entry; 2) restricting to those who were at least 80 years old at the end of follow-up to account for the competing risk of premature death, in which the estimated HR reflects relative hazard of dementia conditional on patients surviving beyond 80 years of age; 3) restricting to participants who entered cohort after 2004, as diabetes data quality was significantly improved in CPRD following the introduction of the Quality and Outcomes Framework indicators for diabetes, and to account for the change of clinical practice and guidelines of diabetes management over time (23); 4) not excluding HbA1c recorded within 2 years prior to dementia incidence or death in the analysis of longitudinal HbA1c; 5) excluding HbA1c records within 5 years prior to dementia incidence or patients who developed dementia within the first 5 years of follow-up to further reduce reverse causality; 6) excluding possible outlier values in HbA1c records (<4% or >12%); 7) adjusting the 3-year CV of HbA1c for the possible influence of number of HbA1c records (24) (adjusted CV = CV/[total records during 3-year follow-up/total records during 3-year follow-up − 1]1/2) and also stratifying the analysis by the median number of HbA1c records; 8) additionally adjusting for the average number of clinical visits per year during follow-up of each patient, since patients who visited a GP more frequently may have a systematically different profile or higher diagnosis rate; 9) additionally adjusting for the identifier of GP practices to account for the practice group variability; and 10) using a more stringent dementia ascertainment criterion that requires at least two dementia diagnosis records within or between data sources or at least one dementia diagnosis record plus one dementia-specific drug prescription (25).

The statistical analyses were performed using Stata (version 15; Stata Corp.). All statistical tests were two-sided, and the significance level was defined as P < 0.05.

Data and Resource Availability

This study is based on data from the CPRD obtained under license from the U.K. Medicines and Healthcare Products Regulatory Agency (protocol approved by the Independent Scientific Advisory Committee, number 19_065R). According to the U.K. Data Protection Act, information governance restrictions (to protect patient confidentiality) prevent data sharing via public deposition. Data extracts can be requested by applying to the CPRD (https://www.cprd.com).

Baseline Characteristics of Study Population

Of the 457,902 patients with type 2 diabetes, 52.1% were male; the mean baseline age was 64.5 (SD 10.8) years (Table 1). At cohort entry, 42.3% of patients had obesity (BMI ≥30 kg/m2), and 19.0% were self-reported current smokers; 17.1% had been prescribed antidiabetes drugs, and the mean baseline HbA1c level was 7.4% (57 mmol/mol) (SD 2.1%). Prior to baseline, only 1,986 (0.4%) patients had hypoglycemic events, and 8,060 (1.8%) had microvascular complications (5,928 subjects with retinopathy, 2,275 with neuropathy, and 569 with nephropathy). During follow-up, 17,524 (3.8%) patients developed hypoglycemic episodes, and 103,188 (22.5%) patients developed microvascular complications (73,615 with retinopathy, 41,920 with neuropathy, and 8,660 with nephropathy).

Among the 372,287 patients included in analyses on longitudinal HbA1c, 216,600 (58.2%) had HbA1c levels <7% (53 mmol/mol) at baseline (Table 1). Compared with patients with HbA1c <7%, those with HbA1c ≥7% were slightly younger, entered the cohort earlier, presented with less comorbidities, and were more likely to be male, deprived, obese, and current smokers; they also had a longer duration of diabetes and higher proportion of antidiabetes drug use and diabetic complications before baseline (P < 0.05). Patients with post–diabetes diagnosis HbA1c record were slightly more likely to have obesity, chronic heart disease, and hypertension at baseline than those without (Supplementary Table 5).

Diabetic Complications in Association With Dementia Risk

During a median of 6 years’ follow-up (ranging from 0–31 years) of the 457,902 patients with diabetes, 28,627 (6.3%) incident dementia cases were recorded. After adjusting for a full set of covariates (model 3), there was evidence for an association between hypoglycemic events and a higher risk of dementia incidence (HR 1.30 [95% CI 1.22–1.39]) (Table 2). The HR estimates were 1.50 in model 1 and 1.44 in model 2 (P < 0.05).

Microvascular complications were also associated with a higher risk of dementia incidence in the fully adjusted model (HR 1.10 [95% CI 1.06–1.14]) as well as in models 1 and 2 (HR 1.22 and 1.21, respectively). Neuropathy and nephropathy had relatively stronger association with dementia incidence (HR 1.25 [95% CI 1.18–1.33] and 1.23 [1.13–1.33]; model 3) than retinopathy (HR 1.07 [95% CI 1.03–1.11]; model 3). The analysis mutually adjusting for neuropathy, nephropathy, and retinopathy revealed similar results; the HRs were 1.25 (95% CI 1.18–1.33), 1.24 (95% CI 1.14–1.35), and 1.06 (95% CI 1.02–1.10), respectively.

