The Relationship Between Glucose Control and Cognitive Function in People With Diabetes After a Lacunar Stroke

Abstract

Context

Lacunar strokes and diabetes are risk factors for cognitive dysfunction. Elucidating modifiable risk factors for cognitive dysfunction has great public health implications. One factor may be glycemic status, as measured by glycated hemoglobin (A_1c).

Objective

The aim of this study was to assess the relationship between A_1c and cognitive function in lacunar stroke patients with diabetes.

Methods

The effect of baseline and follow-up A_1c on the baseline and the change in Cognitive Assessment Screening Instrument (CASI) score over time among participants with a median of 2 cognitive assessments (range, 1-5) was examined in 942 individuals with diabetes and a lacunar stroke who participated in the Secondary Prevention of Small Subcortical Strokes (SPS3) trial (ClinicalTrials.gov No. NCT00059306).

Results

Every 1% higher baseline A_1c was associated with a 0.06 lower standardized CASI z score (95% CI, –0.101 to –0.018). Higher baseline A_1c values were associated with lower CASI z scores over time (P for interaction = .037). A 1% increase in A_1c over time corresponded with a CASI score decrease of 0.021 (95% CI, –0.0043 to –0.038) during follow-up. All these remained statistically significant after adjustment for age, sex, education, race, depression, hypertension, hyperlipidemia, body mass index, cardiovascular disease, obstructive sleep apnea, diabetic retinopathy, nephropathy insulin use, and white-matter abnormalities.

Conclusion

This analysis of lacunar stroke patients with diabetes demonstrates a relationship between A_1c and change in cognitive scores over time. Intervention studies are needed to delineate whether better glucose control could slow the rate of cognitive decline in this high-risk population.

Strokes and diabetes are strong risk factors for dementia and cognitive dysfunction (1-5). Lacunar infarcts account for about 25% of strokes, and approximately 50% of individuals who have a lacunar stroke develop cognitive impairment over subsequent years (6-8). Compared to those without diabetes, people with type 2 diabetes have a 1.5- to 2-fold higher risk of cognitive dysfunction and dementia and an accelerated rate of cognitive decline (2-5) and a 2-fold greater risk for lacunar strokes (9, 10). Understanding potentially modifiable risk factors for cognitive dysfunction in people with type 2 diabetes who experienced a lacunar infarct may have important therapeutic implications. One important factor may be glycemic status as measured by glycated hemoglobin (A_1c). Studies in older people with diabetes have demonstrated a relationship between higher A_1c values and a greater risk for cognitive impairment and dementia and an accelerated rate of cognitive decline (11-16). Less is known regarding this relationship in people with diabetes after stroke and more specifically after a lacunar stroke, which is a known strong risk factor for cognitive impairment.

The Secondary Prevention of Small Subcortical Strokes (SPS3) trial included 1106 individuals with diabetes who had a lacunar infarct (6, 17, 18). Data on cognitive function and A_1c were collected at baseline and at several time points throughout the study. The post hoc epidemiologic analysis assessing the relationship between glycemic status (A_1c) and cognitive function is reported therein.

Materials and Methods

The present study is a post hoc epidemiological analysis of data collected during the SPS3 randomized trial (see supplementary material for the STROBE checklist) (19). The SPS3 trials was a secondary stroke prevention trial conducted in 8 countries between 2001 and 2011. The rationale, design, and main results as well as cognitive results have been previously reported (6, 17, 18, 20, 21). In brief, the study included individuals age 30 years or older with a recent magnetic resonance imaging (MRI)-defined small subcortical ischemic stroke (lacunar stroke) who were randomly assigned to 2 interventions: a) antiplatelet therapy (aspirin + clopidogrel vs aspirin + placebo); and b) blood pressure target control (higher: 130-149 mm Hg systolic vs lower: < 130 mm Hg systolic). Patients with cortical stroke, cardioembolic disease, amendable carotid stenosis, disabling stroke (modified Rankin scale ≥ 4), or significant cognitive impairment (Mini-Mental State Exam [MMSE] > 2 SD below mean for age and education, ie, adjusted score < 24, the generally accepted cutoff for mild dementia) were excluded. Diabetes was defined as at least one of the following: a self-reported history of diabetes; prescribed glucose-lowering medications; documented evidence of an elevated glucose level; or a diagnosis of diabetes within 3 months of study entry. Patients were followed over time for potential stroke and other cardiovascular events, and A_1c and the SPS3 neuropsychological tests (NPTs) were collected at baseline and then annually throughout the study. These analyses included individuals who had a baseline and A_1c value, MRI data, and cognitive data through 5 years of follow-up (Supplementary Fig. S1) (19).

