The Relationship Between the Score on a Simple Measure of Cognitive Function and Incident CVD in People With Diabetes: A Post Hoc Epidemiological Analysis From the ACCORD-MIND Study

Abstract

Context and Objective:

Diabetes is associated with a greater risk for incident cardiovascular disease and cognitive dysfunction. This study aimed to investigate, in people with type 2 diabetes, the association of a simple measure of cognitive function to cardiovascular disease events and mortality.

Design, Setting, Participants, Measurements, and Outcomes:

The Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial included persons with longstanding type 2 diabetes. A substudy of 2977 (Memory in Diabetes) participants aged 55 years or older aimed to test the effect of the interventions on brain structure and function. At baseline, participants were administered a cognitive battery that included the digit symbol substitution test (DSST). The associations of the DSST and the ACCORD primary outcome (the first occurrence of nonfatal myocardial infarction or nonfatal stroke or death from cardiovascular causes) and all-cause mortality were investigated with Cox proportional hazard models adjusting for several demographic and clinical variables.

Results:

Median follow-up time was 4.27 years. An inverse relationship between the incidence of the ACCORD primary outcome and baseline cognitive score was demonstrated. A 1-point higher DSST score was associated with a lower incidence of the primary outcome (hazard ratio, 0.987; 95% confidence interval, 0.977 to 0.998; P = 0.019), after adjustment for demographic and clinical trial factors, additional baseline cardiovascular risk factors, and self-reported need for assistance to follow the protocol.

Conclusion:

Lower scores on the DSST, a simple, sensitive neuropsychological instrument, are associated with a higher incidence of cardiovascular events in persons >55 years old with longstanding diabetes.

Using data from the ACCORD-MIND trial, this study demonstrates in older people with T2D a relationship between a simple measure of cognitive function (the DSS) and incident cardiovascular disease.

Diabetes is a strong, independent risk factor for cardiovascular outcomes and death from all causes. Epidemiological studies have shown that diabetes and the degree of hyperglycemia are inversely associated with measures of cognitive function (1–4) and predict greater cognitive decline in people with diabetes compared with unaffected individuals (1, 5–7). Other epidemiological studies have reported an inverse relationship between cognitive function in both children and adults, as well as the future risk of cardiovascular events (8, 9) or death (10–12). These studies were conducted in population cohorts. It is not known within the population of persons with type 2 diabetes (T2D) whether cognitive function within the “normal range” is inversely related to either cardiovascular events or all-cause mortality.

The Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial was a US and Canadian cardiovascular outcomes trial that assessed the effect of glucose, blood pressure, and lipid-lowering therapy on serious health outcomes in 10,251 middle-aged adults with T2D (13). At the time of randomization, 2977 of these individuals consented to participate in the ACCORD–Memory in Diabetes (MIND) substudy, which included measurement of cognitive function at baseline and periodically during follow-up and did not detect a statistically significant cognitive effect of intensive glucose lowering (14). The primary cognitive measurement for this substudy was the digit symbol substitution test (DSST). This test is a simple and quickly administered instrument that detects cognitive differences in working memory and processing speed with a good distribution in a cognitively intact population. It has been extensively used to measure cognitive function in people with and without diabetes, with lower scores predicting mortality, future cognitive dysfunction, and disability (11, 15, 16). In this prospective analysis, we report the relationship between baseline DSST scores and the subsequent risk of cardiovascular events or death in ACCORD participants with T2D and other cardiovascular risk factors.

Methods

Design

The design and results of the ACCORD trial (17) and the ACCORD-MIND substudy (14) have been reported previously. In brief, participants aged 40 to 79 years with T2D, glycosylated hemoglobin (HbA1C) ≥7.5%, and additional risk factors for cardiovascular outcomes were recruited from 77 sites in the United States and Canada between 21 August 2003 and 16 December 2005. Individuals were excluded if they had frequent or serious hypoglycemia in the previous year or had other serious illness. All 10,251 participants were randomized to either intensive glycemic therapy targeting A1C <6.0% or to standard glycemic therapy targeting HbA1C 7.0% to 7.9%. In a double 2 × 2 factorial design, 4733 of the participants were enrolled in a blood pressure–lowering trial, and 5518 participants were enrolled in a lipid therapy trial. The ACCORD-MIND substudy (18) included 2977 participants aged >55 years and fluent in either English or Spanish.

Measurement of cognitive function at baseline

Certified technicians administered and scored a 30-minute battery comprising four cognitive tests, which were administered to each participant at baseline, after 20 months, and again after 40 months (14).