Longitudinal HbA1c Level in Association With Dementia Risk

During a median of 6 years’ follow-up (ranging from 0–30 years) of the 372,287 patients with diabetes with postdiagnosis HbA1c data, 23,746 (6.4%) incident dementia cases were recorded. HbA1c level was significantly associated with higher risk of dementia incidence, with an HR of 1.08 (95% CI 1.07–1.09) per 1% increment of HbA1c in the fully adjusted model and 1.13 or 1.14 in models 1 and 2, respectively (Table 3). In a separate analysis in which time-varying HbA1c was modeled as a categorical variable with 6–7% as reference group, patients with well-controlled HbA1c (<6%) had lower risk of dementia incidence (HR 0.86 [95% CI 0.83–0.89]), while those with HbA1c levels of 8–9%, 9–10%, and ≥10% had 15% (95% CI 9–21), 26% (95% CI 17–34), and 40% (95% CI 32–49) increased risk of dementia incidence, respectively (Fig. 1).

Long-Term Mean and Variability of HbA1c in Association With Subsequent Dementia Risk

Consistent with the results of time-varying HbA1c analysis, the mean value of 3-year HbA1c measurements was significantly associated with higher risk of subsequent dementia incidence after controlling for HbA1c variability, with an HR of 1.04 (95% CI 1.02–1.06) per 1% increment of HbA1c in model 3 and 1.05 in models 1 and 2 (Table 3). Compared with patients who had a mean HbA1c level at 6–7%, those with mean HbA1c levels of 8–9%, 9–10%, and ≥10% had 9% (95% CI 3–16), 18% (95% CI 8–28), and 30% (95% CI 17–44) increased risk of dementia incidence (model 3), respectively.

The 3-year CV of HbA1c, which reflects the long-term glycemic variability, was also independently associated with higher risk of dementia incidence. After controlling for the 3-year mean HbA1c level, the HRs were 1.03 (95% CI 1.01–1.04) per 1-SD increment of CV in model 3 and 1.02 in models 1 and 2 (Table 3). Compared with patients in the lowest quartile (Q1) of CV, the Q2, Q3, and Q4 groups had 6% (95% CI 1–11), 12% (95% CI 6–18), and 13% (95% CI 6–19) of increased risk of dementia incidence, respectively (Fig. 1).

The subgroup analyses by sex (Supplementary Table 6) showed a stronger association between hypoglycemia and dementia risk in men (HR 1.39) than in women (HR 1.23; Pinteraction = 0.002). Results of all sensitivity analyses were similar to those of the main analyses (Supplementary Table 7).

This is the largest cohort study to comprehensively evaluate the association between longitudinal diabetes control and subsequent dementia risk among patients with type 2 diabetes. Our results strengthen the previous evidence on the association between hypoglycemic events and increased risk of dementia incidence among patients with diabetes. Moreover, the presence of microvascular complications, as well as high and unstable HbA1c levels during follow-up, are also found to be independent risk factors for incident dementia.

Our results on hypoglycemic complications are in line with evidence from previous cohort studies. An EHR-based cohort study of 16,667 patients with type 2 diabetes showed that those with at least one severe hypoglycemic episode had higher risk of dementia than patients without such record (16). Another cohort study, using the earlier version of the U.K. CPRD (from 2003 to 2012) with 53,055 patients with type 2 diabetes, demonstrated that hypoglycemia was associated with a higher risk of subsequent dementia incidence (17). In contrast, as patients with dementia may have reduced functional ability to achieve proper diabetes management, established dementia may in turn lead to frequent hypoglycemic episodes or other diabetic complications. A meta-analysis (18) of nine observational studies revealed a bidirectional association between hypoglycemic episode and dementia. Therefore, we have used a 2-year lag in the main analysis and 5-year lag in the sensitivity analyses to account for reverse causality bias. It has been postulated that hypoglycemia can disrupt the cerebral glucose metabolism and thus contribute to neuropathological changes (26). Previous experimental studies have shown that severe hypoglycemia could trigger apoptosis and neuronal loss (26). In this regard, hypoglycemia may not only serve as a reflection of poor management or progression of diabetes, but may also be directly implicated in the neurodegenerative process.

