The impact of diabetes on cognitive decline: potential vascular, metabolic, and psychosocial risk factors

Although dyslipidemia is common in type 2 diabetes, few observational studies have examined whether an association exists between plasma lipid concentrations and cognitive function (Table 1). Cognitive function has been reported to be significantly poorer in people with type 2 diabetes who have elevated levels of plasma triglycerides [9, 10] and in those with higher cholesterol levels [9], but neither of these observations has been confirmed [11, 12]. Two investigations even reported protective effects: in one cross-sectional study, dislipidemia was associated with better performance on a task of processing speed [13], and higher total cholesterol was found to decrease the risk of subsequent cognitive impairment short of dementia during an 8-year period in the Fremantle Diabetes Study [14]. However, a small intervention study on the effects of physical exercise on insulin resistance over a period of 12 weeks (which was unsuccessful in inducing a change in insulin resistance) [15] and two further prospective observational studies [16, 17] failed to find any association between plasma lipid profiles and subsequent cognitive decline or risk of impairment, with the exception of an apparent association between lower mean high-density lipoprotein during a 6-year period and a steeper-than-expected cognitive decline in a small Japanese study during the same time period [16].

In the Action to Control Cardiovascular Risk in Diabetes-Memory in Diabetes (ACCORD-MIND) randomized controlled trial (RCT), almost 3,000 older people with type 2 diabetes were assigned either to intensive treatment of hyperglycemia or to standard therapy [17]. Around 50 % of participants also entered the only RCT to date to address the effects of a reduction in plasma lipid levels on cognitive decline in people with type 2 diabetes (the other 50 % participated in a trial of anti-hypertensives). Despite a greater reduction in cholesterol levels in patients who received fenofibrate plus simvastatin compared with those receiving placebo plus simvastatin, cognitive function in the two groups declined at similar rates during a 40-month follow-up period [17]. A review of RCTs performed in the general (predominantly non-diabetic) population also concluded that reducing plasma cholesterol does not influence late-life cognitive function [18], consistent with findings from observational studies performed in the general population [2]. The role of dyslipidemia in the development of cognitive impairment in people with diabetes is therefore uncertain.

Hypertension is common in people with type 2 diabetes and, in general, has received more attention than dyslipidemia as a potential risk factor for diabetes-related cognitive impairment (Table 2). Cross-sectional studies have revealed trends for increased prevalence of hypertension in patients with lower cognitive function [11, 13, 19, 20], but cross-sectional analyses of blood pressure as a continuous measure have failed to identify similar associations [11, 14, 21]. On the other hand, some [14, 19, 22], though not all [17, 21], prospective studies have found a relationship between baseline blood pressure or hypertension and the subsequent risk of cognitive decline. In the Fremantle Diabetes Study, higher baseline diastolic blood pressure was associated with an increased risk of incident AD after 8 years [14], and in an investigation of people over 80 years of age, the coexistence of hypertension appeared to exacerbate diabetes-related cognitive decline during a 6-year follow-up and to increase the risk of dementia [19]. Similarly, a retrospective study that examined the hospital records of almost 380,000 older patients with diabetes showed that co-morbid hypertension increased the 2-year risk of dementia; treatment with anti-hypertensive medication (other than α-adrenoceptor blockers, with which the risk of dementia was increased) further diminished the risk of dementia by between 4 % and 24 % depending on the precise type of drug used [22]. In contrast with these findings, the blood pressure trial of the ACCORD-MIND study did not demonstrate a difference in cognitive decline over a period of 40 months between a group of patients who received intensive anti-hypertensive therapy and a group on conventional treatment, despite the success of the trial in producing a difference in blood pressure between the two treatment groups [17]. However, a direct association between blood pressure and cognitive decline was not explored. In the general (non-diabetic) population, the results of observational studies and of RCTs investigating links between hypertension and cognitive impairment have also, in the main, been negative [2, 23]. Therefore, although hypertension causes cerebrovascular disease and, as such, represents a good candidate for a cognitive risk factor, its role in the development of cognitive decline during aging in either the diabetic or non-diabetic population remains unproven.

