Shared Neural Substrates of Cognitive Function and Postural Control in Older Adults

Abstract

Introduction:

Poor cognitive function and postural control co-occur in older adults. It is unclear whether they share neural substrates.

Methods:

Postural sway error during a novel visual tracking (VT) condition and grey matter volume (GMV) were compared between participants with normal cognition (NC), mild cognitive impairment (MCI), or dementia (n=179, mean age 82, 56% female, 56% white). Associations between VT error, cognitive function, and GMV were examined.

Results:

Greater VT error was associated with having dementia compared to NC or MCI (odds ratio [95%CI] =2.15[1.38, 3.36] and 1.58 [1.05, 2.38]). Regions with lower GMV related to greater VT error and worse cognition were: bilateral hippocampi, parahippocampi, entorhinal, and parietal cortices (all p≤0.05). GMV of bilateral hippocampi and left parahippocampus explained >20% of VT error between dementia and NC.

Discussion:

Postural control during visuospatial tasks and dementia may share neural substrates, specifically memory-related regions.

1. Background

Dementia is a widespread neurocognitive disorder, affecting 46.8 million people worldwide, with prevalence predicted to increase exponentially to 131.5 million in the next 30 years.¹ This large public health burden highlights the need for early identification of those at risk for dementia. Early detection of those most at risk is critical to prevent and/or delay the onset of dementia.^2,3 However, current screening approaches are either only useful in those with existing impairment (i.e. neuropsychological assessments), or are too invasive and costly to be used at a population level (i.e. imaging or biomarker measurement). Indicators of early stages of this condition, specifically indicators to distinguish normal cognition (NC) or mild cognitive impairment (MCI) from dementia could be especially important.

There has been a growing interest in examining postural control as a marker of early cognitive impairment.⁴ Clinical measures of poor postural control, such as measures of balance are known to correlate with dementia; specifically, home-dwelling older adults with mild or moderate Alzheimer’s disease (AD), appear to exhibit poorer balance compared to those with subjective cognitive impairment or MCI.⁵ Computerized metrics of postural control, such as maintenance of the center of pressure within the limits of stability computed via forceplate methodologies, provide greater understanding about the control of standing balance compared with observational analysis or timing how long one can maintain stance. It is important to recognize that assessment of standing balance is part of a more comprehensive assessment of balance across multiple domains.⁶ Emerging evidence indicate such computerized measures strongly relate to cognitive impairment. For example, postural sway was greater in older adults with MCI or dementia, compared to those with normal cognition (NC).^7–9 A case-control study also showed posturography data alone discriminated dementia from NC 72–86% of the times.¹⁰ However, the neural correlates of these associations have not been well studied.

Postural control depends on the successful integration of sensory input from visual, vestibular, proprioceptive, and cognitive pathways in the brain.¹¹ Worse postural control has recently been associated with smaller gray matter volume (GMV) of subcortical regions; specifically, with smaller hippocampus in older adults with normal cognition¹², and with the nucleus accumbens in patients with AD.⁹ Atrophy of the temporal horn area correlated with larger postural sway independent of cognitive function.¹³ These areas are also associated with dementia.^13–15 Although it appears plausible that postural control and cognitive function would share neural substrates, such relations have not been examined in detail; studies focused on only few regions, used measures of postural control during conventional tasks, and examined populations with a limited range of cognitive function.

The objective of this study was to identify the shared neural substrates of postural control and cognitive status in older, community-dwelling adults with a range of cognitive function including NC, MCI, and dementia; we used computerized metrics of postural sway while performing tasks requiring varying levels of visuospatial attention. We hypothesized postural sway error would be smaller among older adults with NC or MCI, as compared to those with dementia, and that associations would be stronger for tasks requiring higher levels of visuospatial attention. Additionally, we hypothesized worse postural control (e.g. higher sway error) would be related to lower GMV, especially in regions important for cognitive function and memory, and that these regions would represent a shared neural substrate for postural control and cognitive impairment.