In terms of the relationship between microvascular diabetic complications and dementia risk, there has been one report on a predictive risk score of 10-year dementia risk for patients with type 2 diabetes (27). Among the 45 candidate predictors, the 8 strongest predictors were selected into the final prediction model, including microvascular disease (diabetic retinopathy/end-stage renal disease), diabetic foot, and acute hyper-/hypoglycemic events. Our findings further confirm the associations of overall and individual microvascular complications with risk of dementia incidence. It is likely that such complications merely represent peripheral markers of cerebral microvascular lesions (28) rather than contributing to the development of dementia through direct biological mechanisms. For instance, the retina is traditionally seen as an accessible extension of the central nervous system and may reflect possible cerebral pathology, such as cerebral small vessel disease, resulting from poor diabetes control and chronic hyperglycemia. In this context, our observed relatively weak association between retinopathy and dementia risk is intriguing. On one hand, our findings are in line with previous reports of weak direct evidence on associations between retinal microvascular changes and dementia incidence (29). For example, a report based on the Rotterdam Study found no association between retinopathy and dementia incidence (30). On the other hand, our effect estimate might have been attenuated due to underdiagnosis, as retinopathy may remain asymptomatic and undetected in the absence of routine ophthalmoscopic screening.

Our results of time-varying HbA1c levels and long-term average HbA1c levels confirm previous reports based on observational studies of much smaller sample sizes. A prospective cohort study monitored the glucose levels and dementia incidence of 2,067 older adults for a median of 6.8 years (11). Among participants with diabetes at baseline, elevated average glucose levels were associated with a higher risk of dementia. A 9-year cohort study also demonstrated that elevated longitudinal glucose level was associated with worse cognitive performance among 717 older adults with diabetes (12). A previous cohort study in participants with and without diabetes found a J-shaped association between baseline HbA1c and incident dementia (31). In contrast, our results indicate a linear HbA1c–dementia relationship among patients with diabetes, probably because we adjusted for hypoglycemic events or because the majority of patients in our sample had HbA1c levels >5.5% (37 mmol/mol). However, another cohort study of 2,246 older adults with type 2 diabetes showed that, for those with high or moderate HbA1c at baseline, a large reduction in HbA1c within the 1st year of follow-up was associated with higher incidence of dementia during follow-up (13). A pioneering large randomized controlled trial (32) tested the effects of intensive glycemic control treatment strategies (target HbA1c <6.0%) versus standard-care guidelines (HbA1c <7.0–7.9%) on cognitive outcomes in patients with type 2 diabetes. Although the intensive glycemic therapy group had larger total brain volume at 40 months and a slower rate of gray matter loss, no significant between-group difference in the rate of clinical cognitive decline was observed (32,33). Future adequately powered randomized controlled trials on long-term glycemic control and dementia prevention in patients with diabetes are further warranted.

Our findings also suggest that glycemic variability, an acknowledged indicator for diabetes control and predictor for mortality in older adults (15), is also independently associated with dementia risk. Currently, in the absence of an established “gold standard” measure, the CV has been recommended as a robust marker for glycemic variability (34). In accordance with our findings, a retrospective cohort study of 16,706 Chinese patients with type 2 diabetes suggested that CVs for fasting plasma glucose and HbA1c, calculated during the 1st year of follow-up, were associated with increased risk of AD incidence (35). In this regard, a high and unstable glycemic level may contribute to dementia pathogenesis either through cerebrovascular lesions (28) and cerebral metabolic dysregulation (36) or as an indicator of the underlying “brain insulin resistance,” shown to be associated with the AD pathological signature of brain amyloidosis, tau accumulation, and neurodegeneration (AT[N]) (10).

Studies based on EHRs are subject to potential sources of bias, which we have attempted to account for both in study design and through several sensitivity analyses. Firstly, selection bias is an acknowledged risk for studies based on real-world EHR data. In this study, this risk was mitigated by the use of a national database including data from virtually all patients registered in 8% of primary care/GP practices across the U.K. CPRD covers a wide range of urban and rural areas of diverse socioeconomic strata and has been shown to be well representative of the general population (19). Another possible limitation relates to the potential underreporting or misdiagnosis of dementia cases in EHR data sets. We maximized dementia case ascertainment by data linkage with HES and ONS databases (secondary diagnosis and cause of death). Moreover, assuming that the remaining misclassification in outcome events is independent of exposure variables, our estimates of association (i.e., HRs) are likely to have been biased toward the null rather than overestimated (37). To account for changes in dementia diagnostic criteria and the increasing diagnosis rate of dementia over time (38), we have adjusted for calendar year in all analyses and conducted a sensitivity analysis restricted to data collected after 2004, all of which produced consistent results. Future well-powered clinical studies, allowing for deep phenotyping and biomarker-based characterization of patients with dementia, are warranted to elucidate the precise contribution of diabetes-related factors in the pathogenesis of AD and other LOD forms.