The primary cognitive measure for the ACCORD-MIND study was the DSST, a subset of the third edition of the Wechsler Adult Intelligence Scale (19, 20). This test detects cognitive differences in working memory and processing speed in generally healthy people; it has been extensively used to measure cognitive function in people with and without diabetes, and lower scores have been demonstrated to predict mortality, future cognitive dysfunction, and disability (11, 15, 16). The DSST consists of rows of nine randomly ordered numbers with a blank square underneath and a key at the top of the page that pairs a different symbol with each of the numbers 1 to 9, respectively. Respondents fill in the blank space under each number with the corresponding symbol as quickly as possible over a 2-minute interval. The achieved score is the number of correct digit-symbol pairs within this period; the maximum score is 133.

Other measurements

The ACCORD primary outcome was the first occurrence of nonfatal myocardial infarction or nonfatal stroke or death from cardiovascular cause (17). Death from any cause was a prespecified secondary outcome. Key outcomes were adjudicated by a central committee, whose members were unaware of study group assignment. Biochemical characteristics were measured in the central laboratory of the ACCORD main trial. HbA1C was measured by an automated high-performance liquid chromatography TOSOH G7 (TOSOH Bioscience), and fasting plasma glucose was measured enzymatically on the Hitachi 917 autoanalyzer.

Statistical analysis

ACCORD-MIND enrolled 2977 participants (Table 1), of whom 2957 had a baseline DSST measurement. Twenty participants did not have follow-up for cardiovascular outcomes; thus, these analyses were conducted on 2937 participants. To describe the sample, continuous variables were summarized using means with standard deviations, and binary/categorical variables were summarized using counts and percentages.

Table 1.

Baseline Characteristics of ACCORD-MIND Participants Stratified by Incident Outcomes

Variable	Total (n = 2977)	No Primary (n = 2857)	Primary (n = 100)	P Value	No Death (n = 2845)	Death (n = 132)	P Value
Age, y	62.5 ± 5.8	62.5 ± 5.8	63.4 ± 6.1	0.0948	62.3 ± 5.8	65.6 ± 6.4	<0.0001
Female	1388 (46.6)	1336 (46.8)	43 (43.0)	0.4585	1341 (47.1)	47 (35.61)	0.0094
White	2087 (70.1)	1999 (70.0)	80 (80.0)	0.0309	1991 (70.0)	96 (72.7)	0.5007
Education < high school	392 (13.2)	368 (12.9)	17 (17.0)	0.2289	371 (13.0)	21 (15.9)	0.3407
Current smoking	352 (11.8)	330 (11.6)	18 (18.0)	0.0491	335 (11.8)	17 (13)	0.7010
Alcohol drinks/wk	0.91 ± 2.84	0.9 ± 2.9	1.1 ± 2.7	0.5891	0.91 ± 2.84	0.91 ± 2.71	0.9861
History of CVD	869 (29.2)	818 (28.6)	42 (42.0)	0.0038	811 (28.5)	58 (43.9)	0.0001
Duration of diabetes	10.4 ± 7.4	10.3 ± 7.3	12.9 ± 8.3	0.0006	10.3 ± 7.28	11.92 ± 8.7	0.0148
CHF	150 (5.0)	135 (4.7)	14 (14.0)	<0.0001	127 (4.46)	23 (17.4)	<0.0001
Amputation	45 (1.5)	38 (1.3)	7 (7.0)	<0.0001	41 (1.44)	4 (3.0)	0.1435
Serum creatinine	0.9 ± 0.2	0.9 ± 0.2	0.9 ± 0.2	0.0027	0.9 ± 0.2	1.0 ± 0.3	<0.001
Albumin to creatinine ratio				<0.0001			0.0114
<30	2118 (71.6)	2081 (72.8)	45 (45.0)		2038 (72.1)	80 (61.1)
30 to ≤300	657 (22.2)	617 (21.6)	34 (34.0)		620 (21.9)	37 (28.2)
>300	182 (6.2)	159 ( 5.6)	21 (21.0)		168 (5.9)	14 (10.7)
HbA1c	8.3 ± 1.1	8.3 ± 1.1	8.6 ± 1.0	0.0004	8.3 ± 1.1	8.5 ± 1.1	0.0184
SBP	136 ± 18	135.4 ± 17.7	140.5 ± 20.4	0.0048	136 ± 18	136 ± 20	0.7050
DBP	75 ± 11	74.8 ± 10.6	75.4 ± 12.0	0.5810	75 ± 11	73 ± 12	0.0314
BMI	33.0 ± 5.4	33.0 ± 5.3	32.9 ± 5.6	0.9402	33.0 ± 5.4	32.8 ± 5.2	0.7748
LDL	103.5 ± 33.7	103.4 ± 33.6	106.9 ± 35.4	0.3042	103.7 ± 33.7	98.9 ± 33.5	0.1083
Lipid trial	1538 (51.7)	1476 (51.7)	51 (51.0)	0.8963	1466 (51.3)	72 (54.6)	0.4978
DSST	52.6 ± 15.9	52.7 ± 15.9	48.4 ± 15.2	0.0074	52.8 ± 15.9	47.1 ± 14.4	<0.0001
Stroop	32.0 ± 16.7	31.9 ± 16.7	34.6 ± 15.9	0.1215	31.7 ± 16.6	37.7 ± 18.0	0.0001
Stroop-Winsorized	31.2 ± 13.0	31.1 ± 12.9	34.0 ± 14.3	0.0319	31.8 ± 12.9	36.4 ± 14.0	0.0001
RAVLT	7.5 ± 2.5	7.5 ± 2.6	7.0 ± 2.3	0.0283	7.5 ± 2.5	7.0 ± 2.6	0.0209