2. Methods

2.1. Study Population

A prospective cohort of community-dwelling white and black older adults from Memphis, TN or Pittsburgh, PA were enrolled in the Health, Aging and Body Composition (Health ABC) study, in 1997–98¹⁶. The original cohort has been previously described in detail.¹⁶ Briefly, participants were relatively healthy 70–79 year old adults (51.6% females and 58.4% white) with no self-reported difficulty in walking a quarter mile, climbing ten steps without resting or performing activities of daily living. This study was approved by the institutional review boards at the University of Pittsburgh, the University of Tennessee Memphis, and the Coordinating Center at the University of California San Francisco. All participants signed a written informed consent.

In 2006–07 (year 10 of Health ABC cohort), 819 of the initial 1,527 participants enrolled at the Pittsburgh study site were alive and asked to participate in the Healthy Brain Project (HBP), a neuroimaging sub-study of cognition and mobility. Of the 819, 315 received a brain MRI at 3Tesla, 10 received a brain MRI at 1.5 Tesla, and 17 were not scanned although they were eligible (either because of intervening illnesses, death before the scan, or because they changed their mind). Of the others, 99 were ineligible for a brain MRI, 169 were ineligible for the study because they were walking with a cane and/or did not have mobility performance measures (original study designed to investigate mobility, thus this served as exclusion criteria), and 64 were ineligible for the study because they had been hospitalized for major clinic events in the previous 3 months (fracture, psychiatric problem) or for other reasons (missing data). Eligibility for the study could not be inspected in the remaining 145 participants because they either were not interested or refused. Of the 315 who completed a 3 Tesla brain MRI in 2006–07, 193 participants also completed the instrumented balance examination in 2009–10. A subset of 179 participants also had complete cognitive adjudication data in 2010–11; this comprised our analytic sample. Thus, from the initial sample of 819, 179 participants were included and 640 not included; included participants were about one year younger (82.4 and 83.7 years, p<0.001), and had a slightly higher Modified Mini-Mental State (3MS) Examination score (94 and 92 points, p=0.003) compared with those not included, but were similar for gender, race, body mass index or education.

2.2. Postural Sway

Participants performed a set of four conventional balance tasks and one visual tracking (VT) task, as previously described (Supplemental Figure 1).¹⁷ The four conventional tasks, based on the original Clinical Test of Sensory Interaction on Balance^18,19 included: standing quietly with eyes open and eyes closed, either on a stable surface or on a compliant surface (a 7.5 cm -thick medium-density foam pad). Participants wore a safety harness and stood upright with feet together on a force platform (Bertec Corp), for up to 30-seconds, or until balance was lost. Ground reaction forces were recorded from the force platform using Labview software (National Instruments, Inc.), and the root mean square (RMS) of the center of pressure (COP) in the antero-posterior and medio-lateral directions was computed for trials lasting at least 15 seconds, up to the maximum of 30 seconds. The RMS of the COP is a measure of variability of displacements of the COP (RMS = sqrt(sum(X_i – X_mean)²/number of samples), where X_i is an individual sample, and X_mean is the mean of all samples, equivalent to the standard deviation). Each participant was given 30 seconds of seated rest between each condition. Results are reported for medio-lateral sway for all tasks. The VT task consisted of controlling the movement of one’s medio-lateral COP using visual feedback.¹⁷ Subjects stood on the force platform with feet 16-cm apart, facing a computer monitor approximately 1.5 m away. The computer monitor displayed an open circle moving across the screen along a horizontal line. An ”X” representing the COP was also displayed on the horizontal line. Participants were instructed to keep the ”X” within the open circle to the best of their ability; the target’s movement was scaled to a range of 16-cm. Participants practiced moving the ”X” on the screen by shifting their weight from one foot to the other while keeping both feet in contact with the platform; they practiced until they were comfortable with the task. Participants performed four trials of differing frequencies (0.125, 0.25, 0.50, and 0.75 Hz); each trial was 60 seconds. The 0.25 Hz frequency was chosen as the primary VT variable of interest for this analysis, as it is the closest to the frequency of quiet stance in older adults. The RMS of the error (VT error) between the position of the target and the medio-lateral COP on the horizontal axis was computed and used as an indicator of performance. Smaller values indicate smaller errors (e.g. smaller distances between the “X” and the circle), thus better performance. In this cohort, better performance on this task was associated with higher performance on a cognitive task of visuospatial attention.¹⁷