EHR-based studies are also vulnerable to information bias in exposure assessments. For example, although hospital admissions or emergency department attendances following severe hypoglycemia are typically recorded in GP practices through referral letters or hospital discharge notes, some instances may be missed. Another possible limitation is that measurement of HbA1c may not be regularly conducted by GPs for each patient, which may potentially affect the calculation of the CV of HbA1c. To overcome such limitations, a paradigm shift in data collection methods is warranted, as lack of real-time clinical data is a shared limitation of EHR-based and prospective cohort studies. Future real-time collection of clinical data using patient-administered portable and wearable devices and real-time reporting of adverse outcomes can improve disease management and data quality for research purposes. Furthermore, the possibility of residual confounding bias cannot be ruled out. For instance, there is little information on education level and on physical/social activities in CPRD, which are known risk factors for dementia (3,39). There were 7–19% missing values for IMD, BMI, and smoking status, but little difference was observed for effect estimates in models with and without adjustment for these covariates, suggesting that the effect of residual confounding is likely small. Finally, causality cannot be established from the results of our epidemiological analyses. Further mechanistic studies are required to elucidate the precise biological mechanisms underpinning the effect of poor diabetic control on dementia risk in older adults.

This CPRD-based study has several strengths, including the large sample size (457,902 patients with diabetes and 28,627 incident dementia cases), the long follow-up period of up to 30 years, and the broad age distribution of participants. We were also able to examine multiple indicators of diabetes control (HbA1c level and its variability, hypoglycemia, and microvascular complications) in relation to dementia risk and have adjusted for a large set of potential confounding factors. In addition, we have carefully accounted for reverse causality bias and conducted multiple sensitivity analyses to address issues such as data quality, change of clinical practice over time, and accuracy of dementia diagnosis, the results of which supported the robustness of our main findings.

In conclusion, this large-scale cohort study provides strong evidence that higher or unstable HbA1c levels and the presence of diabetic complications in patients with type 2 diabetes are associated with higher dementia incidence. Given the lack of effective therapies for AD and other LOD forms and the long preclinical disease stage of progressively accumulating pathologies prior to clinical disease onset, the effective management of modifiable risk factors and conditions, such as type 2 diabetes (39,40), may have potential value in reducing the burden of cognitive and functional decline and dementia in the elderly population.

This article contains supplementary material online at https://doi.org/10.2337/figshare.14484852.

Funding. B.Z. was supported by the Imperial College London-China Scholarship Council scholarship.

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

Author Contributions. B.Z. and L.M. contributed to study design. B.Z. carried out data analysis and drafted the first version of the manuscript. B.S. contributed to data analysis. B.Z., I.T., S.A.-A., and L.M. contributed to data interpretation. All authors critically reviewed and edited the manuscript. B.Z. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 145th Annual Meeting of American Neurological Association, Virtual, on 9 October 2020.

