THE CROSS-SECTIONAL ASSOCIATION OF COGNITION WITH DIABETIC PERIPHERAL AND AUTONOMIC NEUROPATHY

Abstract

Background:

Studies examining whether measures of cognition are related to the presence of diabetic peripheral neuropathy (DPN) and/or cardiovascular autonomic neuropathy (CAN) are lacking, as are data regarding factors potentially explaining such associations.

Methods:

Participants were from the Glycemia Reduction Approaches in Diabetes Study (GRADE) that examined 5047 middle-aged people with type 2 diabetes of <10 years of known duration. Verbal learning and immediate and delayed recall (memory) were assessed with the Spanish English Verbal Learning Test; frontal executive function and processing speed with the Digit Symbol Substitution Test; and ability to concentrate and organize data with word and animal fluency tests. DPN was assessed with the Michigan Neuropathy Screening Instrument and CAN by indices of heart rate variability (standard deviation of normal beat to beat variation [SDNN] and root mean square of successive differences [RMSSD]).

Results:

DPN was significantly inversely related to measures of immediate recall and processing speed. The percent of cognitive variation explained by DPN was small. Tests of CAN had an inconsistent or absent association with measures of cognition. Higher waist circumference and urine albumin creatinine (UACR) levels were the strongest correlates in the relationship between DPN and cognitive impairment.

Conclusion:

DPN, but not CAN, is cross-sectionally associated with lower performance in measures of cognition in people with type 2 diabetes of <10 years of known duration. Greater waist circumference and UACR were important variables in this association. The mechanisms underlying the cross-sectional association of DPN with cognitive impairment are unknown.

1. INTRODUCTION

Several studies have demonstrated sub-clinical cognitive deficiencies in frontal lobe organizational function and visuospatial processing speed in people early in the course of type 2 diabetes (1–3). These deficiencies are present in the absence of known cerebrovascular disease and remain significant after adjustment for age and highest attained educational level (4).

Diabetic peripheral neuropathy (DPN) and cardiovascular autonomic neuropathy (CAN) are common complications of diabetes affecting up to 50% of diabetes patients, and are a major cause of disability, pain, poor quality of life, and mortality (5). Although injury of the peripheral and central nervous systems in type 2 diabetes share pathophysiological mechanisms – e.g., hyperglycemia, hyperlipidemia, hypertension, obesity, microvascular dysfunction, and low-grade inflammation (5) - studies linking sub-clinical cognitive changes with DPN and CAN have not been done. Addressing this knowledge gap could help identify patients at risk for cognitive impairments as well as to gain insight into the pathogenesis of early cognitive impairment.

In the present study we examined the cross-sectional association of measures of cognitive function with DPN and CAN. We further examined several candidate factors to determine if they were common to either DPN and/or CAN and lower scores on measures of cognition. We used the Glycemia Reduction Approaches in Diabetes (GRADE): A Comparative Effectiveness Study baseline data set to examine these points (6, 7). GRADE is a prospective NIDDK-sponsored study of 5047 adults with less than 10 years of known diabetes duration treated with metformin who after baseline were randomized to an add-on medication from one of four classes of glucose-lowering medications. All participants were well phenotyped and underwent baseline cognitive testing.

2. METHODS

Eligibility criteria for the GRADE study include less than 10 years of known type 2 diabetes duration, diagnosis at age ≥30 years in non-American Indian (AI) / Alaska Native (AN) patients or age ≥20 for AI/AN, baseline hemoglobin A1c (A1c) between 6.8 and 8.5%, and estimated glomerular filtration rate (eGFR) of ≥30 ml/min/1.73m² at enrollment. Exclusion criteria include any major cardiovascular event in the year prior to recruitment and/or a history of New York Heart Association heart failure stages 3 or 4. The primary outcome in GRADE is the time to primary metabolic failure, defined as an A1C ≥7% (53 mmol/mol), subsequently confirmed, over an anticipated mean observation period of 4.8 (range 4–7) years.