Cognitive/Neuropsychological Assessment

The SPS3 NPTs have been previously described (6). Testing included the following: 1) an assessment of global cognitive function (the Cognitive Assessment Screening Instrument [CASI; primary outcome]; 2) a test of episodic memory (the California Verbal Learning Test: Scores for key subscales were calculated as short delay free/cued recall, delayed free/cued recall, and discriminability; 3) a test of visual construction (the Block Design; 4) a test of perceptual speed (the Symbol Search); 5) a test of motor dexterity (the Grooved Pegboard; 6) a test of verbal fluency (the Controlled Oral Word Association; 7) a test of attention (the Digit Span); and 8) a test of executive function (the Clock Drawing to Command). The reliability and validity of these tests have been previously described (22-26). SPS3-certified examiners blinded to study intervention administered each test at annual visits in the language spoken by the participant (English, Spanish, or French). Cognitive testing was conducted at study entry and annually during the first 5 years of each participant’s follow-up. SPS3 participants were enrolled over a rolling period and followed for a minimum of 1 year to a common termination date. Thus, participants had a median of 2 cognitive assessments (range, 1-5). The number of individuals with a) cognitive test scores and baseline A_1c (S1A) and b) cognitive test scores and time-varying A_1c at each follow-up visit (S1B) are depicted in Supplementary Table S1 (19).

Other Variables

Glycemic status of the participants was evaluated at baseline and throughout the study by local site A_1c measurement. Baseline variables were defined as follows: Education was categorized into 4 groups (0-4, 5-8, and 9-12 years of education, or any college training), and depression was defined as either a Patient Health Questionnaire-9 score greater than 9 or the use of antidepressant medications (selective serotonin reuptake inhibitor, new-generation tricyclic antidepressants) Ethnicity was categorized as either Hispanic White, non-Hispanic White, Black, or other. Medical history was collected by the study physician and included history of hypertension, hyperlipidemia, diabetic retinopathy, nephropathy, cardiovascular disease (CVD, defined as myocardial infarction, angina, coronary revascularization), and obstructive sleep apnea (OSA). Smoking and alcohol use were based on self-report, and the post-stroke level of disability was assessed using the modified Rankin scale and the Barthel index. White-matter abnormalities (WMAs) on baseline MRI were characterized using the age-related white-matter changes scale, and was categorized as none (0), mild (1-4), moderate (5-8), and severe (≥ 9).

Statistical Analysis

This analysis were limited to those participants who had a baseline A_1c, cognitive testing, and MRI data (Supplementary Fig. S1) (19). As previously described (20), all raw NPTs were converted into z scores with reference to the best available normative data standardized when possible for relevant factors including age, education, language of administration, and geographic region. Continuous variables were summarized using either means and SD or as medians with interquartile ranges and compared using t tests. Categorical variables were summarized as counts and percentages and compared using a chi-square test.

The association between baseline A_1c and baseline cognitive function was assessed using linear regression both before and after multivariable adjustment using 6 incremental models. Model 1 included adjustment for age, sex, education, race/ethnicity; model 2 included the earlier mentioned variables and depression; model 3 added hyperlipidemia, hypertension, and BMI; model 4 added CVD; model 5 added OSA; and model 6 added WMA, retinopathy, neuropathy, and insulin use.