2.3. Adjudication of Cognitive Function

An extensive neuropsychological assessment was carried out in 2010–11, concurrently with brain MRI and a neurological exam. The neuropsychological assessment, designed following extensive consultation with neurologists and dementia experts at the Pittsburgh Alzheimer’s Disease Research Center, included (a) Memory: California Verbal Learning Test (CVLT), and Rey-Osterreith figure; (b) Language: Boston Naming test and Verbal fluency (Animals); (c) Executive Function: Verbal fluency (initial letter), Trail Making test part B, Digit Span backwards and Stroop test; (d) Attention: Digit Span forwards, and Trail Making test part a. The neuropsychological battery was sensitive to cognitive impairment, and detailed normative data have previously been obtained through the Cardiovascular Health Study Dementia study²⁰. Cognitive status was clinically adjudicated as dementia, mild cognitive impairment (MCI),or normal cognition (NC), based on medical and cognitive data collected from study entry to time of the brain MRI in 2010–11, and the neuropsychological assessment and the brain MRI in 2010–11. Brain MRIs were visually inspected for presence of medial temporal lobe atrophy, sulcal and/or ventricular enlargement, and pathology of presumed vascular origin, including appearance of white matter hyperintensities, infarcts, and microbleeds. Adjudicated outcomes for the 179 participants included: cognitively normal (n=76), MCI (n=63), dementia (n=40). Those with MCI were further characterized as amnestic multi-domain (45%), non-amnestic single domain (28%), amnestic single domain (18%) and non-amnestic multi-domain (10%).

2.4. Neuroimaging

A 3T Siemens Tim Trio MR scanner with a Siemens 12-channel head coil (Siemens, Munich, Germany) was used for obtaining MRI scans in 2006–07. Image acquisition and analyses have been previously described.²¹ Magnetization-prepared rapid gradient echo (MPRAGE) images were acquired to obtain GMV and white matter hyperintensities (WMH) volume. Atrophy of total brain was computed as the ratio of GMV of total brain by intracranial volume. Gray matter volume of regions of interest (ROI) were defined a priori, and included the medial temporal lobe (hippocampus, parahippocampus, entorhinal cortex), basal ganglia (caudate, putamen, pallidum, thalamus), posterior parietal cortex (superior and inferior parietal lobule), cingulate cortex (anterior, middle, and posterior portion), and cerebellum, based on previous literature demonstrating regional associations with mobility and cognitive function.^22–24 The neuroanatomical boundaries of these regions have been previously published.²⁵ WMH volume was for total brain quantified from T2-weighted fluid-attenuated inversion recovery images with a semi-automated method, as previously described.²⁶

2.5. Covariates

Self-reported demographic data on age, sex, education, body mass index (BMI) and race were recorded at our analytic baseline (2006–07). Prevalent and incident disease algorithms based on both self-report and on physician diagnoses, recorded medications, and laboratory data were used to create time-dependent comorbidity variables indicating the development of diabetes mellitus, coronary heart disease, hypertension, and stroke/TIA during the entire study period. Neuromuscular characteristics included knee extensor strength, joint pain, lower extremity vibration sense as a measure of proprioception²⁷ and ankle/brachial index as a measure of peripheral vascular integrity²⁸; these were measured at our analytic baseline (2006–07).

2.6. Statistical Analysis

The distributions of all continuous variables approximated normality, so parametric tests were used. Demographics, comorbidities, and neuromuscular characteristics were compared across the three cognitive status groups (NC, MCI, and dementia), using analysis of variance (ANOVA) models for continuous variables and chi-square tests (or Fisher’s exact chi-square tests for cell sample sizes <10) for categorical variables. Variables significantly univariately associated with cognitive status entered all subsequent models. To assess the cognitive and postural characteristics of the VT task, Spearman’s correlation coefficients were computed between VT error with postural sway during conventional tasks and with performance on individual cognitive tests.

