Diabetes-Associated Dementia Risk and Competing Risk of Death in the Three-City Study

Abstract

Diabetes is associated with a higher dementia and mortality risk. However, few studies have accounted for death when estimating the association between diabetes and dementia. We estimated absolute and relative risks of all-cause dementia according to diabetes exposure status in older adults while accounting for competing risk of death using illness-death models. Effect modification by specific characteristics (age, gender, education, cardiovascular risk factors, body mass index, cardiovascular history, depressive symptomatology, impaired renal function, and APOE ɛ4 genotype) was also investigated. We analyzed the Three-City study data, a French population-based cohort of adults aged 65 years and above who were followed up for 12 years from 1999–2001. Among 8,328 participants selected in the analytical sample (median age, 73.3 years; 60.3% women), 809 (9.3%) presented with diabetes at baseline. Over a median follow-up period of 8.3 years, 836 participants developed incident dementia. Baseline diabetes was associated with a higher risk of dementia: hazard ratio, 1.79 [95% confidence interval, 1.46–2.19]. No effect modification was shown. Diabetes was associated with a higher 12-year absolute risk of dementia and a lower dementia-free life expectancy (e.g., 14.5% [11.2–18.1] versus 8.7% [7.6–10.2], and 13.4 [12.7–14.1] years versus 16.5 [16.0–17.1] years, respectively, for a 70-year-old woman with the highest level of education). These findings support the potential impact of preventing diabetes on reducing dementia risk in older adults, with a 2-3-year higher dementia-free life expectancy for individuals without diabetes, and inform the design of future interventional trials.

Keywords

Cohort studies death dementia type 2 diabetes mellitus

INTRODUCTION

In the absence of a curative treatment for dementia and Alzheimer’s disease (AD), preventive interventions targeting modifiable risk factors, such as type 2 diabetes, have been promoted. Type 2 diabetes is associated with a 1.5-2-fold increased risk of all-cause dementia and AD-type dementia [1, 2].

Observational studies have suggested that better glycemic control among older adults with type 2 diabetes may lower the risk of dementia [3 –6], but interventional studies targeting intensive glucose control in patients with type 2 diabetes have failed to show a clinically significant protective impact on cognitive decline after up to 6 years of treatment, as compared to usual care [7].

This drawback could partly be explained by an inaccurate estimate of the association between diabetes and dementia in observational studies. As type 2 diabetes is also a major determinant of mortality [8], death is an obvious competing event for dementia. To date, only one study in older Mexican Americans has accounted for the competing risk of death, and the results highlighted a significant 10% decrease in the association between diabetes and 10-year risk of dementia and cognitive impairment, as compared to standard Cox modeling [9]. These results suggest the importance of accounting for the competing risk of death when estimating the relative risk of dementia according to diabetes. Similarly, estimates of the absolute risk (i.e., incidence) of dementia must account for the competing risk of death to identify accurately the target population for interventional trials. As such, a difference in the absolute risk of dementia between individuals with and without diabetes would promote selecting individuals with type 2 diabetes for future interventional trials. In addition, if an intervention leads to a stronger decrease in mortality than in dementia incidence among non-demented participants, a subsequent increase in the incidence of dementia may be observed. Another interpretation of the failure of previous interventions to prevent cognitive decline is that populations targeted by the antidiabetic interventions might not be the population who could benefit. To date, characteristics such as education, physical activity, depression, and a history of cerebrovascular, cardiovascular, or chronic kidney disease have been associated with a higher risk of dementia in participants with diabetes [10 –13], while a synergistic interaction between diabetes and the apolipoprotein E (APOE) genotype has been suggested for cognitive decline but is controversial for all-cause dementia risk [14 –17].

The objectives of this study were to provide estimates of absolute and relative risks of dementia associated with type 2 diabetes, in a population-based cohort of French older adults, considering the competing risk of death. We also investigated whether the association between diabetes and dementia risk was modified by sociodemographic and clinical characteristics.

MATERIAL AND METHODS

Study population

The Three-City (3C) study is a prospective cohort study [18] of the general population. Briefly, it included 9,294 community-dwelling people aged 65 years or over, randomly invited between 1999–2001 to participate from electoral rolls in three French cities (Bordeaux [n = 2,104], Dijon [n = 4,931], and Montpellier [n = 2,259]). During the baseline face-to-face interview, a trained investigator collected data related to sociodemographic characteristics, lifestyle, medical history, and medication use. A fasting blood sample was taken, and physical and cognitive assessments were performed. During the 12-year follow-up period, five follow-up visits took place every 2-3 years and included standardized questionnaires, clinical examinations, and a detailed cognitive assessment. This research adhered to the principles of the Declaration of Helsinki. The ethics committee (“Comité Consultatif de Protection des Personnes se Prêtant aux Recherches Biomédicales”, CCPPRB) of the Kremlin-Bicêtre University Hospital and Sud-Méditerranée III (France) approved the 3C study protocol. All participants gave written informed consent.

Among the 9,078 participants initially free of dementia in the 3C study, 8,328 participants were included in this analysis after excluding participants with prevalent dementia at baseline, missing data on baseline diabetes status or missing data on adjustment covariates (gender, education, smoking status, alcohol consumption, body mass index (BMI), hypertension, or hypercholesterolemia) (Supplementary Figure 1). Compared to the excluded participants, the included participants were younger on average, more likely to be male, had a higher education level and higher professional status, had a higher average Mini-Mental State Examination (MMSE) score, and were less likely to be exposed to risk factors of dementia (Supplementary Table 1).

Diabetes exposure

Participants were classified as having diabetes at the baseline visit either in the presence of fasting blood glucose≥7 mmol/L (≥126 mg/dL) or non-fasting blood glucose≥11.1 mmol/L (≥200 mg/dL) or antidiabetic drug intake (Anatomical Therapeutic Chemical classification system: code A10A “insulins and analogues”, and code A10B “blood glucose lowering drugs, excl. insulins”) or self-reported history of diabetes. Participants with missing data relating to at least one criteria were classified as “diabetes status unknown” and, thus, were not included in the present study.