2.1. Measures of DPN and CAN:

DPN was assessed by the Michigan Neuropathy Screening Instrument (MNSI) (8). It includes two separate assessments: a 15-item self-administered questionnaire and a lower extremity examination consisting of inspection of the feet, and assessment of vibratory sensations and ankle reflexes. DPN was defined either as a continuous variable combining a weighted average of the questionnaire and examination (mean value 2.02; median 1.81 [IQ, 1.03, 2.76]; minimum −0.57, maximum 7.24), or as a dichotomous value, called definite DPN: score >2.54 on the weighted score (8).

CAN was assessed by two time domain indices of heart rate variability derived from a 10 second standard 12-lead ECG recording: the standard deviation of normally conducted R-R intervals (SDNN) and the root mean square of successive differences between normal-to-normal R-R intervals (RMSSD). SDNN and RMSSD values were examined as continuous variables (9) and as dichotomous values using cut-points defining abnormally low variability: <8.2 ms (SDNN) and <8.0 ms (RMSSD) (10). Definite CAN was defined as abnormal values of both SDNN and RMSSD.

2.2. Cognitive assessments:

The cognitive battery measured memory (verbal learning) and frontal-executive abilities. All tests were administered in English or Spanish. Details are presented in the Supplemental Table 1. The measure of memory was the Spanish English Verbal Learning Test (SEVLT) (11). The SEVLT consists of recalling a list of 15 words in 3 trials of immediate recall and 1 trial after a distractor list. For the SEVLT, we examined two outcomes: the sum of the number of words recalled in the first 3 trials (immediate recall; SEVLT1–3), and the score of the 4th trial after the distractor list (delayed recall; SEVLT4). The tests of frontal-executive abilities were the total score in the Digit Symbol Substitution Test (DSST) (12) and number of words generated in the animal and letter (13) fluency tests. The DSST is a test in which participants try to match numbers to symbols in 90 seconds. The total number of correct answers is reported. The animal fluency test asks participants to name as many animals as they can for one minute. The letter fluency test asks participants for as many words as possible with the letter F in English (P in Spanish) in one minute. The total number of correct words is reported for the fluency tests.

2.3. Covariates:

We included as covariates factors reported to be associated with cognitive performance and factors reported to be associated with complications in type 2 diabetes. Covariates associated with cognitive performance included sex, age, smoking status, alcohol consumption, Hispanic ethnicity, racial group, and prevalent cardiovascular disease (myocardial infarction, stroke). Factors associated with type 2 diabetes complications included waist circumference, highly sensitive-CRP (hsCRP), and glycemia. Other measured covariates, common to both conditions, were weight, height, waist circumference, and seated blood pressure. Body mass index (kg/m²; BMI) was calculated. A quality of life questionnaire (Short Form 36) was administered (14). A positive response to the question “are you depressed?” or use of anti-depression medications were considered diagnostic of a history of depression.

2.4. Laboratory:

Fasting glucose (FG) and HbA1c levels were measured at baseline. As part of a substudy 1543 (33.8%) participants had hsCRP measured at baseline. Participants provided a random urine sample for measurement of albumin and creatinine, and blood for creatinine levels. Values ≥30 mg albumin/gram creatinine were classified as albuminuria. An estimated glomerular filtration rate (eGFR) (ml/minute/1.73m²) was calculated using the CKD-EPI equation (15). hsCRP was measured in serum using a latex-particle enhanced immunoturbidimetric assay on the Roche cobas c501 chemistry analyzer (Roche Diagnostics, Indianapolis, IN 46250).