The effect of baseline A_1c on cognitive scores over time was assessed using linear mixed models that accounted for shared variability among the repeated cognitive measures. Of primary interest in this analysis was the interaction between baseline A_1c and time, which allowed us to assess whether the effect of baseline A_1c affected changes in cognitive scores over time. We used the A_1c to determine the appropriate covariance structure and the Toeplitz covariance structure was chosen, which allows the same correlation among any 2 time points the same distance apart (eg, times 1 and 2 and times 3 and 4 would have the same correlation; times 1 and 3 and times 2 and 4 would have the same correlation). The Toeplitz covariance structure is well suited for repeated measures that are collected on a regular time schedule, such as our data, which were collected annually. This analysis was conducted both before and after adjustment for the variables in models 1 to 6 described earlier. The same modeling approach was repeated including A_1c as a time-varying covariate.

Results

A total of 954 of 1106 (86%) individuals with diabetes had baseline CASI and A_1c values; 12 were missing WMA scores, and thus 942 are included in this analysis. At baseline people with a higher A_1c were younger, more likely to be male, have diabetic retinopathy, be insulin users, be more disabled, have more WMAs, and were less likely to be non-Hispanic Whites than those with a lower A_1c (Table 1). Baseline medication use (antidiabetic and other) is provided in Supplementary Table S2 (19). Thirty-one percent were insulin users at baseline, with 73% of these remaining on insulin throughout the study. Of those who were not on insulin at baseline, 11% were on insulin at some point or points during their follow-up.

Table 1.

Baseline characteristics

Variable	Total	A1c < 7.5%	A1c ≥ 7.5
No.	954	388	566
Age, y	62.6 (9.7)	65.3 (9.7)	60.8 (9.3)a
Female	350 (36.7)	127 (32.7)	223 (39.4)b
Education, y
0-4	114 (10.6)	41 (10.6)	73 (12.9)
5-8	178 (18.7)	73 (18.8)	105 (18.5)
9-12	369 (38.7)	155 (39.9)	214 (37.8)
Any college	293 (30.7)	119 (30.7)	174 (30.7)
Ethnicity
Hispanic	342 (35.8)	115 (29.6)	227 (40.1)
Non-Hispanic White	423 (44.3)	220 (56.7)	203 (35.9)
Black	170 (17.8)	44 (11.3)	126 (22.3)
Other/multiple	19 (2.0)	9 (2.3)	10 (1.8)a
Depression	180 (18.9)	77 (19.8)	103 (18.2)
HTN	883 (92.6)	360 (92.8)	523 (92.4)
Hyperlipidemia	588 (61.6)	233 (60.0)	355 (62.7)
BMI	30 (6.3)	29.9 (6.6)	30.1 (6.1)
CVD	154 (16.1)	70 (18.0)	84 (14.8)
OSA	50 (5.2)	24 (6.2)	26 (4.6)
A1c	8.3 (2.1)	6.4 (0.9)	9.6 (1.7)a
Duration of diabetes, y	11.7 (9.1)	10.4 (9.2)c	12.5 (8.9)c
Diabetic retinopathy	122 (12.8)	32 (8.2)	90 (15.9)b
Diabetic nephropathy	93 (9.7)	37 (9.5)	56 (9.9)
Insulin use	298 (31.2)	81 (20.9)	217 (38.3)a
MMSE	27.8 (2.4)	27.8 (2.3)	27.8 (2.5)
Modified Rankin	1.5 (0.8)	1.4( 0.8)	1.5 (0.8)a
Barthel index	93.8 (11.4)	94.5 (11)	93.4 (11.7)
WMAsc
None	49 (5)	20 (5)	29 (5)a
Mild	434 (46)	143 (37)	291 (52)
Moderate	270 (29)	125 (33)	145 (26)
Severe	189 (20)	95(25)	94 (17)

Abbreviations: A_1c, glycated hemoglobin; BMI, body mass index; CVD, cardiovascular disease (myocardial infarction or angina or coronary revascularization); HTN, hypertension; MMSE, Mini-Mental State Examination; OSA, obstructive sleep apnea; WMAs, white-matter abnormalities.

^a P less than .001.

^b P less than .05.

^c Ninety-six participants had missing duration of diabetes: 41 from the A_1c less than 7.5% group; 55 from the 7.5% or greater group.