Dementia assessment

All participants from Bordeaux and Montpellier were examined by a neurologist at baseline. Due to the large number of participants in Dijon, only those who screened positively for dementia underwent further examination. Screening was considered positive if one of the following conditions were met: 1) missing MMSE score, 2) for participants with primary school educational level and lower: MMSE < 25 and Isaacs’ Set Test (total score in 4 categories at 15 seconds) < 33 or missing, or 3) for participants with secondary school educational level and higher: MMSE < 28 and Isaacs’ Set Test < 33 or missing [18].

The diagnosis of dementia at the follow-up examination was made using a standardized three-step procedure, as reported previously [18]. First, trained neuropsychologists administered an extensive battery of neuropsychological tests. Second, all participants with a suspicion of dementia, based on either their neuropsychological performance or a decline relative to a previous examination, were examined by a neurologist. Third, an independent committee of neurologists reviewed all available documentation for suspected cases of dementia and reached a consensus on the diagnosis according to the Diagnostic and Statistical Manual of Mental Disorders-Fourth Edition criteria [19]. Dementia subtyping was based on the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association criteria for AD [20], and on the National Institute of Neurological Disorders and Stroke–Association Internationale pour la Recherche et l’Enseignement en Neurosciences criteria for vascular dementia, [21]. Mixed dementia was defined as a diagnosis of AD with either cerebrovascular lesions on brain imaging or a documented history of stroke and the presence of prominent executive function deficits in addition to an AD-type cognitive profile.

We considered all incident cases that occurred during the 12-year follow-up period for the current analyses.

Vital status

Mortality and date of death were ascertained from the civil registry by systematic request for all subjects not included in follow-up visits [22].

Baseline covariates

Sociodemographic information recorded at baseline included age, gender, education (no or primary school, secondary school, high school, university), and professional status grouped into three categories: high (managers, professionals, technicians and associate professionals), intermediate (clerical, service, shop and market sale workers, armed forces), and low (all other professions including housewife). Lifestyle factors included smoking status (never, former, and current smoker), alcohol consumption (never, former, and current drinker), and regular physical activity (defined by participating in sport regularly or having at least 1 h of leisure or household activity per day, as self-reported by participants). Depressive symptomatology was assessed with the Center for Epidemiological Studies-Depression scale, using scores of≥17 in men and of≥23 in women as indicators of a clinically relevant level of depressive symptomatology. Hypertension was defined by systolic blood pressure≥140 mmHg or diastolic blood pressure≥90 mmHg, antihypertensive treatment, or self-reported history. Hypercholesterolemia was defined by total cholesterol≥7.25 mmol/L, intake of a lipid-lowering drug, or self-reported history, with a centralized measurement of baseline fasting serum total cholesterol performed using an enzymatic method [18]. BMI was categorized as < 20 kg/m² (underweight), 20–24.9 kg/m² (normal weight), 25–29.9 kg/m² (overweight), and≥30 kg/m² (obesity). History of cardiovascular disease was defined as a history of myocardial infarction, angina pectoris, or peripheral artery disease. History of stroke was ascertained based on self-report. Impaired renal function was defined by a glomerular filtration rate < 60 mL/min/1.73 m², estimated using the CKD-EPI equation [23]. The measurement methods for the APOE genotype have been described previously [24]. APOE ɛ4 status was defined as the presence of at least one ɛ4 allele versus absence.

Statistical analyses

Baseline characteristics were compared between selected and non-selected participants for the current study, and according to baseline diabetic status for the study sample. We used the chi-square test (or Fisher exact test when appropriate) and Student’s t-test (or the non-parametric Mann-Whitney-Wilcoxon test when appropriate) for categorical and continuous variables comparisons, respectively.

Age was used as the time scale in all analyses. We first estimated the association between diabetes and death over the 12-year follow-up, expressed as a hazard ratio (HR) and 95% confidence interval (CI), using a multivariable Cox proportional hazard model.

We considered competing risk of death and interval-censoring (due to intermittent follow-up visits in the 3C study and dementia diagnosis potentially missed between visits) for the estimation of the association between diabetes and dementia risk, using a semi-parametric illness-death model. This multistate model relies on three states for participants: disease-free, alive with dementia, and death, and it allows for the possibility of developing dementia between the last visit and death which is not considered in Cox modeling [23]. We used a parametric approach with Weibull baseline intensities and estimation of transition intensities between each state by likelihood maximization. The adjusted effect of diabetes on these transitions was expressed as HRs and their 95% CIs. Using the parameter estimates of this illness-death model, the following predictive parameters were computed according to diabetes status, gender, and education: 1) 12-year absolute risk of dementia; 2) life expectancy without and with dementia at a given age (mean number of years an individual who attained that age without dementia was expected to live free of dementia and with dementia, respectively). Predictions were computed for older adults at the index age of 65 years old (minimum age at inclusion) and 70 years old, to provide information on participants who may benefit from interventions to prevent dementia, i.e., several years before the dementia diagnosis.

We also specified a multivariable Cox proportional hazard model to estimate the association between diabetes and dementia risk without accounting for the competing risk of death, expressed as a HR and its 95% CI. We estimated age at dementia by the age at the midpoint of the interval between the age at the diagnostic visit and age at the previous visit. Participants who have not developed dementia during the follow-up period were censored at the age at their last visit. We considered stratification on study center to allow the baseline hazard to differ across the three study sites, but, as results were unchanged, we present the global results. The proportionality of the HR over time was evaluated by plotting the log of the cumulative hazard versus log of time, and by testing the correlation for Schoenfeld residuals with time. The assumptions were met for the models considered. An interaction term between diabetes and age (time scale) was not retained because it did not reach statistical significance in a univariable Cox model (p = 0.32).