Mean and SD of postural sway measures (4 conventional tasks, and one novel visual tracking task) were computed for NC, MCI and dementia, and statistically compared with ANCOVAs, adjusting for variables significantly associated with cognitive function. These models were also adjusted for age and time between postural sway and cognitive adjudication visits. Given the potential for collinearity among some of these covariates, models were first adjusted for one covariate at a time and the variation inflation factor (VIF) was inspected. VIF remained <4 when one covariate was entered into each model, and in the fully adjusted model, indicating no major collinearity problems. To further assess the strengths of the association of postural sway during the VT task with cognitive status, receiver operating characteristic (ROC) curves were computed for VT error and each cognitive test; the area under the curve (AUC), standard error and asymptotic p value are reported for each model.

To identify the potentially shared neural substrates of VT error and cognitive function, we first computed the correlations of GMV of a-priori ROIs with postural sway (Pearson’s) and with cognitive function (Spearman’s) adjusted for total brain atrophy. Given the large number of correlations examined, all p-values were adjusted for false-discovery rate (FDR). Next, multinomial logistic regression models estimated the association between VT error and cognitive status, before and after adjustment for each ROI GMV which had a significant association with both VT error and cognitive function. The attenuation of odds ratios of postural sway predicting cognitive status after entering each ROI GMV was calculated to examine how much each ROI GMV contributed to the association between VT error and cognitive status. Sensitivity analyses were stratified by MCI subgroups.

3.0. Results

Participants (n=179) were 82.4±2.5 years old, 56% female and 56% white (Table 1). Cognitive status was associated with race, education, history of hypertension, and history of stroke/TIA at p<0.05, but not with age, sex, or neuromuscular factors (Table 1). Pairwise comparisons revealed these differences were significant for individuals with NC as compared to MCI or dementia for race, history of hypertension, and history of stroke/TIA; and for NC as compared to dementia for education. Individuals with NC were also more likely to have a lower BMI compared to MCI. Thus, race, education, history of hypertension, history of TIA/stroke, and BMI were included as covariates in subsequent multivariable models. As expected, the three groups differed on cognitive tests (Table 1); differences were statistically significant for all, except Stroop and the Digit Backwards.

Table 1.

Overall and pairwise comparisons of demographic and neuromuscular characteristics by cognitive status for 179 participants.