Raised blood glucose levels within the non-diabetic or pre-diabetic range have consistently been associated with cognitive impairment, with the strength of the association increasing with advancing age [24]. Given that diabetes is characterized by persistently raised blood glucose levels, a causative role for hyperglycemia in diabetes-associated cognitive decline would seem likely. However, the findings from cross-sectional analyses on the association of HbA1c with cognitive function [9, 11, 13, 14, 20, 25] and cognitive decline [14, 16, 21] in people with type 2 diabetes have been inconsistent (Table 3), potentially due to the different ages of the study populations. Overall, the association of type 2 diabetes with increased cognitive impairment appears to be relatively weak before the age of 70 years, provided that good glycemic control is maintained, and it is only in older patients that cognitive decrements related to chronic hyperglycemia become apparent [26]. More recently, a retrospective analysis of a cohort of people with type 2 diabetes, in whom 12-year data on HbA1c were available from a diabetes register, showed that in addition to increments in blood glucose levels over time, poor glycemic control long-term predicted a lower level of late-life cognitive function, despite a trend toward improved glycemic control by intensifying therapy [27]. This is consistent with the evidence showing damaging effects of mid-life diabetes on the risk of late-life cognitive impairment [4, 5] and suggests that irreversible damage may already have occurred to predispose people to cognitive impairment by the time that aggressive glucose-lowering treatment was commenced.

In one of a number of intervention studies, changes in blood glucose levels due to physical exercise correlated with changes in cognitive function [15]. The ACCORD-MIND study also found a statistically non-significant trend for decelerated decline in processing speed at 20 months in the group with intensive therapy for glycemic control (who achieved relatively greater glycemic control) as compared with the conventional treatment group (with resulting poorer glycemic control), although this difference was no longer apparent at 40 months [28]. Two smaller trials of patients with type 2 diabetes have reported significant associations between improved glycemic control and cognitive function. In one, improvements in glycemic control in both treatment groups due to treatment with either rosiglitazone or glibenclamide (glyburide) correlated with improvement in working memory over a period of 24 weeks [29]. In another, a reduction in post-prandial glucose excursions with repaglinide was associated with a decline in cognitive function over a period of 12 months compared with subjects who received glibenclamide and did not show such a change in glucose excursions; the decline in HbA1c was of similar magnitude in the two treatment groups, suggesting a specific role for post-prandial glucose excursions [30]. Whereas overall a recent systematic review combining the evidence from observational studies and from RCTs concluded that both hyperglycemia and glucose excursions are weakly associated with poorer cognitive function in people with type 2 diabetes [31], a meta-analysis restricted to RCTs suggested that improvement in glycemic control was unrelated to cognitive decline [32], illustrating the need for further evaluation of hyperglycemia as a potentially modifiable cognitive risk factor.

Few studies have investigated the effect of previous exposure to recurrent hypoglycemia on cognitive function in people with type 2 diabetes. Heterogeneity with respect to how ‘hypoglycemia’ has been defined presents a major problem for interpretation of results, with recorded events ranging from asymptomatic biochemical hypoglycemia to severe disabling hypoglycemia (Table 4).

Cross-sectional analyses have reported an association between a history of previous self-reported or medically verified severe hypoglycemia, defined as any episode requiring external help to effect recovery, and cognitive impairment [14, 33, 34] but this could reflect lower cognitive ability in people who go on to experience a higher frequency of severe hypoglycemia. Indeed, in the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) trial, which successfully manipulated the level of glycemic control (intensive versus standard) in patients with type 2 diabetes, each one-unit-lower score on a cognitive screening instrument at baseline was associated with a 10 % greater risk of severe hypoglycemia during follow-up [35]. A lower baseline cognitive function and (for participants who had low processing speed at baseline) a relatively steeper cognitive decline between baseline and the 20-month assessment were also predictive of an increase in the subsequent first-ever hospital admission to treat severe hypoglycemia in ACCORD-MIND, and the group with low processing speed and declining cognitive function had a higher cumulative incidence of severe hypoglycemia over the 4 years of the study [36]. Finally, in two further prospective investigations, a baseline diagnosis of dementia or a diagnosis during the follow-up period in previously unimpaired participants was associated with a two- to three-fold higher rate of hospital admission for emergency medical treatment of hypoglycemia during follow-up [34, 37].