Data are shown as n (%) for counts and mean ± standard deviation for continuous data.

Abbreviations: BMI, body mass index; CHF, congestive heart failure; DBP, diastolic blood pressure; LDL, low-density lipoprotein; RAVLT, Rey Auditory Verbal Learning Test; SBP, systolic blood pressure.

The relationship between the baseline DSST score and both the ACCORD primary outcome and total mortality were analyzed according to categorically defined quartiles of the DSST score, as well as the measured DSST score expressed as a continuous variable. Baseline scores were chosen as they are not affected by the competing risk of death, cardiovascular outcomes, or nonrandom missingness. ACCORD primary outcome-free rates across the DSST quartiles were compared using Kaplan-Meier curves and log-rank test. Cox proportional hazard models were constructed using the ACCORD primary cardiovascular outcome and total mortality as dependent variables. A combined measure of cognitive status in ACCORD has been described previously (18). It was derived by adding the standardized score of the Digit Symbol Substitution (DSS), Stroop, and Rey Auditory Verbal Learning Test, each of which was calculated by subtracting the mean baseline score and dividing the results by the baseline standard deviation. As the distribution of the Stroop variable was highly skewed, a “winsorized Stroop” (i.e., Stroop scores that were below the 5th or above the 95th percentile were reduced to the 5th or 95th percentile, respectively) (21) and a combined measure including the winsorized Stroop were calculated for use in sensitivity analyses. All analyses described above were repeated after substituting these two combined measures for the DSS score. The base model (model 1) was adjusted for age, sex, race, education, history of cardiovascular disease (CVD), participation in a lipid vs blood pressure trial, allocation to fibrate vs placebo in the lipid trial, allocation to intensive vs standard blood pressure in the blood pressure trial, and allocation to intensive vs standard glucose control. Subsequent models included smoking, prior heart failure, prior amputation, site belonging to an integrated health plan, and baseline HbA1C, body mass index, serum creatinine, albumin/creatinine ratio category (model 2), and these additional variables plus a time-dependent covariate representing self-reported need for assistance to follow the protocol were included (model 3). Sensitivity analyses were conducted for all models in which the cardiovascular mortality component of the primary outcome was replaced by mortality from any cause. The proportional hazards assumption for the DSST score variable was tested by entering the interaction between the DSST variable and log-transformed time into the models. The base model was also used to determine whether the relationship between DSST and outcomes differed across predefined subgroups, including age (≤65 vs >65 years), sex, education (high school graduate vs less then high school graduate), HbA1C concentration (≤8% vs >8%), prior congestive heart failure, or prior CVD.

Results

Baseline characteristics

During a median follow-up of 4.27 years, participants in this subgroup sustained 132 deaths (66 in each glycemia arm) and 237 primary end points (137 in the standard arm and 100 in the intensive arm). As noted in Table 1, both outcomes occurred more frequently in individuals with a longer duration of diabetes, with evidence of nephropathy, CVD, congestive heart failure, or a higher baseline HbA1C value. The primary outcome occurred less often in white individuals and more often in smokers, those with a prior amputation, and those with a higher systolic blood pressure. Mortality occurred more often in older participants, men, and individuals with a lower diastolic blood pressure. Both the primary ACCORD outcome and mortality occurred more frequently in individuals with lower baseline DSST scores.