Cross-Sectional Relationship Between Glycated Hemoglobin and Neuropsychological Test Scores

A 1% higher baseline A_1c was associated with a 0.06 lower baseline standardized CASI z score (95% CI, –0.101 to –0.018). This relationship remained statistically significant after adjusting for age, sex, education, race, depression, hypertension, hyperlipidemia, BMI, CVD, OSA, diabetic retinopathy, nephropathy, insulin use, and WMAs (model 6; β coefficient [–0.062, 95% –0.1042 to –0.0193]) (Fig. 1A). There were no statistically significant associations between other NPT scores and baseline A_1c.

Figure 1.

A, The relationship between baseline glycated hemoglobin (A_1c) and baseline Cognitive Assessment Screening Instrument (CASI) z score. B, The relationship between the change in A_1c and change in CASI score over time after adjusting for the following variables: model 1—age, sex, education, and race; model 2—M1 + depression; model 3—M2 + hyperlipidemia, hypertension, body mass index (BMI); model 4—M3 + cardiovascular disease (CVD); model 5—M3 + obstructive sleep apnea (OSA); model 6—M3 + retinopathy, neuropathy, insulin use, white-matter abnormalities (WMA).

Relationship Between Baseline Glycated Hemoglobin and Neuropsychological Test Scores Over Time

Fig. 2 depicts the change in CASI z score at different time points across several baseline A_1c values. There was a statistically significant relationship between baseline A_1c and CASI z score over time (P for interaction = .037), and this relationship persisted after multivariable adjustment (P for interaction = .037) Among the other NPT scores, this relationship was noted only for the discriminability score (model 6 P for interaction = .028).

Figure 2.

The distribution of Cognitive Assessment Screening Instrument (CASI) z scores over time (y axis) according to baseline A_1c (x axis).

Relationship Between Glycated Hemoglobin Over Time and Neuropsychological Test Scores Over Time

A 1% increase in A_1c over time corresponded with a CASI score decrease of 0.021 (95% CI, –0.0043 to –0.038) during follow-up (Fig. 1B). This relationship remained statistically significant after adjusting for age, sex, education, race, depression, hypertension, hyperlipidemia, BMI, CVD, OSA, diabetic retinopathy, nephropathy insulin use, and WMAs (model 6; β coefficient [–0.021, 95% CI, –0.0045 to –0.037]). Using the model 6 adjustments for every 1% increase in A_1c, there was a –0.033 (95% CI, –0.0036 to –0.0327) change in the z score of the clock making test (executive function), a –0.018 change in discriminability (95% CI, –0.035 to –0.00075), and a –0.016 (95% CI, –0.0048 to –0.027) change in the z score of the controlled oral word association (verbal fluency). The associations between A_1c over time and other NPT scores were not statistically significant (Fig. 3).

Figure 3.

The relationship between the change over time in glycated hemoglobin (A_1c) and the change in z scores of several neuropsychological tests after adjustment for age, sex, education, race, hyperlipidemia, hypertension, body mass index, retinopathy, neuropathy, insulin use, and white-matter abnormalities.

Discussion

This analysis of 942 individuals with diabetes who suffered a lacunar infarct demonstrates an inverse relationship between baseline A_1c and baseline cognitive function, subsequent change in cognitive scores over time, and between change in A_1c and the subsequent change in cognitive test scores over time. The fact that these relationships persisted after adjusting for many potentially confounding variables supports the hypothesis of an independent relationship between glucose control and the change in cognitive function in people with diabetes after a lacunar infarct.

Previous studies in people with diabetes support these findings. In a 10-year longitudinal study of 187 people with diabetes, a 1% increment in A_1c was associated with faster decline in memory (14). In a prospective study that included 232 participants with diabetes older than 65 years, those whose mean glucose level was 190 mg/dL (10.5 mmol/L) had a 1.4 times greater risk of dementia during 6.8 years of follow-up than those whose mean glucose level was 160 mg/dL (8.9 mmol/L) (27). Moreover, in a cohort of 717 people with diabetes followed for an average of 9 years there was an association between higher mean A_1c levels and lower mean cognitive scores (11). Finally, in another study of 575 individuals who were assessed 24 months following a stroke or transient ischemic attack, type 2 diabetes and a higher A_1c level were both independent risk factors for lower memory, executive function, attention, and total cognitive score (28). To the best of our knowledge, this is the first study to demonstrate this relationship in individuals with diabetes after a lacunar infarct.