	Normal Cognition n=76		Mild Cognitive Impairment n=63		Dementia n=40		p-value2
	Mean (SD) or n (%)	p-value1 NC vs MCI	Mean (SD) or n (%)	p-value1 MCI vs Dementia	Mean (SD) or n (%)	p-value1 NC vs Dementia
Age	82.2 (2.3)	0.93	82.2 (2.5)	0.12	83.0 (2.9)	0.10	0.18
Female Sex	41 (54%)	0.85	35 (56%)	0.66	24 (60%)	0.53	0.82
White Race	54 (71%)	0.003	29 (46%)	0.92	18 (45%)	0.006	0.003
BMI	26.8 (4.6)	0.02	28.5 (3.5)	0.61	28.1 (5.2)	0.17	0.06
Education (≤high school)	27 (36%)	0.15	30 (48%)	0.05	27 (67.5%)	0.001	0.005
History of Diabetes	17 (22%)	0.63	12 (19%)	0.47	10 (25%)	0.75	0.77
History of Hypertension	58 (76%)	0.01	58 (92%)	0.94	37 (92.5%)	0.03	0.01
History of Coronary Heart Disease	14 (18%)	0.44	15 (24%)	0.28	6 (15%)	0.64	0.52
History of Stroke/TIA	0 (0%)	0.001	8 (13%)	0.07	1 (2.5%)	0.17	0.002
Joint Pain	2 (3%)	0.50	0 (0%)	N/A	0 (0%)	0.99	0.69
Knee Extensor Strength (Nm)	86 (32)	0.99	86 (25)	0.34	81 (34)	0.38	0.59
LE Vibration Sense	18 (24%)	0.97	14 (22%)	0.07	15 (38%)	0.07	0.13
Ankle/Brachial Index	1.06 (0.18)	0.49	1.04 (0.17)	0.40	1.01 (0.18)	0.17	0.35
Modified Mini-Mental State	94 (6)	<0.210	92 (13)	<0.867	92 (6)	<0.041	<0.274
List A Sum trials 1–5	46.99 (9.11)	<0.001	38.12 (10.49)	<0.001	24.48 (11.95)	<0.001	<0.001
List B Immediate Recall	6.1 (1.7)	<0.001	4.5 (1.9)	<0.001	2.9 (2.0)	<0.001	<0.001
List A Short delay recall	8.5 (2.7)	<0.001	6.0 (3.1)	<0.001	2.6 (3.2)	<0.001	<0.001
List A Long delay free Recall	9.4 (2.5)	<0.001	5.8 (3.2)	0.36	4.4 (11.5)	0.001	<0.001
Rey Figure, Immediate Recall	16.9 (4.1)	<0.001	11.5 (4.5)	<0.001	7.2 (5.1)	<0.001	<0.001
Rey Figure, Delayed Recall	16.1 (4.1)	<0.001	11.3 (4.1)	<0.001	7.2 (5.1)	<0.001	<0.001
Rey Figure, Percent Retained	79.5 (17.1)	<0.001	57.6 (22.3)	0.007	41.7 (28.5)	<0.001	<0.001
Digit Spans Forwards	6.2 (1.0)	0.18	5.9 (1.2)	0.11	5.4 (1.0)	0.002	0.01
Digit Spans Backwards	3.4 (1.2)	0.12	3.0 (1.1)	0.38	2.8 (1.3)	0.03	0.06
Rey Figure, copy	21.5 (1.8)	0.006	20.3 (2.8)	<0.001	17.1 (3.4)	<0.001	<0.001
Boston Naming (modified) Spontaneous Correct	25.8 (3.3)	0.005	22.6 (4.4)	0.002	16.2 (6.9)	<0.001	<0.001
Fluency Animal	17.0 (4.9)	<0.001	13.6 (4.3)	<0.001	9.6 (4.3)	<0.001	<0.001
Fluency FAS	39.8 (12.8)	0.005	33.4 (13.0)	<0.001	19.7 (10.9)	<0.001	<0.001
Stroop Color Word Test Interference Score	22.4 (20.6)	0.97	22.5 (10.7)	0.005	11.0 (6.8)	0.12	0.12
Trails A Time	50.5 (19.4)	0.01	60.2 (23.6)	<0.001	96.0 (42.1)	<0.001	<0.001
Trails B Time	119.7 (52.7)	0.004	157.8 (92.1)	0.01	219.1 (125.9)	<0.001	<0.001

¹Pairwise p-values based on between group-comparisons, using ANOVAs for continuous variables and chi-square tests for categorical variables.

²Overall p-values based on across group-comparisons, using ANOVAs for continuous and chi-square tests for categorical variables (Significant differences in bold). NC= Normal Cognition; MCI= Mild Cognitive Impairment; TIA=transient ischemic attack; LE=lower extremity.

Spearman’s correlations were statistically significant between VT error and cognitive tests of attention, concentration, construction/visual discrimination, executive control function, and language (p<=0.01); associations with VT error were not significant at p<0.05 with memory, nor with postural-sway during conventional tasks (Supplemental Table 1). Medio-lateral postural sway variability (RMS) during the conventional tasks differed by cognitive status in unadjusted models (Figure 1), but differences were not statistically significant for any of the groups (fully adjusted p>0.05), thus RMS sway during the conventional tasks was not considered for further analyses. Analyses were similar for antero-posterior sway and when stratified by MCI subgroup (data not shown). The RMS VT error during the VT task was greater for those with dementia (mean: 4.53, SD=1.65) compared to those with MCI (mean: 4.00, SD=1.57), or with NC (mean: 3.55, SD=1.21) (Figure 1).

Figure 1. Mean and standard deviation of postural sway of conventional and novel tasks by cognitive status.