Relationship between baseline DSST score and ACCORD primary outcome

Mean (standard deviation) DSST score for the entire cohort was 52.6 (15.9). There was an inverse relationship between the incidence of the ACCORD primary outcome and quartiles of DSST score, with the lowest incidence occurring in the highest quartile (Fig. 1). When treated as a continuous variable [Fig. 2(a)], a 1-point higher DSST score was associated with a lower incidence of the primary outcome [hazard ratio (HR), 0.98; 95% confidence interval (CI), 0.97 to 0.99; P = 0.001], adjusting for demographic and clinical trial factors (model 1). Adding additional cardiovascular risk factors and health systems factors (model 2) slightly attenuated the association (HR, 0.987; 95% CI, 0.977 to 0.998; P = 0.02), as did the addition of the self-reported need for assistance to follow the protocol as a time-dependent covariate (model 3) (HR, 0.987; 95% CI, 0.977 to 0.998; P = 0.02).

Figure 1.

The risk for ACCORD primary outcome according to DSST baseline quartiles is shown (DSS ≥ 63, 53 ≤ DSS < 63, 42 ≤ DSS < 53, DSS < 42). Q, quartile.

Figure 2.

The HR per 1-point higher DSST score is shown for the (a) ACCORD primary outcome and (b) all-cause death. HRs in each panel are adjusted for the following: model 1: age, sex, race, education, history of CVD, participation in the lipid vs blood pressure trial, allocation to fibrate vs placebo in the lipid trial, allocation to intensive vs standard blood pressure lowering in the blood pressure trial, and allocation to intensive vs standard glucose lowering; model 2: model 1 plus smoking, prior heart failure, prior amputation, site belonged to an integrated health plan, and baseline HbA1C, body mass index, serum creatinine, and albumin to creatinine ratio category; and model 3: model 2 plus whether the participant required help from others to follow medical instructions.

In sensitivity analysis in which death from cardiovascular cause was replaced by total mortality, the relationship between baseline DSST score and the primary outcome was similar (i.e., HR, 0.986, after adjusting for model 3 variables; 95% CI, 0.977 to 0.996; P = 0.004).

Similar conclusions were obtained when the standardized combined cognitive outcome measure (18) was substituted for the DSS score (HR, 0.93, after adjusting for model 3 variables; 95% CI, 0.87 to 0.99; P = 0.029) (Table 2).

Table 2.

The Relationship Between the Standardized Combined Cognitive Measure and the ACCORD Primary Outcome/All-Cause Mortality After Adjustment for Several Models

Outcome/Model	Standardized Cognition Composite		Winsorized Stroop First
	HR (95% CI)	P Value	HR (95% CI)	P Value
MACE
Model 1	0.91 (0.85 to 0.97)	0.004	0.89 (0.83 to 0.96)	0.001
Model 2	0.93 (0.87 to 0.99)	0.027	0.91 (0.85 to 0.98)	0.010
Model 3	0.93 (0.87 to 0.99)	0.029	0.91 (0.85 to 0.98)	0.011
All-cause mortality
Model 1	0.92 (0.83 to 0.98)	0.017	0.90 (0.82 to 0.98)	0.022
Model 2	0.95 (0.87 to 1.03)	0.209	0.93 (0.85 to 1.01)	0.096
Model 3	0.95 (0.87 to 1.04)	0.267	0.93 (0.85 to 1.02)	0.129

The base model (model 1) was adjusted for sex, age, race, education, history of CVD, participation in the lipid vs blood pressure trial, allocation to fibrate vs placebo in the lipid trial, allocation to intensive vs standard blood pressure in the blood pressure trial, and allocation to intensive vs standard glucose control. Subsequent models included smoking, prior heart failure, prior amputation, site belonging to an integrated health plan, and baseline HbA1C, body mass index, serum creatinine, and albumin/creatinine ratio category (model 2); these additional variables plus a time-dependent covariate representing self-reported need for assistance to follow the protocol were included (model 3).

Abbreviation: MACE, major adverse cardiac events.

Relationship between baseline DSST score and ACCORD primary outcome in subgroups

The inverse relationship between DSST score and the ACCORD primary outcomes was consistent across all subgroups except for those with prior heart failure [Fig. 3(a)] (P for interaction = 0.020), where power to detect relevant associations was limited due to a small number of participants with prior heart failure.

Figure 3.

The HR per 1-point higher DSST score is shown for (a) ACCORD primary outcome and (b) all-cause death. All analyses are adjusted for model 1 variables: age, sex, race, education, history of CVD, participation in the lipid vs blood pressure trial, allocation to fibrate vs placebo in the lipid trial, allocation to intensive vs standard blood pressure lowering in the blood pressure trial, and allocation to intensive vs standard glucose lowering. CHF, congestive heart failure.