This study has several limitations. First, despite adjustment for many variables the possibility of residual confounding cannot be eliminated. For example, the analysis did not adjust for diabetes duration or different drug classes used by participants Second, not all individuals had follow-up A_1cs measured, and not all individuals had all cognitive test scores at follow-up. Indeed, a recent study conducted in poststroke participants has demonstrated that including only those with in-clinic visits underestimates cognitive impairment rates (29). Thus, it may well be that the observed relationship is actually an underestimation of the true association. Third, because SPS3 participants were individuals able to take part in a randomized controlled trial, with a mean age of 62 years and a mean baseline MMSE of 27.8, it is not clear whether these results may apply to a broader older population with more severe cognitive impairment. Additionally, follow-up was relatively short, thus limiting the ability to detect substantial cognitive decline. Finally a clinical correlate (dementia, mild cognitive impairment) of the decline observed in cognitive test scores is not provided; however, previous studies have demonstrated a relationship between the rate of decline and subsequent risk for dementia (30). This study also has several strengths, including a well-characterized cohort of individuals with small subcortical strokes verified by MRI and the detailed cognitive assessment conducted.

There are several explanations for the observed relationship. First, it may be that individuals with cognitive impairment have difficulties managing their disease and thus have worse glucose control. Indeed, previous studies have shown that cognitive dysfunction impedes self-care management capacities (31-33). The fact that a consistent relationship was observed mainly for a test of general cognitive abilities and not for tests of specific cognitive domains may strengthen this explanation. Second, hyperglycemia may accelerate the rate of cognitive decline by either reducing capillary perfusion (34, 35), accelerating larger vessel disease, or directly damaging the brain. Indeed, postmortem studies have revealed increased numbers of senile plaques and other neuropathological markers of Alzheimer dementia and metabolic oxidation products associated with hyperglycemia (36, 37).

To summarize, this analysis of 942 individuals with diabetes after a lacunar stroke demonstrates a relationship between higher A_1c and greater decline in cognitive test scores over time. More studies are needed to confirm these results, and intervention studies are needed to delineate whether better glucose control could slow the rate of cognitive decline in this high-risk population.

Glossary

Abbreviations

A_1c

glycated hemoglobin

BMI

body mass index

CASI

Cognitive Assessment Screening Instrument;

CVD

cardiovascular disease

MMSE

Mini-Mental State Exam

MRI

magnetic resonance imaging;

NPT

neuropsychological test

OSA

obstructive sleep apnea

SPS3

Secondary Prevention of Small Subcortical Strokes

WMAs

white-matter abnormalities

Additional Information

Disclosures: T.C.Y. reports receipt of speaker honoraria from Sanofi, Eli Lilly, Medtronic, and MSD. L.A.M. has nothing to disclose. T.R. has nothing to disclose. J.B. is on advisory boards for Bayer. H.C.G. holds the McMaster-Sanofi Population Health Institute Chair in Diabetes Research and Care. He reports research grants from Eli Lilly, AstraZeneca, Merck, Novo Nordisk and Sanofi; honoraria for speaking from AstraZeneca, Boehringer Ingelheim, Eli Lilly, Novo Nordisk, and Sanofi; and consulting fees from Abbott, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Merck, Novo Nordisk, Janssen, Sanofi, Kowa and Cirius; MS Advisory board, consulting and research: Bayer and BMS; and consulting: Daiichi Sankyo. Honoraria: Servier, Bayer, and BMS. O.B. has nothing to disclose.

Data Availability

Publicly available SPS3 data can be found at https://www.ninds.nih.gov/Current-Research/Research-Funded-NINDS/Clinical-Research/Archived-Clinical-Research-Datasets. The data sets generated during and/or analyzed during the present study are not publicly available but are available from the corresponding author on reasonable request.