1.
Alzheimer’s Association
.
2020 Alzheimer’s disease facts and figures
.
Alzheimers Dement
2020
;
16
:
391
460
2.
Ahmadi-Abhari
S
,
Guzman-Castillo
M
,
Bandosz
P
, et al
.
Temporal trend in dementia incidence since 2002 and projections for prevalence in England and Wales to 2040: modelling study
.
BMJ
2017
;
358
:
j2856
3.
Livingston
G
,
Huntley
J
,
Sommerlad
A
, et al
.
Dementia prevention, intervention, and care: 2020 report of the Lancet Commission
.
Lancet
2020
;
396
:
413
446
4.
Gudala
K
,
Bansal
D
,
Schifano
F
,
Bhansali
A
.
Diabetes mellitus and risk of dementia: a meta-analysis of prospective observational studies
.
J Diabetes Investig
2013
;
4
:
640
650
5.
Zhang
J
,
Chen
C
,
Hua
S
, et al
.
An updated meta-analysis of cohort studies: diabetes and risk of Alzheimer’s disease
.
Diabetes Res Clin Pract
2017
;
124
:
41
47
6.
Chatterjee
S
,
Peters
SAE
,
Woodward
M
, et al
.
Type 2 diabetes as a risk factor for dementia in women compared with men: a pooled analysis of 2.3 million people comprising more than 100,000 cases of dementia
.
Diabetes Care
2016
;
39
:
300
307
7.
Moran
C
,
Beare
R
,
Phan
TG
,
Bruce
DG
,
Callisaya
ML
;
Alzheimer’s Disease Neuroimaging Initiative (ADNI)
.
Type 2 diabetes mellitus and biomarkers of neurodegeneration
.
Neurology
2015
;
85
:
1123
1130
8.
Li
W
,
Risacher
SL
,
Huang
E
;
Alzheimer’s Disease Neuroimaging Initiative
.
Type 2 diabetes mellitus is associated with brain atrophy and hypometabolism in the ADNI cohort
.
Neurology
2016
;
87
:
595
600
9.
Biessels
GJ
,
Staekenborg
S
,
Brunner
E
,
Brayne
C
,
Scheltens
P
.
Risk of dementia in diabetes mellitus: a systematic review
.
Lancet Neurol
2006
;
5
:
64
74
10.
Arnold
SE
,
Arvanitakis
Z
,
Macauley-Rambach
SL
, et al
.
Brain insulin resistance in type 2 diabetes and Alzheimer disease: concepts and conundrums
.
Nat Rev Neurol
2018
;
14
:
168
181
11.
Crane
PK
,
Walker
R
,
Hubbard
RA
, et al
.
Glucose levels and risk of dementia
.
N Engl J Med
2013
;
369
:
540
548
12.
Yaffe
K
,
Falvey
C
,
Hamilton
N
, et al
.
Diabetes, glucose control, and 9-year cognitive decline among older adults without dementia
.
Arch Neurol
2012
;
69
:
1170
1175
13.
Lee
ATC
,
Richards
M
,
Chan
WC
,
Chiu
HFK
,
Lee
RSY
,
Lam
LCW
.
Higher dementia incidence in older adults with type 2 diabetes and large reduction in HbA1c
.
Age Ageing
2019
;
48
:
838
844
14.
Mukai
N
,
Ohara
T
,
Hata
J
, et al
.
Alternative measures of hyperglycemia and risk of Alzheimer’s disease in the community: the Hisayama study
.
J Clin Endocrinol Metab
2017
;
102
:
3002
3010
15.
Gorst
C
,
Kwok
CS
,
Aslam
S
, et al
.
Long-term glycemic variability and risk of adverse outcomes: a systematic review and meta-analysis
.
Diabetes Care
2015
;
38
:
2354
2369
16.
Whitmer
RA
,
Karter
AJ
,
Yaffe
K
,
Quesenberry
CP
 Jr
,
Selby
JV
.
Hypoglycemic episodes and risk of dementia in older patients with type 2 diabetes mellitus
.
JAMA
2009
;
301
:
1565
1572
17.
Mehta
HB
,
Mehta
V
,
Goodwin
JS
.
Association of hypoglycemia with subsequent dementia in older patients with type 2 diabetes mellitus
.
J Gerontol A Biol Sci Med Sci
2017
;
72
:
1110
1116
18.
Mattishent
K
,
Loke
YK
.
Bi-directional interaction between hypoglycaemia and cognitive impairment in elderly patients treated with glucose-lowering agents: a systematic review and meta-analysis
.
Diabetes Obes Metab
2016
;
18
:
135
141
19.
Herrett
E
,
Gallagher
AM
,
Bhaskaran
K
, et al
.
Data resource profile: Clinical Practice Research Datalink (CPRD)
.
Int J Epidemiol
2015
;
44
:
827
836
20.
Schneider
JA
,
Arvanitakis
Z
,
Bang
W
,
Bennett
DA
.
Mixed brain pathologies account for most dementia cases in community-dwelling older persons
.
Neurology
2007
;
69
:
2197
2204
21.
Toledo
JB
,
Arnold
SE
,
Raible
K
, et al
.
Contribution of cerebrovascular disease in autopsy confirmed neurodegenerative disease cases in the National Alzheimer’s Coordinating Centre
.
Brain
2013
;
136
:
