Table 3 Studies of hyperglycemia or hyperinsulinemia and cognitive function in type 2 diabetes

Abbatecola et al. [30] (2006)

Patients with diabetes free of vascular disease, receiving repaglinide or glibenclamide; Italy

12-month trial on glycemic control

Mean 74 ± 2 years

Variation in post-prandial blood glucose, fasting plasma glucose, and HbA1c. (No difference between groups in reduction of HbA1c and plasma glucose during trial. Decline in variation in post-prandial glucose only in group treated with repaglinide.)

Composite score of attention/executive function (Trail-Making Test, Digit Span, and verbal fluency), MMSE, cortical atrophy, and white matter lesions

Age, education, physical activity, depression, blood pressure, cIMT, insulin resistance, and body mass index

Association of higher variation in fasting plasma glucose and post-prandial blood glucose with lower cognitive function across groups at baseline (fully adjusted analyses). Composite score and MMSE declined in glibenclamide but not in repaglinide group during trial (analyses controlling only for HbA1c and variation in fasting plasma glucose).

Bruce et al. [14] (2008)

Patients with type 2 diabetes participating in the Fremantle Diabetes Study; Australia

8-year retrospective, observational

Mean 76 ± 5 years

HbA1c at baseline and 8 years earlier

Dementia and MCI identified from screening instruments/clinical interview

None

No association

Bruce et al. [21] (2008)

Patients with type 2 diabetes participating in the Fremantle Diabetes Study; Australia

8-year retrospective, 2-year prospective, observational.

Mean 75 ± 4 years

HbA1c 8 years prior to baseline cognitive assessment

Dementia and MCI identified from screening instruments/clinical interview at baseline and at 2-year follow-up. ‘Cognitive decline’ defined as downward conversion between ‘normal’, MCI, and dementia.

None

No association

Chen et al. [9] (2011)

Patients with type 2 diabetes; China

Cross-sectional, observational

Mean 63 ± 8 years

HbA1c

MCI identified on the basis of cognitive screening instrument

None

Higher HbA1c in group with MCI compared with group free of MCI

Chen et al. [11] (2012)

Patients with type 2 diabetes; China

Cross-sectional, observational

Mean 55 ± 7 years

HbA1c

MCI identified on the basis of cognitive screening instrument

None

No association

Cukierman-Yaffe et al. [13] (2009)

Patients with type 2 diabetes participating in ACCORD-MIND; North America

Cross-sectional analysis of trial on blood pressure, lipids, and glycemic control

Mean 63 ± 6 years

HbA1c and fasting plasma glucose

Digit Symbol Coding (primary outcome), MMSE, Rey Auditory Verbal Learning, and Stroop (secondary outcomes)

Total of 18 demographic and clinical risk factors

Higher HbA1c associated with lower Digit Symbol Coding. Findings for other cognitive tests did not survive full adjustment. No findings for fasting plasma glucose.

Launer et al. [28] (2011)

Patients with type 2 diabetes participating in ACCORD-MIND trial, with HbA1c targets of <6.0 % versus 7.0 % to 7.9 %; North America

40-month trial on blood pressure, lipids and glycemic control

Mean 62 ± 6 years

Successful manipulation of glycemic control. (Treatment groups differed in glycemic control following intervention.)

Total brain volume and abnormal white matter at baseline and 40 months. Digit Symbol Coding (primary outcome), MMSE, Rey Auditory Verbal Learning, and Stroop (secondary outcomes) at baseline and 20 and 40 months.

Second trial assignment (lipid or blood pressure trials), group allocation within second trial assignment, clinical center, and history of cardiovascular disease

No difference in 20- or 40-month cognitive decline between intervention groups (fully adjusted analyses). Total brain volume declined at slower rate in intensively treated compared with standard treatment groups (independent of adjustment variables and of age, sex, duration of diabetes, Digit Symbol Coding). Greater abnormal white matter in intensively treated compared with standard treatment group at 40 months.

Manschot et al. [20] (2006)

Patients with type 2 diabetes participating in the Utrecht Diabetic Encephalopathy Study; The Netherlands

Cross-sectional, observational

Mean 66 ± 6 years

HbA1c

Composite scores on five cognitive domains from 11 cognitive tests, estimate of pre-morbid ability, cortical atrophy, and white matter lesions

Age, sex, and estimated pre-morbid ability

Association of higher HbA1c with steeper estimated lifetime decline in processing speed. No association of HbA1c with brain imaging data.