All models were adjusted for potential confounding factors: gender and education in a basic adjusted model and, additionally, for alcohol consumption and smoking to capture lifestyle habits, BMI, hypertension, and hypercholesterolemia in an extended adjusted model. We did not adjust our analyses for depression, APOE genotype, cardiovascular history, or impaired renal function as those variables were not considered likely causes of diabetes and, thus, did not meet the definition of confounding factors. We explored residual confounding by estimating the E-value, which expresses the minimum strength of association that an unmeasured confounder needs to reach with both the exposure and the outcome to fully explain away a specific exposure-outcome association [26].

To investigate the potential effect modification by each characteristic considered, we performed stratified analyses according to age at inclusion (<75 versus≥75 years, selected because it was close to the median age in the study population, and relevant for aging studies), gender, education (no or primary school versus secondary school or higher), professional status, smoking status (current, never, and former smoker), APOE ɛ4 genotype, hypertension, hypercholesterolemia, BMI, cardiovascular history, history of stroke, depressive symptomatology, and impaired renal function. The significance of the interaction term between diabetes and the characteristic considered was evaluated using the likelihood ratio test.

Sensitivity analyses

First, we considered different definitions of the main exposure to diabetes, to explore the consistency of measures of association to dementia, by excluding a self-reported history of diabetes, or considering baseline impaired fasting glucose as a separate category of exposure, defined according to the WHO 2006 definition (i.e., fasting plasma glucose 6.1–6.9 mmol/L) [27]. We also repeated the main analysis after excluding participants with possible type 1 diabetes, identified by a self-reported age at diabetes onset < 40 years, from the population. Second, we excluded participants with dementia that occurred during the first 4 years of follow-up to decrease the likelihood of reverse causality between incipient dementia and diabetes. Third, to explore the possibility of selection bias, missing values of baseline diabetes and covariates were imputed by multiple imputation (MI) using chained equations with a fully conditional specification (15 imputed data sets), assuming missingness at random [28]. Fourth, we explored the impact of residual confounding due to baseline physical activity (defined as low if reported sports participation was less than once a week and self-reported low walking activity) by adjusting for it in a subsample of participants from Dijon and Montpellier with available and standardized data (n = 5,477 [88%] of 6,210 participants included from both centers). We also performed an additional adjustment for depressive symptomatology because of its baseline imbalance according to diabetes.

Analyses were performed using SAS software version 9.3 (SAS Institute, Cary, NC, USA), and the R v3.6.0 SmoothHazard package for illness-death models [29]. P-values were two sided, and a p-value≤0.05 was considered significant.

RESULTS

Study population

Compared to participants without diabetes at baseline, participants with diabetes (809, 9.7%) were more likely to be men, ever smokers, and to have a lower education level. They were also more likely to have hypertension, hypercholesterolemia, overweight or obesity, history of cardiovascular disease, and depressive symptomatology (Table 1). At baseline, 66.0% of participants with diabetes were using antidiabetic medications (oral antidiabetic agents, 61.8%; insulin, 5.8%). Diabetes status was based solely on self-report in 121 participants (15.0%). The median self-reported duration of diabetes was 10.3 years (interquartile range, 3.6–19.8 years).

Table 1

Comparison of baseline characteristics^a according to baseline diabetes – 3C Study, France, 1999–2012 (n = 8,328)

	Baseline diabetes		p^b
	No (n = 7,519)	Yes (n = 809)
Age (y)	73.2 (69.6–77.5)	73.5 (69.9–77.7)	0.23
Female gender	61.9	45.2	<0.0001
Education			0.0007
No or primary school	32.0	38.1
Secondary school	30.0	30.7
High school	13.6	10.6
University	24.4	20.6
Professional status			0.14
Low	27.1	30.2
Intermediate	31.3	31.0
High	41.6	38.8
Smoking status			<0.0001
Never smoker	62.0	50.3
Former smoker	32.5	43.9
Current smoker	5.5	5.8
Alcohol consumption			0.13
Current drinker	80.2	77.3
Never drinker	16.9	19.6
Former drinker	2.8	3.1
Mini-Mental State Examination score	28.0 (26.0–29.0)	27.0 (26.0–29.0)	<0.0001
Hypertension	76.0	90.8	<0.0001
Hypercholesterolemia	40.9	53.9	<0.0001
History of cardiovascular disease	8.3	17.9	<0.0001
History of stroke	3.9	5.6	0.02
Body mass index (kg/m²)			<0.0001
<20	6.2	1.6
20–24.9	43.2	26.3
25–29.9	38.7	46.0
≥30	11.8	26.1
Depressive symptomatology	13.3	17.5	0.001
Impaired renal function	18.7	19.4	0.60
Fasting blood glucose (mmol/L)	4.9 (4.6–5.2)	7.1 (5.7–8.4)	<0.0001
APOEɛ4 carrier	20.2	19.0	0.46

^aValues are median (interquartile range) for continuous variables or percentage for qualitative variables. ^bWilcoxon test for continuous variables and χ² test for qualitative variables. Missing data: Mini-Mental State Examination, 37; fasting blood glucose, 26; depressive symptomatology, 97; impaired renal function, 27; APOE genotype, 61; history of stroke: 84; professional status: 23.

Follow-up

Over a median follow-up period of 8.3 years (interquartile range, 3.6–11.0 years), 836 participants developed dementia (incidence, 1.38/100 person-years [95% CI 1.29–1.47]). The etiology of dementia was AD for 576 (68.9%) cases, mixed dementia for 99 (11.8%) cases, and vascular dementia for 65 (7.8%) cases. In addition, 2,337 participants died (mortality rate, 2.87/100 person-years [2.76–2.99]), among whom 1,999 died free of dementia (Supplementary Figure 1).