2.5. Statistical Analyses:

Baseline characteristics of the cohort are presented by the presence or absence of definite DPN and definite CAN, the main outcomes in this analysis. The associations of definite CAN and definite DPN with measures of cognition were examined using sequential linear regression modeling; each test of cognition, separately, was the outcome variable, and definite CAN alone and definite DPN alone, respectively, were the main effects variables. Similarly, sequentially adjusted linear regression models were constructed to examine the association of DPN as a continuous variable and of SDNN and RMSSD, respectively, with the measures of cognition. SDNN and RMSSD were log transformed to address skewness and to improve normality for the mediation analyses. The associations of continuous DPN and RMSSD with changes in tests of cognition were examined together to determine if the presence of one impacted the association of the other with measures of cognition. The same was done for continuous DPN and SDNN. Adjusted R² values were examined to quantify the proportion of variability of the test of cognition that was explained by definite DPN and CAN as dichotomous variables and by DPN, SDNN, and RMSSD as continuous variables. Values were multiplied by 100 to convert to percentages.

In addition to fitting the individual regression models, we performed a pooled analysis using the Wei-Lachin test (16) to assess the overall effect of DPN on cognition and CAN on cognition. In this analysis, the regression effects from the fully adjusted models are standardized and combined into an overall 1 degree freedom test of the null hypothesis of no effect in any regression model versus the alternative hypothesis of effects in at least some of the regression models, all of them in the same direction.

In exploratory analyses, we examined whether hyperglycemia (fasting glucose [FG] and HbA1c levels), inflammation (hsCRP levels), and urine albumin creatinine ratio [ACR]) impacted the relationships involving DPN (as continuous and dichotomous variables) with measures of cognition. We also examined LDL cholesterol, waist circumference, systolic blood pressure, and amount of alcohol consumed per week. The effect of hsCRP on cognition was measured in the 1543 participants with this measurement at baseline.

Analyses were done using the R Program, 2019 (17).

3. RESULTS

Of the 5047 participants in GRADE, 4559 (90.3%) with all data identified a priori as necessary for hypothesis testing were included in this study (Figure 1).

Figure 1:

Analytic cohort from the GRADE study.

Baseline characteristics of the cohort, categorized by the presence or absence of definite DPN and definite CAN, are shown in Table 1. There were 1255 participants (27.5% of the cohort) with definite DPN and 447 (9.8% of the cohort) with definite CAN. Participants with definite DPN and definite CAN were older, were more likely male, had higher waist circumferences, increased systolic blood pressure, lower eGFR and increased urine albumin creatinine levels, and more depression than their counterparts without either of these diagnoses. Participants with definite DPN had lower education, increased BMI, and more prevalent CVD than participants without definite DPN. Participants with definite CAN were more likely to be White compared to participants without definite CAN. Participants with definite CAN had more definite DPN and participants with definite DPN had lower SDNN measures compared to their respective counterparts. Finally, participants with definite DPN had lower levels of measures of cognition than participants without definite DPN. Conversely, participants with definite CAN had approximately similar levels of measures of cognition compared to those without definite CAN.

Table 1:

Baseline characteristics of the 4559 GRADE cognition participants by the presence or absence of definite diabetic peripheral neuropathy (DPN) and definite cardiovascular autonomic neuropathy (CAN).