Manschot et al. [25] (2007)

Patients with type 2 diabetes participating in the Utrecht Diabetic Encephalopathy Study; The Netherlands

Cross-sectional, observational

Mean 66 ± 6 years

HbA1c and plasma insulin

Composite score from 11 cognitive tests, estimate of pre-morbid ability, cortical atrophy, and white matter lesions

Age, sex, and estimated pre-morbid ability

Association of higher HbA1c with steeper estimated lifetime decline in overall cognitive function. Association of higher insulin with greater severity of white matter lesions.

Ravona-Springer et al. [27] (2014)

Patients with type 2 diabetes participating in the Israel Diabetes and Cognitive Decline Study; Israel

12-year retrospective observational

Mean 73 ± 5 years

Data on HbA1c from diabetes register (mean 18 ± 10 measurements per patient). Six HbA1c trajectories identified (for example, high/increasing and high/stable).

Measured at 12 years only: CDR, MMSE, and battery of seven cognitive tests. Sum of z-scores calculated for four cognitive domains.

Age, sex, education, cardiovascular disease, years in diabetes register, anti-diabetes treatment, and depression

Associations of HbA1c trajectories with level of overall cognitive function, semantic categorization, and executive function. Relatively poorest cognitive function in high/decreasing group and high/increasing groups. Relatively highest performance in low/stable group.

Ryan et al. [29] (2006)

Patients with type 2 diabetes, receiving rosiglitazone or glibenclamide; USA

24-week trial

Mean 60 ± 1 years

Successful manipulation of glycemic control and insulin sensitivity. (Treatment groups differed in fasting plasma glucose and fasting serum insulin following intervention.)

CANTAB, Digit Symbol Coding, Rey Auditory Verbal Learning, and estimate of pre-morbid ability

Age, center, pre-morbid ability, and baseline measurement of fasting plasma glucose/insulin

Cognitive function improved equally in both treatment groups (fully adjusted analysis). Correlation of reduction in fasting plasma glucose with improvements in working memory across groups (unadjusted analysis). No finding for insulin.

Seaquist et al. [41] (2013)

Patients with type 2 diabetes participating in ACCORD-MIND, with HbA1c targets of less than 6.0 % versus 7.0 % to 7.9 %; North America

40-month trial on blood pressure, lipids, and glycemic control

Mean 62 ± 6 years

Treatment with insulin at enrolment and during trial

Digit Symbol Coding at baseline and 20 and 40 months

Total of 21 demographic, lifestyle, and clinical covariates

Association of insulin use at enrolment with lower baseline cognitive function. Loss of statistical significance upon full adjustment. No association of insulin use during trial with 40-month cognitive decline in standard treatment group. Association of insulin use with steeper 40-month cognitive decline in intensive treatment group; loss of statistical significance upon full adjustment.

Umegaki et al. [16] (2014)

Patients with type 2 diabetes; Japan

6-year prospective, observational

Mean 74 ± 5 years

Mean of HbA1c and plasma immunoreactive insulin at baseline and annual follow-ups.

Composite score from MMSE, Digit Symbol Coding, Stroop, and word recall. Analyses of ‘decliners’ versus ‘non-decliners’ on bases of composite score and individual cognitive tests.

None

No associations for HbA1c or insulin, except for higher mean 6-year insulin in ‘decliners’ compared with ‘non-decliners’ on Stroop.

Yanagawa et al. [15] (2011)

Patients with diabetes receiving exercise program four times/week versus none; Japan

12-week trial on physical exercise intervention

Mean 71 ± 4 years

HbA1c, fasting blood glucose, GIR, MCR in euglycemic clamp, and immunoreactive insulin. (No intervention effect on any of these measurements.)

MMSE, word recall, Digit Symbol Coding, Stroop, and Trail-Making Test

Age, education, and body mass index

No difference in cognitive function between groups following intervention. Across groups, changes in HbA1c and changes in GIR correlated with changes in word recall. Changes in fasting blood glucose correlated with changes in Trail-Making.