RMS=root mean square; COP=center of pressure. NC= Normal Cognition; MCI= Mild Cognitive Impairment. *Indicates p-values ≤0.05. **Indicates p-value ≤0.001. P-values are from multinomial regression models adjusted for age, race, education, BMI, history of hypertension, history of stroke/TIA and time between visits.

In multinomial models adjusted for age, race, education, BMI, history of hypertension, history of stroke/TIA, and time between visits, odds were significantly greater for dementia compared to NC (OR=2.30; 95% CI=1.40, 3.77, p=0.001) or MCI (OR=1.87; 95%CI= 1.16, 3.02; p=0.01), but not for MCI compared to NC (OR=1.45; 95% CI=0.99, 2.13; p=0.30). Results of ROC analyses indicate the association of VT error with cognitive status was comparable to that of the 3MS score (e.g. similar AUC, see Table 2) or other tests (not shown).

Table 2.

Results of receiving operating characteristic curves, predicting the probability of having dementia as compared to normal cognition or mild cognitive impairment for Visual Tracking Task error and Modified Mini-Mental State Examination tests.

	Dementia compared to Normal Cognition	Dementia compared to Mild Cognitive Impairment
	Area Under the Curve (Standard Error), p value
Visual Tracking Task	0.725 (.051), p<0.0001	0.655 (0.056), p=0.009
Modified Mini-Mental State	0.870 (.037), p<0.0001	0.774 (0.051), p<0.0001

Correlations of neuroimaging measures of total brain by cognitive status and postural sway were statistically significant for GMV but not WMH. Total brain atrophy significantly and negatively correlated with VT error in postural sway (r=−0.32, FDR- and total brain atrophy-adjusted p<0.0001), and with cognitive status (r=−0.16, FDR- and total brain atrophy-adjusted p=0.04). Total brain WMH were not statistically significantly associated with VT error or with cognitive status (r=0.04, FDR- and total brain atrophy-adjusted p=0.76 and r=0.16, FDR- and total brain atrophy-adjusted p=0.08, respectively).

Correlations of ROI GMV with both VT error and cognitive status were statistically significant for bilateral hippocampi, parahippocampi, entorhinal cortex, and for left inferior and superior parietal lobes (r range −0.19 to −0.32; FDR- and total brain atrophy-adjusted p-value ≤0.05 for all; see Supplemental Table 2). These associations were robust to further adjustment for covariates (age, race, education, BMI, history of hypertension, history of stroke/TIA, and time between visits) for bilateral hippocampi and parahippocampi, right entorhinal cortex and left inferior parietal lobule, but not for left entorhinal cortex or left superior parietal lobule (Table 3). Of note, the GMV of the primary motor, supplementary, cingulate cortices (anterior, middle and posterior) and cerebellum were also significantly and negatively associated with postural sway, but not with cognitive function (Supplemental Table 2).

Table 3.

Results of multivariable linear regression models testing the associations between Visual Tracking Task error and gray matter volume of selected regions of interest.

Regions of interest1	β coefficient2 (p value)
Hippocampus Left	−0.64 (0.001)
Hippocampus Right	−0.91 (0.0002)
Parahippocampus Left	−0.56 (0.001)
Parahippocampus Right	−0.45 (0.007)
Entorhinal Cortex Left	−0.64 (0.08)
Entorhinal Cortex Right	−0.82 (0.03)
Inferior Parietal Lobe Left	−0.29 (0.03)
Superior Parietal Lobe Left	−0.21 (0.15)

All models are adjusted for atrophy, age, race, education, BMI, history of hypertension, history of stroke/TIA and time between visits.

¹Regions of interest were selected based on associations significant after adjustment for brain atrophy and false discovery rate, see Supplemental Table 1.

²βs are multiplied by a factor of 1000 for ease of presentation.

In multinomial logistic regression analyses of postural sway predicting cognitive function adjusted for total brain atrophy, we found that further adjustment for either bilateral hippocampus GMV or left parahippocampus GMV led to >20% attenuation in the odds of dementia compared to NC (Table 4). Similarly, adjustment for left hippocampus GMV also led to >20% attenuation in the odds of dementia compared to MCI (Table 4). Adjustments for other ROIs yield more modest attenuations of the odds of dementia compared to NC or MCI (Table 4).