2697
2706
22.
National Institute for Health and Care Excellence
.
Donepezil, galantamine, rivastigmine and memantine for the treatment of Alzheimer’s disease
.
2011
.
Accessed 29 March 2021. Available from https://www.nice.org.uk/guidance/ta217
23.
Wilkinson
S
,
Douglas
I
,
Stirnadel-Farrant
H
, et al
.
Changing use of antidiabetic drugs in the UK: trends in prescribing 2000-2017
.
BMJ Open
2018
;
8
:
e022768
24.
Kilpatrick
ES
,
Rigby
AS
,
Atkin
SL
.
A1C variability and the risk of microvascular complications in type 1 diabetes: data from the Diabetes Control and Complications Trial
.
Diabetes Care
2008
;
31
:
2198
2202
25.
Pujades-Rodriguez
M
,
Assi
V
,
Gonzalez-Izquierdo
A
, et al
.
The diagnosis, burden and prognosis of dementia: a record-linkage cohort study in England
.
PLoS One
2018
;
13
:
e0199026
26.
Suh
SW
,
Hamby
AM
,
Swanson
RA
.
Hypoglycemia, brain energetics, and hypoglycemic neuronal death
.
Glia
2007
;
55
:
1280
1286
27.
Exalto
LG
,
Biessels
GJ
,
Karter
AJ
, et al
.
Risk score for prediction of 10 year dementia risk in individuals with type 2 diabetes: a cohort study
.
Lancet Diabetes Endocrinol
2013
;
1
:
183
190
28.
Sonnen
JA
,
Larson
EB
,
Brickell
K
, et al
.
Different patterns of cerebral injury in dementia with or without diabetes
.
Arch Neurol
2009
;
66
:
315
322
29.
Heringa
SM
,
Bouvy
WH
,
van den Berg
E
,
Moll
AC
,
Kappelle
LJ
,
Biessels
GJ
.
Associations between retinal microvascular changes and dementia, cognitive functioning, and brain imaging abnormalities: a systematic review
.
J Cereb Blood Flow Metab
2013
;
33
:
983
995
30.
Schrijvers
EM
,
Buitendijk
GH
,
Ikram
MK
, et al
.
Retinopathy and risk of dementia: the Rotterdam Study
.
Neurology
2012
;
79
:
365
370
31.
Rawlings
AM
,
Sharrett
AR
,
Albert
MS
, et al
.
The association of late-life diabetes status and hyperglycemia with incident mild cognitive impairment and dementia: the ARIC study
.
Diabetes Care
2019
;
42
:
1248
1254
32.
Launer
LJ
,
Miller
ME
,
Williamson
JD
, et al.;
ACCORD MIND investigators
.
Effects of intensive glucose lowering on brain structure and function in people with type 2 diabetes (ACCORD MIND): a randomised open-label substudy
.
Lancet Neurol
2011
;
10
:
969
977
33.
Erus
G
,
Battapady
H
,
Zhang
T
, et al
.
Spatial patterns of structural brain changes in type 2 diabetic patients and their longitudinal progression with intensive control of blood glucose
.
Diabetes Care
2015
;
38
:
97
104
34.
DeVries
JH
.
Glucose variability: where it is important and how to measure it
.
Diabetes
2013
;
62
:
1405
1408
35.
Li
TC
,
Yang
CP
,
Tseng
ST
, et al
.
Visit-to-visit variations in fasting plasma glucose and HbA1c associated with an increased risk of Alzheimer disease: Taiwan Diabetes Study
.
Diabetes Care
2017
;
40
:
1210
1217
36.
Biessels
GJ
,
Despa
F
.
Cognitive decline and dementia in diabetes mellitus: mechanisms and clinical implications
.
Nat Rev Endocrinol
2018
;
14
:
591
604
37.
Wang
L
,
Hubbard
RA
,
Walker
RL
,
Lee
EB
,
Larson
EB
,
Crane
PK
.
Assessing robustness of hazard ratio estimates to outcome misclassification in longitudinal panel studies with application to Alzheimer’s disease
.
PLoS One
2017
;
12
:
e0190107
38.
Donegan
K
,
Fox
N
,
Black
N
,
Livingston
G
,
Banerjee
S
,
Burns
A
.
Trends in diagnosis and treatment for people with dementia in the UK from 2005 to 2015: a longitudinal retrospective cohort study
.
Lancet Public Health
2017
;
2
:
e149
e156
39.
Norton
S
,
Matthews
FE
,
Barnes
DE
,
Yaffe
K
,
Brayne
C
.
Potential for primary prevention of Alzheimer’s disease: an analysis of population-based data
.
Lancet Neurol
2014
;
13
:
788
794
40.
Wheeler
MJ
,
Dempsey
PC
,
Grace
MS
, et al
.
Sedentary behavior as a risk factor for cognitive decline? A focus on the influence of glycemic control in brain health
.
Alzheimers Dement (NY)
2017
;
3
:
291
300
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/content/license.