		Definite DPN				Definite CAN
		Yes N=1255	No N=3304	P Value		Yes N=447	No N=4112	P Value
DEMOGRAPHIC FACTORS
Age (mean, SD)		58.6 ± 9.3	56.2 ± 10.0	<0.001		60.1 ± 8.8	56.5 ± 9.9	<0.001
Male sex (%)		65.7	61.6	0.01		68.8	62.1	0.006
Race (%)				<0.001				<0.001
White		68.3	65.0			77.4	64.6
Black		19.6	19.6			11.6	20.5
Asian		1.8	4.2			3.1	3.6
Other		10.3	11.2			7.8	11.4
Hispanic Ethnicity (%)		17.2	19.6	0.18		14.1	19.4	0.003
Current smoker (%)		15.2	12.7	0.03		16.8	13.0	0.03
Education (%)				<0.001				0.96
High school		35.9	25.1			27.5	28.1
College		50.6	56.8			55.3	55.1
Graduate school		13.5	18.2			17.2	16.8
DIABETES
Duration (years; mean, SD)		4.2 ± 2.7	4.0 ± 2.8	0.03		4.3 ± 2.8	4.0 ± 2.7	0.06
HbA1c screening >8% (%)		34.4	35.2	0.66		34.9	38.8	0.27
HbA1c baseline (%)				0.17				0.90
<7.0		15.6	13.7			13.9	14.3
7.0–7.9		63.1	65.7			66.0	64.9
8.0–8.5		21.3	20.6			20.1	20.9
ANTHROPOMETRICS
BMI (kg/m2)		34.9 ± 6.8	34.0± 6.8	<0.001		34.4 ± 7.0	34.2 ± 6.8	0.72
Waist (cm)		114.6 ±16.0	111.1 ± 15.4	<0.001		114.1± 16.1	111.9 ± 15.6	0.006
SBP ≥140 or on HTN treatment (%)		77.3	71.4	<0.001		79.2	72.3	0.002
DBP (mmHg)		76.9 ± 9.9	77.6 ± 9.8	0.045		77.9 ± 10.5	77.3 ± 9.8	0.29
LIPIDS (mg/dl)
Total cholesterol		164.0 ± 37.5	164.3±38.1	0.90		166.1 ± 39.9	163.9 ± 37.7	0.27
HDL		42.7 ± 10.6	43.7 ± 10.5	0.01		42.6 ± 10.1	43.5 ± 10.6	0.08
LDL		90.9 ± 31.2	90.6 ± 32.1	0.83		88.9 ± 29.8	90.9 ± 32.0	0.18
Triglycerides		160.4± 147.6	152.9± 113.1	0.11		178.2 ± 153.6	152.5 ± 119.6	<0.001
RENAL TESTS
eGFR (mean, SD)		93.3 ± 17.0	95.8 ± 16.6	<0.001		91.1 ± 17.5	95.6 ± 16.6	<0.001
eGFR <60 ml/minute/1.73m2 (%)		3.5	2.1	0.01		4.9	2.2	<0.001
ACR (mean, SD)		40.2 ± 208.5	27.9 ± 108.2	0.045		43.3 ± 185.7	30.0 ± 137.6	0.14
ACR ≥ 30 mg/g creat (%)		16.9	14.2	0.02		18.8	14.5	0.02
FASTING GLUCOSE & INSULIN
Fasting glucose (mg/dl)		151.6 ± 31.8	151.9 ±30.9	0.80		155.1 ± 33.7	151.5 ± 30.8	0.03
Fasting insulin (mU/L)		22.8 ± 16.3	20.9 ± 14.1	0.002		21.3 ± 13.4	21.5 ± 14.9	0.80
PREVALENT CVD (%)
Myocardial infarction		6.9	4.0	<0.001		6.7	4.6	0.06
Stroke		2.8	1.4	0.003		2.2	1.8	0.58
GENERAL FACTORS (%)
Alcohol				0.18				0.002
Never		35.1	32.0			33.8	32.7
Sometimes		60.9	64.5			59.5	64.0
Daily		4.0	3.5			6.7	3.3
Depression		16.7	12.0	<0.001		18.8	12.7	<0.001
Self-Rated Health				<0.001				0.21
Excellent, very good, good		71.7	85.4			77.9	82.0
Fair, poor		28.3	14.6			22.1	18.0
NEUROPATHY
Mean MNSI DPN score		3.7 ± 0.8	1.4 ± 0.6	<0.001		2.2 ± 1.3	2.0 ± 1.2	0.002
Definite DPN (%)				21.3	15.5	0.002
Definite CAN (%)	11.2	9.3	0.052
SDNN (ms)	18.7 ± 16.8	20.0 ± 13.8	0.01	5.8 ± 1.5	21.2 ± 14.7	<0.001
RMSSD (ms)	21.1 ± 24.2	21.8 ± 18.4	0.31	5.6 ± 1.4	23.4 ± 20.5	<0.001
COGNITIVE TEST RESULTS (mean, SD)
SEVLT 1–3	24.8 ± 6.0	25.6 ± 5.8	<0.001	24.9 ± 5.9	25.4 ± 5.8	0.08
SEVLT 4	9.2 ± 2.7	9.5 ± 2.6	0.002	9.3 ± 2.6	9.4 ± 2.7	0.28
DSST	44.0 ± 13.2	47.2 ± 13.9	<0.001	44.5 ± 12.9	46.5 ± 13.8	0.003
Word Fluency Letter	11.9 ± 4.3	12.6 ± 4.4	<0.001	12.1 ± 4.4	12.4 ± 4.4	0.18
Word Fluency Animal	18.9 ± 5.3	19.4 ± 5.4	0.002	19.5 ± 5.1	19.2 ± 5.4	0.27