Whether exposure to hypoglycemia precedes cognitive decline and may even be a causal risk factor for this condition is less clear. In the Edinburgh Type 2 Diabetes Study (ET2DS) of more than 1,000 adults between 60 and 75 years of age, a history of severe hypoglycemia was associated with lower cognitive function when the estimated pre-morbid cognitive function before exposure to hypoglycemia was compared with post-hypoglycemia cognitive function, with evidence of an acceleration of late-life cognitive decline that was independent of the potential influence of further episodes of hypoglycemia [38] (Fig. 2). However, these observations conflict with the findings of the Fremantle Diabetes Study [21] and with the evidence from RCTs. In ACCORD-MIND and ADVANCE, cognitive function declined at similar rates by 40 months and 5 years in patients in the intensive treatment groups (in whom the incidences of hypoglycemia were significantly higher) compared with those in the standard treatment arms [28, 35]. However, in both of these trials, the management of diabetes was manipulated to attain pre-determined glycemic targets. It is plausible that any detrimental effect of hypoglycemia was counterbalanced by an improvement in cognitive function occurring through specific beneficial effects of the assigned intervention.

Two retrospective investigations [39, 40] have suggested that a dose-response relationship may exist between the frequency of exposure to severe hypoglycemia and the subsequent risk of dementia. However, these studies relied on hospital records, a suboptimal method of identifying hypoglycemia, and the suggestion that exposure to a single episode of hypoglycemia would induce dementia is biologically implausible. In the observational analysis of the Fremantle Diabetes Study, a history of severe hypoglycemia failed to predict the 5-year risk of dementia [34]. In the prospective Health Aging and Body Composition Study, participants with incident hypoglycemia had a two-fold risk of subsequent dementia over a period of 12 years, but in combination with the analysis showing an increased risk of subsequent hypoglycemia in patients who were diagnosed with dementia during follow-up, the data were overall interpreted as showing bidirectional causality [37]. It is essential that the role of hypoglycemia in either causing or accelerating cognitive decline be clarified in view of the current policy to use intensive therapy to achieve near-normoglycemia to minimize the development of diabetic complications.

Hyperinsulinemia from endogenous hypersecretion of insulin is common in the early stages of type 2 diabetes as a ‘pathophysiological’ response to insulin resistance; it also occurs as a consequence of exogenous insulin therapy. Hyperinsulinemia has been associated with cognitive impairment, but a systematic review of observational studies that included people with and those without diabetes concluded that the evidence for an association of elevated plasma insulin concentrations with impairment of cognition was weak, because it is possible that any association of plasma insulin with cognition in such samples had been influenced by the inclusion of people with diabetes [24]. Very few studies have been performed in non-diabetic populations or exclusively in people with type 2 diabetes (Table 3). In one observational study, a higher mean insulin during a 6-year period was associated with a steeper rate of concurrent cognitive decline based on a test of executive function [16], and in a small intervention study, which was unsuccessful in inducing a difference in insulin sensitivity in two treatment groups through physical exercise, improvement in memory performance correlated with improvements in insulin resistance [15]. By contrast, in a larger 24-week trial in middle-aged to older patients with type 2 diabetes (mean age of 60 years), an improvement in plasma insulin levels and insulin sensitivity had no effect on concurrent change in cognitive function [29]. Similarly, in the ACCORD-MIND study, treatment with insulin on study entry or during the trial was relatively unrelated to 40-month cognitive change, but plasma insulin levels as such were not considered [41]. This is despite the fact that compliance by participants is difficult to ascertain. Inter-relationships between plasma insulin concentration, insulin resistance, and the quality of glycemic control further complicate attempts to evaluate associations of any of these risk factors with cognitive impairment.

Chronic low-grade inflammation is a characteristic feature of both diabetes and AD and appears to interact with diabetes in its association with cognitive impairment. This suggests a common biological mechanism [42]. Circulating markers of inflammation include C-reactive protein (CRP), interleukin-6 (IL-6), fibrinogen, and tumor necrosis factor-alpha (TNF-α), some of which have been associated with cognitive dysfunction in people with diabetes (Additional file 1: Table S1). Elevated levels of CRP have been associated with lower cognitive function in small studies of hospitalized patients (for example, [9]). In the ET2DS, higher levels of fibrinogen, TNF-α, and IL-6 but not CRP were associated with lower measures of cognitive function [43, 44]; higher baseline levels of fibrinogen and IL-6 additionally predicted a steeper 4-year cognitive decline [45, 46]. CRP levels were also unrelated to cognitive decline in a further prospective study with a 6-year follow-up [16]. In support of associations (particularly, causal) between inflammation and cognition, genetic variants that influence circulating levels of inflammatory markers have been associated with cognitive impairment, but this finding has not been consistent [43, 47].