Table 4.

Results of multinomial regression models testing the association of Visual Tracking Task error with cognitive status: dementia, mild cognitive impairment, or normal cognition.

	Dementia compared to Normal Cognition	Dementia compared to Mild Cognitive Impairment
	OR (95% CI), % Attenuation of OR after addition of region of interest
Model 1	2.15 (1.38, 3.36) n/a	1.58 (1.05, 2.38) n/a
Model 1 + Hippocampus Left	1.84 (1.16, 2.93) 26. 96%	1.46 (0.96, 2.24) 20.69%
Model 1 + Hippocampus Right	1.81 (1.14, 2.86) 29.57%	1.53 (1.00, 2.34) 8.62%
Model 1 + Parahippocampus Left	1.90 (1.20, 3.02) 21.74%	1.56 (1.01, 2.40) 3.45%
Model 1 + Parahippocampus right	2.03 (1.28, 3.21) 10.43%	1.55 (1.00, 2.39) 5.17%
Model 1 + Entorhinal Cortex Right	2.10 (1.33, 3.31) 4.35%	1.58 (1.01, 2.39) 0%
Model 1 + Inferior Parietal Lobe Left	1.97 (1.25, 2.12) 15.65%	1.56 (1.08, 2.34) 3.45%

Model 1 is adjusted for total brain atrophy. Regions of interest are selected based on models presented in Table 2. NC=normal cognition; MCI=mild cognitive impairment.

4.0. Discussion

Our results indicate that lateral postural control while performing a VT task can distinguish between older adults with dementia and those with either NC or MCI, with a predictive value comparable to that of other cognitive tests, and robust to adjustment for other dementia-related factors. The association of VT-related postural control and cognition is likely due to shared selected neural substrates, including memory-related regions in the medial temporal lobe, e.g. the bilateral hippocampus. We further found GMV of dementia-related ROIs^22–24, namely the bilateral hippocampi and left parahippocampus, significantly attenuated the postural control-cognition relationship by more than 20%. These results suggest regional GMV loss is contributing both to cognition and postural control during a VT task. Our results add to existing literature on physical clinical biomarkers of cognitive impairment with the use of a novel VT task, which increases postural control requirements through greater need for cognitive input, including attention.^29,30 Further, we investigated a wide range of ROIs to identify the regions related to both postural control and a range of cognitive function. Taken together, our findings add important evidence to the ongoing initiatives to unravel the shared mechanisms underlying postural, gait, and cognitive changes in aging and neurodegenerative diseases.⁴ Better understanding of these mechanisms can help develop novel tools for early dementia detection.

We did not find significant associations between cognitive status and postural sway during conventional tasks (e.g. while standing on level or foam surfaces with eyes open or closed). This negative finding is consistent with prior studies in similarly aged populations without motor disorders^7,9,13. Of note, Kido et al¹³ found associations with one-leg standing time but not with postural control during quiet stance; Deschamps et al⁷ found associations with velocity-based but not position-based metrics of postural control. Our results underscore the importance of examining comprehensive metrics of postural control obtained during conditions that challenge posture. The VT task of our study appeared to challenge postural control to a greater degree as compared to other conventional tasks (e.g. eyes closed or standing on foam), likely because it included both postural and cognitive elements; in fact, VT-related postural control was strongly correlated with performance on visual tasks and timed attention tasks.

Our study found that both VT-related postural control and cognitive status were related to smaller GMV of regions in the medial temporal (bilateral hippocampi, parahippocampi, entorhinal cortex) and parietal lobe (inferior, superior). While the association of these medial temporal regions with memory is well estabilished¹⁴, the association with postural control in older adults has only been recently shown^12,9,13. A study including several regions of interest, found that worse postural control was associated with the nucleus accumbens in those with AD, but not with the hippocampus or any of the other regions hereby identified after adjustment⁹; however, this study did not report associations with NC or MCI. The parietal lobe is engaged with both higher-order motor planning³¹ and conversion to dementia³²; however, no study to date has examined these associations with both cognitive and postural control in the same population.