Association between baseline diabetes, mortality, and incident all-cause dementia

In the fully adjusted model, baseline diabetes was associated with a higher risk of death during the 12-year follow-up (HR, 1.52 [95% CI 1.35–1.72]). Baseline diabetes was associated with a higher risk of dementia at any time during the 12-year follow-up period when accounting for the competing risk of death and interval-censoring (HR, 1.79 [1.46–2.19]) (Table 2). The association between diabetes and dementia was similar when estimated with a Cox proportional hazards model. Baseline diabetes was associated with a higher risk of death in non-demented participants (HR, 1.45 [1.22–1.73]), but was not associated with a higher risk of death in demented participants (HR, 1.15 [0.88–1.50]). Calculation of the E-value suggested that an unmeasured confounder would need to be associated with both diabetes and all-cause dementia with a magnitude of at least 2.98 to fully explain away the observed association. The absolute risk of dementia was significantly higher for individuals with diabetes (Table 3). At index ages of 65 and 70 years, regardless of gender and education level, the life expectancy without dementia was significantly lower in individuals with diabetes compared to individuals without diabetes, whereas there was no difference in the life expectancy with dementia according to diabetes (Table 4).

Table 2

Association between baseline diabetes and incident all-cause dementia and death over a 12-year follow-up period – 3C Study, France, 1999–2012 (n = 8,328 participants, with 836 dementia cases and 2,337 deaths)

	Illness-death model		Cox model^a
	HR (95% CI)	p	HR (95% CI)	p
Model 1^b
Dementia	1.83 (1.51–2.23)	<0.0001	1.87 (1.54–2.28)	<0.0001
Mortality without dementia	1.57 (1.32–1.86)	<0.0001	/	/
Mortality with dementia	1.18 (0.91–1.54)	0.20	/	/
Model 2^c
Dementia	1.79 (1.46–2.19)	<0.0001	1.84 (1.50–2.24)	<0.0001
Mortality without dementia	1.45 (1.22–1.73)	<0.0001	/	/
Mortality with dementia	1.15 (0.88–1.50)	0.31	/	/

^aCox model: 7,702 participants with at least one follow-up visit, with 836 dementia cases. ^bAdjusted for age (as time scale), gender, and education. ^cAdjusted for age (as time scale), gender, education, smoking status, alcohol consumption, body mass index, hypertension, and hypercholesterolemia. HR, hazard ratio; 95% CI, 95% confidence interval.

Table 3

Twelve-year absolute risk in percent (and 95% confidence interval) of all-cause dementia after controlling for competing risk of death and according to age, gender, education, and diabetes*

		No or primary school	University
Women
65 years	No diabetes	6.4 (5.5–7.4)	4.3 (3.6–5.1)
	Diabetes	11.1 (8.7–13.6)	7.6 (5.6–9.8)
70 years	No diabetes	12.8 (11.2–14.3)	8.7 (7.6–10.2)
	Diabetes	20.8 (16.9–24.7)	14.5 (11.2–18.1)
Men
65 years	No diabetes	6.6 (5.6–7.6)	4.5 (3.8–5.3)
	Diabetes	11.6 (9.2–14.3)	8.0 (5.9–10.4)
70 years	No diabetes	13.6 (11.9–15.1)	9.3 (8.1–10.8)
	Diabetes	22.8 (18.6–27.3)	16.0 (12.5–19.9)

*Table should be read as follow: a 65-year old woman with university level of education (upper right cells) has a 12-year absolute risk of all-cause dementia of 4.3% (95% confidence interval, 3.6–5.1%) if without diabetes, and of 7.6% (95% confidence interval, 5.6–9.8%) if with diabetes.

Table 4

Life expectancy without and with dementia (and 95% confidence interval) in years, according to age, gender, education, and diabetes*

		No or primary school		University
		Dementia-free LE	LE with dementia	Dementia-free LE	LE with dementia
		Year (95% CI)	Year (95% CI)	Year (95% CI)	Year (95% CI)
Women
65 years	No diabetes	22.6 (22.1–23.1)	3.3 (2.9–3.7)	24.2 (23.5–24.9)	2.3 (2.0–2.8)
	Diabetes	18.9 (18.0–19.7)	3.8 (3.1–4.6)	20.5 (19.4–21.3)	2.7 (2.0–3.4)
70 years	No diabetes	18.1 (17.6–18.6)	3.2 (2.8–3.6)	19.6 (19.0–20.3)	2.3 (2.0–2.7)
	Diabetes	14.7 (13.8–15.3)	3.6 (3.0–4.4)	16.1 (15.1–16.9)	2.6 (1.9–3.3)
Men
65 years	No diabetes	19.9 (19.5–20.4)	2.0 (1.8–2.3)	20.8 (20.3–21.4)	1.3 (1.1–1.6)
	Diabetes	16.6 (15.8–17.2)	2.3 (1.9–2.8)	17.5 (16.7–18.2)	1.5 (1.2–2.0)
70 years	No diabetes	15.6 (15.2–16.1)	1.9 (1.7–2.2)	16.5 (16.0–17.1)	1.3 (1.1–1.5)
	Diabetes	12.6 (11.8–13.1)	2.1 (1.8–2.6)	13.4 (12.7–14.1)	1.4 (1.1–1.9)

*Table should be read as follow: a 65-year old woman with no education or primary school education and without diabetes (upper left cells) is expected to live free of dementia for an average of 22.6 years (95% confidence interval, 22.1–23.1), and with dementia for an average of 3.3 years (2.9–3.7). LE, life expectancy; 95% CI, 95% confidence interval.

