Introduction
People with type 2 diabetes are at greater risk of cognitive impairment in later life than their non-diabetic peers [
1]. With the prevalence of diabetes expected to reach 15% in the USA within the current decade [
2], researchers have attempted to identify underlying pathophysiological mechanisms and potentially modifiable risk factors that may be responsible for the observation. For instance, based on competition of insulin with β-amyloid peptide for degradation and recent reports identifying the amyloid precursor amylin, which is co-secreted with insulin from the pancreas, in brains of patients with diabetes (but not in diabetes-free controls), increased concentrations of β-amyloid and amylin are candidate links between type 2 diabetes and cognitive impairment [
3,
4]. The role of vascular disease has also received increasing attention of late due to an increased prevalence in type 2 diabetes and plausible biological mechanisms linking the two conditions. Amylin, for instance, is known to contribute to cardiovascular disease, so similar effects on the vascular system of the brain may be plausible [
4]. Recent epidemiological evidence has also shown a direct association between cognitive decline and markers of symptomatic and asymptomatic vascular disease in people with type 2 diabetes [
5]. These observations suggest a potential role for smoking, hypertension, adverse lipid profiles and hyperglycaemia, all of which are established risk factors for adverse cardiovascular outcomes, in the development of cognitive impairment in people with type 2 diabetes. If such a role is repeatedly demonstrated, then future clinical trials targeting these risk factors should consider inclusion of cognitive endpoints (e.g. to determine whether a change in the threshold for existing clinical intervention may be indicated).
In the general population, hypercholesterolaemia, hypertension, smoking and hyperglycaemia have each been associated with a poorer level of cognitive function [
6‐
8]. Lower childhood intelligence, however, potentially leads to increased vascular risk, as well as to lower cognitive ability and greater cognitive decline in later life [
9]. Therefore, studies with a prospective design and, preferably, with a measure of pre-morbid cognitive ability, are required to help determine the direction of relationships between vascular risk factors and cognitive decline. Compared with previous prospective studies in the general population that suggest a degree of complexity and do not always provide a consistent picture of vascular risk and cognitive impairment [
10,
11], such studies in diabetic populations are much more scarce and often sub-optimal in design not least due to the neglect of consideration of pre-morbid ability [
12]. For smoking in particular, this type of evidence was lacking until a recent analysis of the Fremantle Diabetes Study, which showed that midlife smoking predicted an increased risk of future cognitive impairment [
13].
Using data from a large prospective cohort of older adults with type 2 diabetes, the Edinburgh Type 2 Diabetes Study (ET2DS), we tested the associations of serum cholesterol, BP, glycaemic control and smoking with late-life cognitive decline. Given known strong correlations among vascular risk factors themselves, it was important to evaluate the relative independence of their contributions in our cohort. Our analyses had the additional advantage of using ‘historical’ data on BP and HbA1c covering the time between diabetes diagnosis and attendance at the clinic.
Discussion
Smoking history, long-term exposure to raised BP and poorer glycaemic control were independently associated with an accelerated late-life cognitive decline short of dementia. Overall effect sizes were modest but, compared with the findings on BP, those for a measure of mean HbA1c collected over a time period of up to 20 years prior to cognitive testing and for lifetime smoking history appeared to be more robust across the total sample. Independent of age and sex, a history of moderate to heavy smoking and each percentage increase in HbA1c (an increase that is equivalent to 11 mmol/mol) was associated with a 64% and 21% increased risk of an accelerated cognitive decline during 4 year follow-up, respectively. Cholesterol was associated only cross-sectionally with cognition. While our findings are not directly transferable to a clinical setting (since our definition of accelerated decline was arbitrary and sample-specific), they should serve to increase awareness of cognitive impairment and the potential importance of treatment of vascular risk factors in patients with type 2 diabetes.
To date, the relationships between cardiovascular risk factors and cognition have only infrequently been studied in people with diabetes. The present study has extended earlier evidence of cross-sectional associations of hypertension [
24,
25] and poorer glycaemic control [
26,
27] with reduced cognitive function in type 2 diabetes by demonstrating prospective associations with accelerated late-life cognitive decline. The association of smoking history with cognitive decline supports recent findings from the Fremantle Diabetes Study, in which midlife smoking was associated with the risk of cognitive impairment in later life [
13]. Findings on glycaemic control in the present cohort are also consistent with a recent analysis of the Atherosclerosis Risk in Communities (ARIC) study, which reported that people with diabetes in midlife declined cognitively at faster rates during a 20 year follow-up period compared with diabetes-free individuals [
28] and which additionally suggested a role for severity of hyperglycaemia [
28]. In our study, the corresponding association of historical blood glucose levels with cognitive decline in people with diabetes in later life was particularly evident for tests of executive function, which again is consistent with findings from the ARIC study [
28] and with reports of vulnerability of that domain in people with diabetes [
29]. The suggestion of stronger findings for glycaemic control in insulin-treated patients in the present study may be due to the advanced disease stage and comorbidity in these patients and, in women, merits further investigation.