The link between hippocampus, postural control and dementia is of particular interest. A recent study has also identified the hippocampus as a shared neural substrate of gait slowing and cognitive impairment.³³ Gait speed is dependent on, among other things, sufficient postural control.³⁴ These convergent results strengthen the hypothesis that there are shared neural pathways between physical and cognitive functions that may explain parallel declines in these functions with age.²⁴

We found smaller GMV in several regions was associated with VT-related sway but not with cognitive status, specifically primary motor, supplementary, cingulate cortices (anterior, middle and posterior) and cerebellum. A recent review of neuroimaging of postural control³¹ reports postural control is regulated by a widely distributed network of cortical and subcortical network of regions. However most studies to date are from younger adults, or clinical populations, and often focused only on few regions. New studies in older adults with a broader number of regions are needed to replicate our results.

The negative findings of our study are also of interest. We did not find a significant association between VT-related postural sway and basal ganglia or thalamus. While the role of these regions in balance control is well studied for clinical populations and Parkinson’s Disease³¹, they are largely understudied in older adults without neurologically overt motor disorders. Accurate segmentation is challenging for these small regions and may explain at least in part these negative results³⁵. The lack of association between VT-related postural control and white matter hyperintensities was counter to what we expected; it cannot be excluded that this was due to the limited sample size or a limited range of WMH volumes. Of note, the association between white matter hyperintensities and dementia was also not significant. Likewise, although VT-related postural control significantly differed between dementia and MCI, and between dementia and NC groups, differences between MCI and NC groups were more modest. This could be due to the heterogeneity of the MCI group; however, analyses were similar when non-amnestic and amnestic subgroups were examined. Future studies should examine whether other biomarker-supported MCI groups would differ from CN in postural control.

There are several strengths but also limitations that should be considered when interpreting these results. A comparison of postural control between three adjudicated cognitive groups allowed us to expand on prior studies that only assessed NC and/or dementia; specifically, we were able to assess a gradient of strength in the relation of posture with cognitive function. Standardized methods were used for measurements of postural control while performing tasks of varying difficulty. Cognitive function adjudication was based on robust clinical measurement and extensive review of medical history. The neuroimaging methodologies have been well validated and allowed for careful inspection a wider range of regions compared to prior studies. Several limitations should also be considered. First, we have assessed only one of several components serving postural control, and other postural control metrics may have a stronger association with cognition and should be explored in future studies. In addition, postural sway was measured at only one time; therefore, decline in postural control with decline in cognition could not be examined. There was a gap of time between postural control measurement and cognitive adjudication; however, results remained significant after adjustment for this interval of time between visits. Nonetheless, participants who died or were lost to follow-up during that interval of time were necessarily excluded from the analyses; this could have biased our estimates toward the null, since participants who are less cognitively healthy are more likely to die or be lost to follow-up. Lastly, this study did not include amyloid or tau as biomarkers of cognitive status; however, volumetric measures of the hippocampus, combined with extensive neuropsychological assessment and clinical history and adjudication are well-recognized predictors of cognitive impairment and dementia.³⁶

4.1. Conclusions

Current neuroimaging and blood biomarkers for cognitive function have limited utility at the population level due to their invasive nature and restricted eligibility (e.g. due to metal implants, difficulty lying flat, etc.).³⁷ Similarly, comprehensive cognitive assessments require repeated measurements over time, and referral to specialists for assessment and interpretation. Therefore, there is a need for novel, easily obtainable, non-invasive biomarkers of cognitive decline in older adults. Clinical markers to distinguish those with NC or MCI from those with dementia may be of particular use, because they could point to those who are optimal targets for early interventions and/or treatments. In this cross-sectional study, postural control under specific conditions appeared to have a similar predictive value when compared to the 3MS test, which is a simple and widely used test of cognition. Longitudinal studies should assess whether postural control, in combination with 3MS test, may be a potentially clinically useful biomarker of dementia in population studies.