Sensitivity analyses

Results of sensitivity analyses are described in Supplementary Table 2. Excluding a self-reported history of diabetes from the definition slightly increased the magnitude of the association between diabetes and dementia (HR, 1.85 [95% CI 1.50–2.29]). The association between diabetes and dementia risk was similar when excluding participants with baseline impaired fasting glucose from the population without diabetes. Untreated diabetes and treated diabetes were both associated with a higher risk of dementia (HR, 1.51 [1.06–2.16] and 1.94 [1.54–2.45], respectively). Participants with diabetes and a fasting glucose level≥5.71 mmol/L were suggested to have an increased risk of dementia compared to participants with diabetes and a fasting glucose level < 5.7 mmol/L. The association between diabetes and dementia risk remained similar after excluding 25 participants with possible type 1 diabetes. When excluding 246 participants (29% of observed dementia cases) who developed dementia during the first 4 years of follow-up, the association between diabetes and dementia decreased by nearly 10% (HR, 1.63 [1.28–2.09]) but remained significant. Imputation of missing baseline diabetes status and covariates yielded results consistent with the primary analysis (HR, 1.72 [1.16–2.19]). Additional adjustment for physical activity did not change the association between diabetes and dementia in the subsample of participants from Dijon and Montpellier with available data (HR, 1.43 [1.10–1.87] versus 1.44 [1.10–1.88], with and without adjustment for physical activity, respectively). Additional adjustment for depressive symptomatology yielded results similar to the primary analysis (HR, 1.80 [1.45–2.24]).

Investigation of effect modification

None of the baseline characteristics considered (age, gender, education, professional status, smoking, hypertension, hypercholesterolemia, BMI, history of cardiovascular disease, history of stroke, impaired renal function, depressive symptomatology, or APOE ɛ4 genotype) modified the effect of baseline diabetes on all-cause dementia risk (Fig. 1).

Fig.1

Adjusted association^a between diabetes and incident dementia over a 12-year follow-up period, for strata of characteristics considered (illness-death model with delayed entry on age) – 3C Study, France, 1999–2012 (n = 8,328). ^aAll models were adjusted for age (as time scale), gender, education, smoking status, alcohol consumption, body mass index, hypertension and hypercholesterolemia, except when the model was stratified on the considered variable. HR, hazard ratio; 95% CI, 95% confidence interval.

DISCUSSION

In this French population-based cohort study, diabetes at baseline was associated with an 80% increase in all-cause dementia risk over 12 years in older adults, after considering the competing risk of death and interval-censoring. Moreover, the effect of diabetes on dementia risk was not modified according to several baseline characteristics of the participants. The absolute 12-year risk of dementia was significantly higher and the dementia-free life expectancy was significantly lower for individuals with diabetes, as compared to individuals without diabetes, regardless of age, gender, and education.

Mechanisms by which type 2 diabetes contributes to dementia are not fully understood, but may involve chronic hyperglycemia and insulin dysregulation (insulin deficiency and insulin resistance), both of which are associated with dementia and accelerated cognitive decline [30, 31]. Chronic hyperglycemia is associated with the accumulation of advanced glycation end products and atherosclerosis, oxidative stress, and microvascular and macrovascular brain lesions [32]. Insulin resistance is likely to induce neurodegeneration, notably through brain hypometabolism, and may also inhibit expression of the insulin-degrading enzyme, which is involved in amyloid-β clearance in the brain [32]. This latter pathway is increasingly recognized, as diabetes is associated with brain atrophy and brain hypometabolism independently of its effect on cerebrovascular disease [33, 34].

While this study is one of the few that estimated the association between diabetes and long-term risk of dementia in older French adults, our results are consistent with previous work investigating this association in various countries. Diabetes is a consistent risk factor for dementia in various observational studies of older adults worldwide (Europe, United States, Mexico, and Asia), with a 50–100% increased risk for all-cause dementia [1, 2]. Our study results were robust to changes in the definition of diabetes, to competing risk of death and to selection and residual confounding, as evaluated by sensitivity analyses.

Our results confirm the observed association between diabetes and dementia risk in observational studies, even after accounting for the competing risk of death. We must emphasize that our results are different from a previous study investigating the association between diabetes, dementia, and competing risk of death. In the study by Mayeda et al., the association between diabetes and the 10-year risk of dementia and cognitive impairment among older Mexican-Americans decreased by more than 10% when modelled with the Fine and Gray model, as compared to standard Cox modeling [9]. Our different results cannot be fully explained by the inability of Fine and Gray modeling to consider interval-censoring. The association between diabetes and death was much higher in the study by Mayeda et al. than in the present study (HR, 2.15 and 2.12, for treated and untreated diabetes respectively; HR, 1.52 in the present study), resulting in a stronger competing risk of death on the observed association between diabetes and dementia in the first study. Taken together, the results highlight the importance of considering mortality associated with diabetes when estimating the potential impact of an antidiabetic intervention on preventing dementia. For example, any intervention leading to a greater decrease in mortality among non-demented participants than the decrease in dementia incidence may result in an increase in dementia incidence. Similarly, the observed dementia prevalence in the older population may also depend on the impact of this intervention on mortality among demented persons, as ascertained by simulation studies [35]. As a consequence, the impact of a given intervention on dementia incidence and prevalence will depend on its impact on the population-specific associations among diabetes, dementia and death.

The present study also showed that the absolute risk of all-cause dementia was substantial for older adults with diabetes, regardless of gender: between 7% (high education) and 10% (low education) from 65 to 77 years old, and between 14% (high education) and 22% (low education) from 70 to 82 years old. These estimates are similar in magnitude to the absolute 20-year risk of stroke for 65-year-old women with diabetes (11%) observed in a recent pooled analysis of US cohorts and nearly half the 20-year absolute risk of cardiovascular disease (25%) [36]. These estimates are also similar in magnitude to the estimates observed for carriers of the ɛ4/4 genotype in two prospective cohort studies of Danish older adults [37]: the absolute 10-year risk of all-cause dementia was nearly 10% from the age of 60 years and nearly 20% from the age of 70 years.