Figure 2 shows percent changes in measures of cognition from regression models of DPN as a continuous and as a dichotomous variable. In unadjusted analyses, all cognition scores were significantly lower with higher DPN values (continuous and dichotomous). As a continuous variable, adjustment for covariates attenuated the associations between DPN with SEVLT 4 and with word and animal fluency tests. The DSST and the SELVT1–3 tests remained significantly lower. As a dichotomous variable, DSST remained significantly lower with adjustment for covariates.

Figure 2:

Percent change of tests of cognition with peripheral diabetic neuropathy as a continuous and as a dichotomous variable (Definite DPN). Open circles are not statistically significant.

The tests of cognition were the immediate (trials 1–3) and delayed recall (trial 4) in the Spanish English Verbal Learning Test (SEVLT), the digit symbol substitution test (DSST), and the letter and animal fluency tests. Model 1: no adjustments; Model 2: age, race, and female sex; Model 3: in addition, waist circumference, diabetes duration, current smoking, alcohol status, history of stroke, depression, hypertension, highest education level, and hyperlipidemia.

The proportion of cognitive test variation explained by DPN alone (Model 1) was small, varying from 0.1% (Animal fluency) to 1.2% (DSST) (the adjusted R² values) (Supplemental Table 2). The Wei-Lachin test, applied to Model 3, revealed a significant trend for continuous DPN (p=0.02) but not for dichotomous (definite) DPN.

Figure 3 shows percent changes in tests of cognition from regression models of SDNN, RMSSD and definite CAN. For SDNN and RMSSD, values are per unit increase of the log transformed variable (original variable in milliseconds). SDNN was significantly associated with the SEVLT 1–3 and DSST on unadjusted analyses. With adjustment, these associations were attenuated. RMSSD had a significant association with the Word Fluency Animals on unadjusted analyses but not with other tests. Adjustment for covariates resulted in statistically significant associations with SEVLT4, DSST and Word Fluency Animals tests. When definite CAN was examined, there were borderline and statistically significant associations with SEVLT1–3 and DSST on unadjusted analyses, respectively. Adjustment attenuated these findings. The Wei-Lachin test applied to Model 3 for SDNN, RMSSD and definite CAN showed significant trend for RMSSD (p=0.001) but not for SDNN or definite CAN.

Figure 3:

Percent change of tests of cognition with SDNN, RMSSD and as a dichotomous variable (Definite CAN). Open circles are not statistically significant.

The tests of cognition were the immediate (trials 1–3) and delayed recall (trial 4) in the Spanish English Verbal Learning Test (SEVLT), the digit symbol substitution test (DSST), and the letter and animal fluency tests. Model 1: no adjustments; Model 2: age, race, and female sex; Model 3: in addition, waist circumference, diabetes duration, current smoking, alcohol status, history of stroke, depression, hypertension, highest education level, and hyperlipidemia.

The proportion of variation of tests of cognition explained by SDNN as a continuous variable was stronger than that explained by RMSSD (Supplemental Table 2). However, most of the tests were not statistically significant and the amount of cognitive test variation explained by SDNN or RMSSD was negligible.

The associations of DPN as a continuous variable and of RMSSD and of SDNN with tests of cognition when adjusted for one another are shown in Supplemental Tables 3 A&B. β coefficients were similar to coefficients when each of the tests was examined separately.