Because of the homology between retinal and cerebrovascular cells, the state of small vessels in the retina closely mirrors that of the cerebral microvasculature, suggesting that diabetic retinopathy can be used as a marker for the presence of microangiopathy within the brain. A systematic review of cross-sectional and prospective observational studies concluded that people from the general population and people with diabetes who exhibit retinal microvascular abnormalities appear to be at increased risk of cognitive impairment, including dementia, compared with people who have no retinal microvascular abnormalities [48], although subsequent studies have given conflicting results [25, 35, 49] (Additional file 1: Table S2). However, in support of the findings of the systematic review, baseline presence of retinopathy was recently identified as a predictor of steeper rates of cognitive decline during 40-month (but not 20-month intermittent) follow-up in ACCORD-MIND [50]. Overall, diabetic retinopathy may be a putative surrogate marker for cognitive impairment in people with diabetes, in which cerebral microvascular disease may have an important pathogenetic role.

An association of CHD with a lower level of cognitive function was observed in the ET2DS [51] (Additional file 1: Table S3). However, in the ACCORD-MIND study, the evidence for an association of CHD with cognitive dysfunction was limited [13], and in all other cross-sectional investigations [14, 35] and in all prospective analyses, including the ET2DS [21, 51], the results have been negative. The only significant prospective association was in the direction of cognitive ability predicting worsening of CHD. In ADVANCE, after multivariate adjustment, a baseline presence of ‘mild cognitive dysfunction’ and ‘severe cognitive dysfunction’ increased the 5-year risk of a major coronary event by 31 % and 70 %, respectively [35].

The inactive metabolite N-terminal pro-brain natriuretic peptide (NT-proBNP) is a biomarker of the cardiac stress associated with ventricular dysfunction and congestive heart failure. In the ET2DS, an association of small effect size was found between a higher baseline NT-proBNP and a lower cognitive ability and with a steeper cognitive decline later in life [51] (Additional file 1: Table S4). In the general population and in people with cardiovascular disease, associations of small to large unadjusted effect size have been reported relatively consistently between elevated levels of natriuretic peptide and lower cognitive function (for example, [52]) and with the presence of dementia or milder forms of impairment [53, 54]. Null findings are rare [55] and in some studies may have resulted from the nature of the cognitive screening instruments that were applied. Some results have suggested an independence of these associations from symptomatic macrovascular disease, including stroke [52], which was also observed in the ET2DS [51]. Prospective investigations of the general population were, until recently, restricted to a single cohort (of people over 75 years of age) in which the findings were inconclusive [53, 54]. However, a recent large Finnish study (n = 7,000 participants) that examined the relationship of natriuretic peptides with cognitive function has not provided definitive evidence [56]. In that study, each standard deviation above baseline NT-proBNP predicted a 48 % increased risk of dementia during 14-year follow-up after multivariate adjustment in men, but no such association was found in women.

An association between a lower level of cognitive function and cerebral infarction has been a consistent finding in populations with diabetes [13, 51] (Additional file 1: Table S3). For example, in the Fremantle Diabetes Study, cognitive impairment diagnosed on the basis of a screening instrument and follow-up clinical interview was associated with a history of cerebrovascular disease [14]. The impact of stroke on cognitive function was demonstrated in the diabetic subpopulation of a Dutch study and in the ET2DS, in which an association between stroke and diminished cognitive function persisted after adjustment for estimated pre-morbid ability [20, 25, 51].

In the ET2DS and in the Fremantle Diabetes Study, a history of stroke was associated with a steeper decline in cognitive ability [14, 51], but this observation differed from those of several other prospective analyses (for example, [21]), in which no such association was found. In the ADVANCE study, evidence of a prospective association in the direction of lower cognitive function predisposing patients to an increased risk of infarction was observed. After multivariate adjustment, individuals with ‘mildly reduced’ cognitive function at baseline had a 5-year risk of sustaining a major stroke which was 34 % greater than that of individuals who had a higher level of cognitive ability; people with ‘severe cognitive dysfunction’ had a 71 % greater risk [35]. The relationship between cerebral infarction and cognition in diabetes may therefore be bidirectional.