Our findings also suggest that diabetes in older adults is associated with an average of three dementia-free life years lost, whether from age 65 or 70 years, and regardless of gender and education. In the present study, the impact of diabetes was stronger than the impact of education, as lower education (no or primary school) was associated with an average of 1.5 dementia-free life years lost in women and 0.9 years lost in men, compared to higher education (university). In an analysis of the population-based Rotterdam study cohort, Wolters et al. estimated that ApoE ɛ4 presence was associated with a mean decrease of dementia-free life expectancy at age 65 of 2.4 years in women, and of 1.6 years in men (similar results at age 70 years) [38]. Taken together, these findings support the importance of diabetes in dementia risk.

In the current study, we did not find any effect modification for the association between diabetes and dementia risk for the characteristics considered. Specifically, we did not find an interaction between gender and diabetes on dementia, which is consistent with a recent meta-analysis of 14 observational studies representing more than 2 million participants [2]. Our study sample did not suggest an effect modification by the APOE ɛ4 genotype of the association between diabetes and all-cause dementia risk, while previous results were inconclusive [14, 15]. Depression or depressive symptoms have already been suggested to interact with diabetes on the risk of cognitive decline and dementia [12, 39], but we did not find an effect modification by depressive symptomatology in the present study. Chronic depression has been suggested to reduce self-care and medication adherence in older adults with diabetes [12], but we did not observe a worse fasting glucose level at baseline when comparing participants with diabetes according to depressive symptomatology (data not shown). In summary, the current results did not allow us to identify a high-risk population that should be targeted to decrease the effect of diabetes on dementia risk, even if we acknowledge a potential lack of statistical power. This finding is consistent with the results observed in the ACCORD-MIND trial, which compared global cognitive decline, assessed by the MMSE, between participants with diabetes receiving either intensive glycemic treatment targeting glycated hemoglobin (HbA1c) < 6.0% or standard glycemic treatment targeting HbA1c to 7.0–7.9% [40]. No significant difference was found between interventional arms, and no subgroup differences were detected according to a history of cardiovascular disease, gender, diabetes duration, age or baseline psychomotor speed.

The present study had several strengths. First, this is the first study to model the association between diabetes and dementia while accounting simultaneously for the competing risk of death and interval-censoring. We used an illness-death model to consider the interval-censored data, as it has been demonstrated that using a Cox model with censorship of dead participants at the last visit without dementia can induce a biased estimate of the association between a given exposure and dementia, with direction and magnitude of the bias depending on the effects of the exposure on dementia and death [23]. In addition, accounting for competing risk of death avoids overestimating the absolute risk of dementia and helps to accurately identify participants at-risk for dementia [23]. As described above, the ability to consider the competing risk of death is of the utmost importance in a study of late-life outcomes, such as dementia. Furthermore, the potential impact of a competing risk of death cannot be predicted as it varies according to exposure-death, exposure-dementia, and dementia-death associations. Thus, we encourage future researchers to use the best available methods to consider the competing risk of death when relevant, and to take advantage of the widespread availability of statistical software facilitating these analyses. Second, we adjusted our analyses for major confounding factors (age, education, lifestyle factors, and comorbidities). Physical activity was not standardized for the whole analytical sample; the sensitivity analysis performed on the subsample of participants with available self-reported measurement suggested no significant impact of this additional adjustment, even though this subsample slightly differed from the 3C population, as suggested by the smaller association between diabetes and dementia in this subsample. Third, we corrected for potential selection bias due to missing exposure and covariates data by the use of MI and there was no significant impact on the estimated association.

We must also discuss some limitations. First, even though the 3C study is a large population-based cohort study, participants who volunteered to participate had a higher educational level than the general population. Selective enrollment cannot completely be ruled out, with a possible higher participation rate for participants with less severe diabetes, and less diabetes-associated complications. Moreover, the possibility of selective attrition due to loss to follow-up may have underestimated the association between diabetes and dementia risk. Second, we must also acknowledge that diet is a major risk factor for diabetes and dementia risk and, hence, a potential confounding factor. For example, the Mediterranean diet is suggested to have a protective effect on type 2 diabetes [41], as well as cognitive decline and dementia [42]. Nevertheless, calculation of the E-value for our main results suggested that the observed association between baseline diabetes and incident dementia could be explained by residual confounding if the strength of association of an unmeasured confounder with both diabetes status and incident dementia was at least 2.96. We are unaware of such a strong association between dietary factors or other potential confounding factors and dementia risk. Third, we must emphasize the modeling assumptions of the illness-death model: transition intensities were assumed to be proportional, attrition was considered non-informative as death is assumed to be the only source of dependent censoring (whereas participants lost-to-follow-up may have different risks of dementia or death), and mortality rate in demented subjects was assumed to depend only on age and covariates, and not on disease duration. Visit times were also assumed to be independent of the illness-death process, which is likely in cohort studies with repeated follow-up visits such as the 3C study.

In conclusion, we found that type 2 diabetes was associated with an 80% increased risk of all-cause dementia over 12 years in French community-dwelling older adults, even after accounting for the competing risk of death. These results add to the growing evidence of a causal link between diabetes and dementia. These findings highlight the potential impact of diabetes prevention, early diagnosis and treatment on reducing the dementia risk, with a potential of 2-3-year increase in dementia-free life expectancy for older adults. Future interventional trials aimed at preventing diabetes should evaluate its effect on dementia and mortality to confirm the present results.

Footnotes

ACKNOWLEDGMENTS

The authors acknowledge the contribution of the 3C study participants.