Exploratory Analysis:

Candidate factors that could alter the relationships involving DPN with tests of cognitive performance were explored (Table 2). In fully adjusted models hsCRP, HbA1c, systolic blood pressure, and alcohol intake were not significantly associated with DPN, i.e. neither continuous nor definite DPN. Alcohol use was associated with improved cognitive function. LDL-cholesterol was significantly associated with definite DPN but not continuous DPN. The strongest associations were for waist circumference (associated with both continuous and definite DPN) and urine ACR (associated with continuous DPN but not definite DPN).

Table 2:

Mediation analyses assessing which factors could impact relationships between diabetic peripheral neuropathy (DPN) and measures of cognition. Data shown are p values. A plus (+) sign signifies a positive association; a negative (−) sign is an inverse association.

	Diabetic Peripheral Neuropathy			Measures of Cognition
				SEVLT 1–3	SEVLT 4	DSST	Letter Fluency	Animal Fluency
	Continuous	Dichotomous
CRP*	0.71	0.70		0.04 −	0.25	0.59	0.43	0.04 −
FG	0.45	0.99		0.15	0.12	0.005 +	0.31	<0.001 +
HbA1c	0.91	0.91		0.07	0.19	0.15	0.62	0.20
UACR	<0.006 +	0.10		0.10	0.30	0.004 −	0.30	0.01 −
Waist circumference (cm)	<0.001 +	<0.001 +		<0.001 +	<0.001 +	<0.001 +	<0.001 +	<0.001 +
SBP (mmHg)	0.83	0.84		0.99	0.90	0.62	0.35	0.33
LDL cholesterol (mg/dl)	0.07	0.03 +		0.40	0.89	0.01 −	0.12	0.80
Alcohol (drinks/week)	0.23	0.63		<0.001 +	<0.001 +	<0.001 +	<0.001 +	<0.001 +

^*CRP – performed in 1543 participants only; FG – fasting glucose; UACR – urine albumin creatinine ratio

The tests of cognition were the immediate (trials 1–3) and delayed recall (trial 4) in the Spanish English Verbal Learning Test (SEVLT), the digit symbol substitution test (DSST), and the letter and animal fluency tests. All p values are adjusted for age, race, sex, waist circumference, diabetes duration, current smoking, alcohol use, history of stroke, depression, hypertension, hyperlipidemia, and highest level of attained education. Statistically significant values are highlighted.

4. DISCUSSION

There are two salient outcomes in this cross-sectional study of mostly middle-aged diabetic individuals with type 2 diabetes duration of <10 years. First, in unadjusted analyses, participants with higher DPN scores, when measured as a continuous variable, had lower scores on several measures of cognition, suggesting more cognitive dysfunction. With adjustment for confounding factors, measures of memory and learning (SEVLT1–3) and of visuospatial processing speed (DSST) remained significantly associated with DPN. As a dichotomous variable, definite DPN had a significant association with DSST and a borderline significant association with letter fluency. The proportion of cognition variation explained by DPN was small.

The question arises how to interpret these findings. It may be argued that the association of DPN with cognition is marginal and therefore is of little to no clinical significance. In addition, the cognitive deficits were subtle and of uncertain clinical significance, possibly owing to the short duration of known diabetes, the relatively good glycemic control, and the relatively young age of the cohort. The large size of the GRADE cohort likely contributed to our ability to detect these small associations.

Previous studies of the relationship of DPN with cognition have been done mostly in small cohorts (18–24). Results have been conflicting, which is not surprising given the wide array of ages of the examined participants and durations of known diabetes, as well as the heterogeneity of cognitive measures that were used. Several studies reported no significant associations (19–21), while others found significant inverse associations (22–24). Our finding of an association of DPN with measures of frontal lobe executive function (DSST) is consistent with the work of Rucker et al (23) who used the MNSI as the measure of DPN (with a scoring method similar to our own) together with nerve conduction studies in a young cohort. Additionally, a recent study using measures of cognition assessed with functional MRI showed changes in cognitive areas of the brain with peripheral nerve stimulation in people with DPN (as confirmed by EMG). This study suggests a connection between peripheral nerve function and areas of the brain associated with cognitive function (25). However, the mechanisms linking DPN with cognitive dysfunction in diabetes are not known.