In people with type 2 diabetes, greater carotid intima-media thickness (cIMT) has been associated with a lower level of cognitive function [9, 11], but its association with an estimated steeper decline of lifetime cognitive function has been inconsistent [25, 51] (Additional file 1: Table S3). To date, the ET2DS appears to be the only prospective study to examine cIMT and cognition in people with type 2 diabetes. This identified an association of cIMT with a steeper decline in late-life cognitive function, which was independent of a preceding history of stroke [51]. In the population in general, an association between a higher cIMT and an increased risk of cognitive impairment has been established [57], and so a similar association is likely to exist in people with type 2 diabetes.

In people with type 2 diabetes, a low ankle-brachial pressure index (ABI) - a measure of peripheral arterial disease (PAD) of the lower limbs and of more generalized atherosclerosis - and PAD diagnosis have been associated with lower cognitive function [9, 51] and with dementia [14] (Additional file 1: Table S3). In the Fremantle Diabetes Study, 38 % of cognitively ‘normal’ individuals, 45 % of people with reduced cognitive function, and 75 % of people with frank dementia had evidence of coexisting PAD [14]. In one study, cross-sectional findings for ‘any vascular event’ (which was partly defined by PAD) remained significant after adjustment for an estimate of peak pre-morbid ability [20, 25], but after such an adjustment was made in the analysis of ABI and symptomatic PAD in the ET2DS, it did not quite achieve statistical significance [51]. However, in the latter, each standard deviation of a lower baseline ABI was also associated with a 0.12-standard deviation increment in subsequent 4-year decline on a composite measure of cognitive function [51]. In the Fremantle Diabetes Study, PAD measured 8 years earlier also predicted an increased risk of cognitive impairment [14] though it was not associated with the risk of cognitive decline in the subsequent 2-year follow-up period [21].

Overall, the evidence for an association between macrovascular disease and cognitive impairment in diabetes is inconsistent and varies according to the area of the vasculature considered. As might be expected, evidence for a relationship with cerebrovascular disease, especially stroke, is stronger than that for vascular sites which are more distant from the brain, including the heart. Evidence for an association with the most distal presentations of macrovascular disease, such as PAD of the lower limbs, is particularly limited, is likely to reflect widespread atherosclerosis as a marker for cognitive impairment in people with diabetes, and would suggest that any true associations have a small effect size.

Cross-sectional studies of cognitive function in people with diabetes, with or without depressive symptoms or clinical depression, have been inconclusive (Additional file 1: Table S5). One investigation of older people with type 2 diabetes reported a statistically non-significant trend for negative correlations between scores on a cognitive screening instrument and scores on a self-administered screening instrument for depression [58]. In a cross-sectional analysis of ACCORD-MIND, patients with depression (based on scores on screening instrument or on self-report) also scored lower on a cognitive screening instrument (though not on more detailed neuropsychological tests) compared with patients who were free of depression [13]. Additive detrimental effects have been suggested by another study of people with type 2 diabetes and healthy controls who were 30 to 80 years of age (the mean age was 60 years across groups), to whom more detailed neuropsychological testing was applied along with clinical interviews to diagnose depression. The patients with co-morbid diabetes and depression performed less well on tests of attention and processing speed compared with participants with diabetes but without depression. Relative to the latter, there was also a trend just short of statistical significance for lower cognitive function overall in the group with co-morbid diabetes and depression [59]. In a prospective analysis of a large cohort of Americans, co-morbidities of diabetes and depression were also linked to a 100 % increased risk of dementia over a period of 3 to 5 years when compared with people with diabetes but without depression [60]. Finally, ACCORD-MIND revealed associations of higher scores on a screening instrument for depression and a steeper 40-month cognitive decline [61]. In the general population, the association of depression with cognitive impairment appears to be well established [60], and so it seems likely that depression has a contributing role in promoting diabetes-associated cognitive impairment.

Diabetes-associated cognitive impairment may partly reflect reverse causality. Consistent with the assumption that individuals who have lower cognitive ability may be predisposed to have lower late-life cognitive function and to be at increased risk of developing diabetes as they get older, an analysis of the Lothian Birth Cohort (a group of people who were born in 1936) found that cross-sectional associations of diabetes with lower late-life cognitive ability disappeared following adjustment for cognitive ability that had been measured at age 11 [62] (Fig. 3).

However, where diabetes is associated with a steeper late-life cognitive decline in prospective analyses, the role of pre-morbid ability is as yet unclear, particularly as its role in promoting late-life cognitive decline per se is uncertain. Some prospective investigations have indicated that individuals with lower cognitive ability decline more rapidly as they get older [63], but this has not been confirmed [64].