The 3C Study is conducted under a partnership agreement among the Institut National de la Santé et de la Recherche Médicale (INSERM), the University of Bordeaux, and Sanofi-Aventis. The Fondation pour la Recherche Médicale funded the preparation and initiation of the study. The 3C Study is also supported by the Caisse Nationale Maladie des Travailleurs Salariés, Direction Générale de la Santé, Mutuelle Générale de l’Education Nationale, Institut de la Longévité, Conseils Régionaux of Aquitaine and Bourgogne, Fondation de France, Ministry of Research–INSERM Programme Cohortes et collections de données biologiques, the Fondation Plan Alzheimer (FCS 2009–2012), and the Caisse Nationale pour la Solidarité et l’Autonomie, Agence Nationale de la Recherche ANR PNR 2006 (ANR/DEDD/PNRA/PROJ/200206-01-01) and Longvie 2007 (LVIE-003-01). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

Cheng

, Noble

, Tang

, Schupf

, Mayeux

, Luchsinger

(2011) Type 2 diabetes and late-onset Alzheimer’s disease. Dement Geriatr Cogn Disord 31, 424–430.

Chatterjee

, Peters

, Woodward

, Mejia Arango

, Batty

, Beckett

, Beiser

, Borenstein

, Crane

, Haan

, Hassing

, Hayden

, Kiyohara

, Larson

, Li

, Ninomiya

, Ohara

, Peters

, Russ

, Seshadri

, Strand

, Walker

, Xu

, Huxley

(2016) 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 39, 300–307.

Yaffe

, Falvey

, Hamilton

, Schwartz

, Simonsick

, Satterfield

, Cauley

, Rosano

, Launer

, Strotmeyer

, Harris

(2012) Diabetes, glucose control, and 9-year cognitive decline among older adults without dementia. Arch Neurol 69, 1170–1175.

Tuligenga

, Dugravot

, Tabàk

, Elbaz

, Brunner

, Kivimäki

, Singh-Manoux

(2014) Midlife type 2 diabetes and poor glycaemic control as risk factors for cognitive decline in early old age: A analysis of the Whitehall II cohort study. Lancet Diabetes Endocrinol 2, 228–235.

Hsu

C-C

, Wahlqvist

, Lee

M-S

, Tsai

H-N

(2011) Incidence of dementia is increased in type 2 diabetes and reduced by the use of sulfonylureas and metformin. J Alzheimers Dis 24, 485–493.

Cheng

, Lin

C-H

, Tsai

Y-W

, Tsai

C-J

, Chou

P-H

, Lan

T-H

(2014) Type 2 diabetes and antidiabetic medications in relation to dementia diagnosis. J Gerontol A Biol Sci Med Sci 69, 1299–1305.

Areosa Sastre

, Vernooij

, Gonzàlez-Colaço Harmand

, Martínez

(2017) Effect of the treatment of Type 2 diabetes mellitus on the development of cognitive impairment and dementia. Cochrane Database Syst Rev 6, CD003804.

Tancredi

, Rosengren

, Svensson

, Kosiborod

, Pivodic

, Gudbjörnsdottir

, Wedel

, Clements

, Dahlqvist

, Lind

(2015) Excess mortality among persons with type 2 diabetes. N Engl J Med 373, 1720–1732.

Mayeda

, Haan

, Kanaya

, Yaffe

, Neuhaus

(2013) Type 2 diabetes and 10-year risk of dementia and cognitive impairment among older Mexican Americans. Diabetes Care 36, 2600–2606.

10.

Exalto

, Biessels

, Karter

, Huang

, Katon

, Minkoff

, Whitmer

(2013) Risk score for prediction of 10 year dementia risk in individuals with type 2 diabetes: A cohort study. Lancet Diabetes Endocrinol 1, 183–190.

11.

Shih

I-F

, Paul

, Haan

, Yu

, Ritz

(2018) Physical activity modifies the influence of apolipoprotein E ɛ4 allele and type 2 diabetes on dementia and cognitive impairment among older Mexican Americans. Alzheimers Dement 14, 1–9.

12.

Katon

, Pedersen

, Ribe

, Fenger-Grøn

, Davydow

, Waldorff

, Vestergaard

(2015) Effect of depression and diabetes mellitus on the risk for dementia: A national population-based cohort study. JAMA Psychiatry 72, 612–619.

13.

Haroon

, Austin

, Shah

, Wu

, Gill

, Booth

(2015) Risk of dementia in seniors with newly diagnosed diabetes: A population-based study. Diabetes Care 38, 1868–1875.

14.

Peila

, Rodriguez

, Launer

; Honolulu-Asia Aging Study (2002) Type 2 diabetes, APOE gene, and the risk for dementia and related pathologies: The Honolulu-Asia Aging Study. Diabetes 51, 1256–1262.

15.

Irie

, Fitzpatrick

, Lopez

, Kuller

, Peila

, Newman

, Launer

(2008) Enhanced risk for Alzheimer disease in persons with type 2 diabetes and APOE ɛ4: The Cardiovascular Health Study Cognition study. Arch Neurol 65, 89–93.

16.

Haan

, Shemanski

, Jagust

, Manolio

, Kuller

(1999) The role of APOE epsilon4 in modulating effects of other risk factors for cognitive decline in elderly persons. JAMA 282, 40–46.

17.

Dore

, Elias

, Robbins

, Elias

, Nagy

(2009) Presence of the APOE epsilon4 allele modifies the relationship between type 2 diabetes and cognitive performance: The Maine-Syracuse Study. Diabetologia 52, 2551–2560.

18.

3C Study Group (2003) Vascular factors and risk of dementia: Design of the Three-City Study and baseline characteristics of the study population. Neuroepidemiology 22, 316–325.

19.

American Psychiatric Association (1994) Diagnostic and statistical manual of mental disorders: DSM-IV. 4th ed. American Psychiatric Association, Washington, DC.

20.