The second major finding of this study was a lack of association between the measures of cognition with measures of CAN. There was also no impact by RMSSD or SDNN on the association of DPN with tests of cognition. These findings are similar to the Whitehall II Study (26). It, too, failed to demonstrate an association of CAN, assessed by heart rate variability, with prevalent or incident cognitive dysfunction in adults, mean age 55 years. It did not report separately on people with diabetes. In contrast, in a small study of 20 people with type 2 diabetes and autonomic neuropathy (mean age 60 years) there was poorer performance on tests of visual memory compared to participants without CAN (27).

We explored potential factors that could explain the relation of DPN to measures of cognition. Baseline HbA1c, systolic blood pressure, hsCRP levels and alcohol use were not associated with DPN. Hence, these factors are unlikely to be mediators explaining how DPN could be associated with cognition. On the other hand, greater waist circumference had an association with most measures of peripheral and autonomic neuropathy and with measures of cognition. Obesity is associated with peripheral and autonomic neuropathy (28, 29). Larger waist circumference is associated with increased production of markers of inflammation and metabolic factors not measured here that could impact cognition (e.g., non-esterified fatty acids [30]). Obesity is also reported to be associated with reduced measures of executive function (31). Thus, factors associated with elevated adiposity may explain our findings.

We also found urine ACR levels were strongly related to DPN (defined as a continuous variable), suggesting potentially common pathways of associations between DPN with cognitive impairment. A possible explanation is that urine ACR is a manifestation of a systemic disorder of the microvasculature and/or of endothelial dysfunction (32) which impairs blood flow or blood flow regulation to peripheral and autonomic nerves, and to the brain (33, 34).

We note that low to moderate alcohol use (the majority of GRADE participants drank “sometimes”) was associated with improved cognitive function. This has been previously reported in middle aged and older adults (35, 36). We also found no association of low alcohol intake with DPN. Only excessive and prolonged alcohol intake is associated with peripheral neuropathy in the setting of diabetes (37). These two findings support the internal validity of our study results regarding the association of DPN with cognition.

Strengths of this study include a large, well characterized cohort and a broad spectrum of participants which was racially and ethnically diverse. The tests of DPN, CAN, and cognition are widely used. We classified DPN and CAN as continuous and categorical variables so as to examine associations with measures of cognition in as broad a manner as possible. Our finding that DPN was associated with a low level of attained education – a surrogate marker of socioeconomic status - is consistent with prior studies (31, 38) and supports the validity of our results. We also present data on the inter-relationship between DPN and CAN. Limitations of this study should be acknowledged. The measures of CAN used here are based only on time domain variables. Other domains are available (39) which could offer a broader characterization of the relationship of CAN with cognition. Our association analyses used representative markers of candidate pathway, none of which was a comprehensive marker of its pathway. For example, one value of blood pressure or fasting glucose, may not reflect the overall level or severity of either factor. Also, vitamin B12 levels and TSH levels were not available at baseline. Low B12 levels are associated with metformin use (40). Low B12 levels and abnormal TSH levels can impact cognition and peripheral nerve function (41–43). Last, this is a cross-sectional study from which causal associations cannot be derived.

In conclusion, we found that sub-clinical cognitive impairment is associated with DPN in middle-aged adults with known diabetes duration <10 years. This suggests empirically that sub-clinical cognitive dysfunction may be an early complication of diabetes that can arise with DPN and may share common pathologic pathways as of yet unidentified.

HIGHLIGHTS.

• The association of diabetic peripheral neuropathy (DPN) and autonomic neuropathy (AN) with cognitive impairment was examined cross-sectionally in 5047 middle-aged people with type 2 diabetes of <10 years duration.

• DPN was significantly and inversely associated with measures of immediate recall and processing speed. Higher urine albumin creatinine (UACR) levels and waist circumference were the strongest correlates of this association.

• AN had an inconsistent or absent association with measures of cognition.