Many previous studies on a similar range of risk factors have investigated single factors, despite complex relationships among them. The present study makes an important advance by showing that associations of each of the risk factors identified as being predictive of cognitive decline were independent of one another. Although observational findings do not correspond to causality, it is possible that each factor is associated with cognition in separate pathophysiological pathways. Their modifiable nature suggests the potential for each risk factor as a separate target for intervention to reduce patients’ risk of cognitive impairment. Observational data from studies of people with type 2 diabetes appear to support this possibility (e.g. with associations of antihypertensive treatment with reduced risk of dementia as an endpoint of cognitive decline [
30]) but current evidence from trials is limited and inconclusive. In the Memory in Diabetes study of the Action to Control Cardiovascular Risk in Diabetes (ACCORD-MIND) trial, a trend was observed for decelerated cognitive decline in the treatment arm that was subjected to intensified glucose-lowering treatment [
31]. However, some evidence suggests that hypoglycaemia (a side effect of more intensive therapy) may increase the rate of cognitive decline [
32‐
35], thereby potentially counteracting the beneficial effects of improved glycaemic control. Future trials comparing glucose-lowering agents that cause different degrees of hypoglycaemia may help to clarify this. Trials on smoking cessation are complicated by confounding and are scarce even in the general population, with a single study showing decelerated 2 year cognitive decline in people who successfully quit smoking compared with those who were unsuccessful [
36]. A Cochrane review of randomised trials in the general population concluded that antihypertensive therapy was of no benefit in slowing cognitive decline [
37] and in the ACCORD-MIND study, intensive lowering of BP in people with type 2 diabetes had no effect on cognitive decline [
38].
The present findings are consistent with those of a number of studies that have reported links of higher midlife or late-life cholesterol, or of lower HDL-cholesterol, with lower late-life cognitive function or with a steeper cognitive decline in people with diabetes [
11,
39,
40]. However, these studies did not adjust for pre-morbid ability. Here, we found evidence for reverse causality: lower pre-morbid ability predisposed participants to a higher late-life level of cholesterol and to lower late-life cognitive function, and so confounding of previous cross-sectional (and potentially prospective) evidence is likely. In line with the present null findings for prospective associations of either cholesterol (without regard to lipid-lowering treatment) or of presence of hypercholesterolaemia (partly defined by use of lipid-lowering treatment), ACCORD-MIND revealed no effect of addition of fenofibrate to simvastatin therapy on 3 year cognitive decline [
38].
The prospective design and extensive clinical and cognitive characterisation are clear strengths of the ET2DS. Adjustment for estimated peak pre-morbid ability helped to counteract potential reverse causality. The dropout of 22% over 4 years was relatively high but acceptable considering the diabetic status and age of the cohort. The associated survivor bias favouring healthier individuals to attend follow-up is likely to have led to an underestimate in effect sizes, so that associations may be stronger than reported here. Our overall results are further strengthened by the use of historical data. Despite data heterogeneity (historical data were available for between 1 and 65 readings), which may decrease the overall reliability of our findings, the approach substantially reduced the influence of measurement error compared with single measurement data. The calculation of a
g factor allowed the analysis of decline in overall cognitive function that was independent of the specific test battery used to measure cognitive function [
41]. A resulting neglect of test-specific variance [
42] was partly offset by additional consideration of findings from individual cognitive tests. Multiple testing will have increased the risk of type I statistical error and subgroup analysis by sex will have reduced statistical power, limiting the reliability of all findings and of those stratified by sex in particular. Replication in other cohorts with type 2 diabetes is therefore essential. Readers should also note that the category of accelerated cognitive decline was specific to the current, predominantly dementia-free cohort, and should not be mistaken for the presence of dementia. Finally, we did not consider the role of treatment of vascular risk, and so potential moderation of risk factor associations with cognitive decline by treatment cannot be excluded. The contributions of insulin or risk factor trajectories between baseline and year 4 follow-up to findings were also not assessed; such analyses are currently planned as part of future follow-up waves of the ET2DS.
Should associations of long-term exposure to vascular risk with cognitive decline prove to be causal, the present findings suggest that interventions targeting a range of vascular risk factors (rather than single factors such as hyperglycaemia only) may be needed to help reduce the risk of cognitive impairment in people who either have diabetes or who go on to develop type 2 diabetes in older age. Our findings support recent evidence that late-life cognitive change may be determined over the course of decades prior to the age at which deficits typically occur [
28].
Overall, we have provided strong observational evidence showing that smoking history, higher BP and poorer glycaemic control are independently associated with accelerated late-life cognitive decline in people with type 2 diabetes. Further research into the usefulness of each risk factor as a target for intervention to reduce age-related cognitive decline in this patient population is now warranted.
Acknowledgements
We thank all participants of the ET2DS.