McKhann

, Drachman

, Folstein

, Katzman

, Price

, Stadlan

(1984) Clinical diagnosis of Alzheimer’s disease: Report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer’s Disease. Neurology 34, 939–944.

21.

Romàn

, Tatemichi

, Erkinjuntti

, Cummings

, Masdeu

, Garcia

, Amaducci

, Orgogozo

, Brun

, Hofman

(1993) Vascular dementia: Diagnostic criteria for research studies. Report of the NINDS-AIREN International Workshop. Neurology 43, 250–260.

22.

Alpérovitch

, Bertrand

, Jougla

, Vidal

, Ducimetière

, Helmer

, Ritchie

, Pavillon

, Tzourio

(2009) Do we really know the cause of death of the very old? Comparison between official mortality statistics and cohort study classification. Eur J Epidemiol 24, 669–675.

23.

Helmer

, Stengel

, Metzger

, Froissart

, Massy

, Tzourio

, Berr

, Dartigues

(2011) Chronic kidney disease, cognitive decline, and incident dementia: The 3C Study. Neurology 77, 2043–2051.

24.

Dufouil

, Richard

, Fiévet

, Dartigues

, Ritchie

, Tzourio

, Amouyel

, Alpérovitch

(2005) APOE genotype, cholesterol level, lipid-lowering treatment, and dementia: The Three-City Study. Neurology 64, 1531–1538.

25.

Leffondré

, Touraine

, Helmer

, Joly

(2013) Interval-censored time-to-event and competing risk with death: Is the illness-death model more accurate than the Cox model? Int J Epidemiol 42, 1177–1186.

26.

VanderWeele

, Ding

(2017) Sensitivity Analysis in Observational Research: Introducing the E-Value. Ann Intern Med 167, 268–274.

27.

World Health Organization, International Diabetes Federation (2006) Definition and diagnosis of diabetes mellitus and intermediate hyperglycemia: Report of a WHO/IDF consultation. World Health Organization, Geneva.

28.

van Buuren

(2007) Multiple imputation of discrete and continuous data by fully conditional specification. Stat Methods Med Res 16, 219–242.

29.

Touraine

, Gerds

, Joly

(2017) SmoothHazard: An R Package for Fitting Regression Models to Interval-Censored Observations of Illness-Death Models. J Stat Softw 79, 1–22.

30.

Ohara

, DOI

, Ninomiya

, Hirakawa

, Hata

, Iwaki

, Kanba

, Kiyohara

(2011) Glucose tolerance status and risk of dementia in the community: The Hisayama Study. Neurology 77, 1126–1134.

31.

Schrijvers

EMC

, Witteman

JCM

, Sijbrands

EJG

, Hofman

, Koudstaal

, Breteler

MMB

(2010) Insulin metabolism and the risk of Alzheimer disease: The Rotterdam Study. Neurology 75, 1982–1987.

32.

Mayeda

, Whitmer

, Yaffe

(2015) Diabetes and Cognition. Clin Geriatr Med 31, 101–115, ix.

33.

, Risacher

, Huang

, Saykin

, Alzheimer’s Disease Neuroimaging Initiative (2016) Type 2 diabetes mellitus is associated with brain atrophy and hypometabolism in the ADNI cohort. Neurology 87, 595–600.

34.

Schneider

ALC

, Selvin

, Sharrett

, Griswold

, Coresh

, Jack

CR Jr

, Knopman

, Mosley

, Gottesman

(2017) Diabetes, prediabetes, and brain volumes and subclinical cerebrovascular disease on MRI: The Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). Diabetes Care 40, 1514–1521.

35.

Jacqmin-Gadda

, Alperovitch

, Montlahuc

, Commenges

, Leffondre

, Dufouil

, Elbaz

, Tzourio

, Ménard

, Dartigues

, Joly

(2013) 20-Year prevalence projections for dementia and impact of preventive policy about risk factors. Eur J Epidemiol 28, 493–502.

36.

Bancks

, Ning

, Allen

, Bertoni

, Carnethon

, Correa

, Echouffo-Tcheugui

, Lange

, Lloyd-Jones

, Wilkins

(2019) Long-term absolute risk for cardiovascular disease stratified by fasting glucose level. Diabetes Care 42, 457–465.

37.

Rasmussen

, Tybjærg-Hansen

, Nordestgaard

, Frikke-Schmidt

(2018) Absolute 10-year risk of dementia by age, sex and APOE genotype: A population-based cohort study. CMAJ 190, E1033–E1041.

38.

Wolters

, Tinga

, Dhana

, Koudstaal

, Hofman

, Bos

, Franco

, Ikram

(2019) Life expectancy with and without dementia: A population-based study of dementia burden and preventive potential. Am J Epidemiol 188, 372–381.

39.

Demakakos

, Muniz-Terrera

, Nouwen

(2017) Type 2 diabetes, depressive symptoms and trajectories of cognitive decline in a national sample of community-dwellers: A prospective cohort study. PloS One 12, e0175827.

40.

Launer

, Miller

, Williamson

, Lazar

, Gerstein

, Murray

, Sullivan

, Horowitz

, Ding

, Marcovina

, Lovato

, Margolis

, O’Connor

, Lipkin

, Hirsch

, Coker

, Maldjian

, Sunshine

, Truwit

, Davatzikos

, Bryan

; ACCORD MIND investigators (2011) 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 10, 969–977.

41.

Salas-Salvadó

, Guasch-Ferré

, Lee

C-H

, Estruch

, Clish

, Ros

(2016) Protective effects of the Mediterranean diet on type 2 diabetes and metabolic syndrome. J Nutr 146, 920S–927S.

42.

Féart

, Samieri

, Allès

, Barberger-Gateau

(2013) Potential benefits of adherence to the Mediterranean diet on cognitive health. Proc Nutr Soc 72, 140–152.