Relationship between baseline DSST score and total mortality

A statistically significant relationship was demonstrated between a higher DSST score and a lower risk for mortality after adjustment for several covariates, including age, sex, race, education, history of CVD, participation in the lipid vs blood pressure trial, allocation to fibrate vs placebo in the lipid trial, allocation to intensive vs standard blood pressure in the blood pressure trial, and allocation to intensive vs standard glucose control (model 1 HR, 0.985; 95% CI, 0.971 to 0.999; P = 0.031). This relationship, however, was somewhat attenuated and no longer significant after further adjustment (Fig. 2). Similar conclusions were obtained when the standardized combined cognitive outcome measure (18) was substituted for the DSS score (HR, 0.95, after adjusting for model 3 variables; 95% CI, 0.87 to 1.04; P = 0.267) (Table 2). The relationship was consistent across all subgroups [Fig. 3(b)].

Discussion

This analysis of 2977 middle-aged and older individuals with T2D demonstrates an inverse relationship between cognitive function as measured by a simple cognitive instrument (DSST) that is sensitive to differences in the normal cognitive range and incident cardiovascular events. The fact that this relationship persisted after adjustment for demographic, cardiovascular risk factors as well as for self-care factors suggests measuring cognition with a simple test provides additional information that informs about an individual’s future health.

The importance of cognitive function as an informative assessment of a person’s health has been demonstrated in previous studies. An analysis of ∼31,000 individuals with preexisting CVD or at high risk for CVD followed for 56 months demonstrated an inverse relationship between the score on the Mini Mental State Examination score (a screening instrument for dementia) and incident cardiovascular events after adjustment for demographic and cardiovascular risk factors (9). In a study of 11,140 individuals with T2D who were followed for a median of 5 years, there was an inverse relationship between the Mini Mental State Examination score at baseline and incident CVD events (22). The Mini Mental State Examination, however, is a global test for cognitive impairment and a good screening test for dementia (23) (which is associated with a higher risk for death) but has less ability to discriminate between cognitively intact individuals. The current analysis broadens the finding of these studies to cognitive performance with an instrument sensitive to minor differences within the cognitively intact range and accounts for possible confounding factors.

We considered several explanations for the inverse relationship between cognitive function and incident CVD among middle-aged and older participants with diabetes. Lower DSST scores are associated with brain atrophy, an underlying substrate for poor cognitive performance that predicts dementia and death (24). Cognitive function may reflect preexisting cerebrovascular or other diabetes-related pathology in the brain or systemically. However, the observation that the relationship persisted in models that adjusted for prior CVD and diabetes complications suggests the measure is reflecting other aspects of health or resilience. Cognitive function is also associated with education and socioeconomic status, which are associated with subsequent risk for CVD. Differences in cognitive function may be associated with different self-management parameters (including adherence to prescribed medication and a higher propensity for severe hypoglycemia episodes) (25, 26). However, we found that the addition of “self-reported need for assistance to follow the protocol” only minimally attenuated the relationship, making it unlikely that factors related to self-management largely accounted for the findings. Other intervening factors, which we did not measure, include physical activity or diet, which may have a positive effect on cognition (27). In this diabetic sample, specific factors that may predispose to both decreased cognitive function and incident CVD include mitochondrial dysfunction (28, 29), the sortilin pathway (30), activation of the hypothalamic-pituitary-adrenal axis, and the relative lack of insulin and/or inflammation.

The study has several strengths, including its prospective design, the uniform administration of the DSST by trained individuals, and almost 100% adjudication of CVD outcomes. Like all observational analyses, however, we cannot eliminate the possibility of residual confounding. Furthermore, the fact that this is a sample of persons with longstanding T2D and a high risk for CVD means that the results may not be generalizable to persons who have had diabetes for a shorter duration, are better controlled, and have fewer comorbidities.

Our finding of a higher incidence of cardiovascular events with lower levels of cognitive function reflects an important link between cognition and health. On a clinical level, our findings suggest that differences in cognitive function, even in the normal range, are a sensitive marker of both systemic and central nervous system integrity and should be monitored over time. We have previously shown that lower cognitive function is associated with a higher likelihood of hypoglycemic events (31). These findings support recent guidelines that recommend routine screening for cognitive impairment in older people with diabetes (32–34). They also support the notion that a simple, easily administered neuropsychological instrument may provide additional information on a patient’s health status.