Intraindividual Variability in Executive Functions but Not Speed of Processing or Conflict Resolution Predicts Performance Differences in Gait Speed in Older Adults

Abstract

Background.

The relationship between executive functions (EF) and gait speed is well established. However, with the exception of dual tasking, the key components of EF that predict differences in gait performance have not been determined. Therefore, the current study was designed to determine whether processing speed, conflict resolution, and intraindividual variability in EF predicted variance in gait performance in single- and dual-task conditions.

Methods.

Participants were 234 nondemented older adults (mean age 76.48 years; 55% women) enrolled in a community-based cohort study. Gait speed was assessed using an instrumented walkway during single- and dual-task conditions. The flanker task was used to assess EF.

Results.

Results from the linear mixed effects model showed that (a) dual-task interference caused a significant dual-task cost in gait speed (estimate = 35.99; 95% CI = 33.19–38.80) and (b) of the cognitive predictors, only intraindividual variability was associated with gait speed (estimate = −.606; 95% CI = −1.11 to −.10). In unadjusted analyses, the three EF measures were related to gait speed in single- and dual-task conditions. However, in fully adjusted linear regression analysis, only intraindividual variability predicted performance differences in gait speed during dual tasking (B = −.901; 95% CI = −1.557 to −.245).

Conclusion.

Among the three EF measures assessed, intraindividual variability but not speed of processing or conflict resolution predicted performance differences in gait speed.

The relationship between specific cognitive functions and gait performance in nondemented older adults is reproducible, robust, and evidenced by studies that involve different methodologies (1,2). Studies that utilize neuropsychological tests to assess the relationship between cognitive functions and mobility reveal that attention and executive functions (EF) play a key, though not exclusive role, in predicting variance in gait performance and decline (1,3–6). Dual-task paradigms that involve walking while performing a cognitive interference task can be conceptualized as an attention-demanding mobility stress test that was recently shown to predict the risk of frailty, disability, and mortality in older adults (7).

EF is a multifaceted construct that subsumes cognitive abilities such as planning, organizing, set shifting, inhibition, and conflict resolution. Allocating attention to competing task demands, as assessed via dual-task paradigms, is a facet of EF (8) that is rooted in classic cognitive psychology (9) and is distinct from other aspects of this construct (10). EF are often evaluated using clinical neuropsychological tests that by design are multidetermined. Consequently, identifying specific cognitive component(s) that underlie the relationship between EF and gait has been an unresolved issue to date. Determining, however, the components of EF that are most relevant to gait has critical implications with respect to the discovery of shared brain substrates, etiology, risk assessment, and cognitive interventions for mobility decline and disability.

One key cognitive determinant in many measures of EF is speed of processing. The effect of decreased processing speed on performance differences on a wide range of cognitive functions including EF in older adults has been well documented (11). It is, therefore, reasonable to expect that speed of processing might account for the relationship between EF and gait speed. Indeed, processing speed was a key component in latent neuropsychological factors (1,3) and individual measures such as the Digit Symbol Modalities Test (4,6) that were used to assess the relationship between EF and mobility outcomes. Furthermore, a recent structural magnetic resonance imaging study reported that the relationship between prefrontal cortex (PFC) volume and gait speed was attenuated by performance on the Digit Symbol Modalities Test (12). This finding was interpreted by the authors as evidence that shared brain substrates subserved cognitive and motor tasks that were dependent on processing speed. However, two important caveats should be considered. First, processing speed is one of several cognitive processes that are involved in EF in general and in Digit Symbol Modalities Test performance in particular. Second, regions in the PFC subserve complex cognitive functions that include but are not limited to conflict resolution, allocation of attention to competing task demands as well as executive control and monitoring of the efficiency of responses during cognitive task performance (13,14). We have recently demonstrated that functional involvement of the PFC in walking was increased under dual-task gait conditions (15). These findings suggest that processing speed alone might not account for the relationship between EF and gait in older adults, especially in attention-demanding locomotive tasks.

One approach to identifying key EF component(s) that predict gait performance is to use more refined cognitive paradigms that empirically separate processing speed from other EF measures. The flanker task (16) requires individuals to inhibit prepotent responses and resolve conflicting visual information and is mediated by the anterior cingulate and dorsolateral PFC (17). Importantly, cognitive paradigms that employ variants of the flanker task separate processing speed from conflict resolution and other measures of attention and EF (18). Intraindividual variability, considered as inconsistency in cognitive performance within a person, can be directly assessed using the flanker task. The taxonomy of intraindividual variability includes several definitions. However, variability on repeated trials of a single cognitive task, which is increased in aging (13) and dementia (19), is considered a defined facet of EF that measures the individual’s ability to monitor and optimize the consistency of responses during acute mental effort (20).

Using the flanker task, the current study was designed to determine whether processing speed, conflict resolution, and intraindividual variability in EF predicted variance in gait performance as assessed in single- and dual-task conditions in nondemented older adults. In light of recent studies implicating the functional involvement of the PFC in locomotion tasks (15), we hypothesized that when assessed simultaneously conflict resolution and intraindividual variability but not processing speed would predict gait performance in older adults. We further hypothesized that the relationship between EF measures and gait would be more pronounced in dual compared with single task conditions.

Methods

Participants

Participants in this study were recruited from an ongoing cohort study of older adults entitled “Central Control of Mobility in Aging.” The primary aims of the study are to determine cognitive and brain predictors of mobility performance, mobility decline, and disability in aging. Potential participants 65 and older, identified from a population list of lower Westchester county, New York, were first contacted by mail and then by telephone inviting them to participate. A structured telephone screening interview was administered to potential participants to assess for eligibility. The telephone interview consisted of verbal consent, a brief medical history questionnaire, mobility questions, and validated cognitive screens to exclude dementia. Exclusion criteria were inability to speak English, inability to independently ambulate, dementia, significant loss of vision and/or hearing, current or history of neurological or psychiatric disorders, recent or anticipated medical procedures that may affect mobility, and receiving hemodialysis. After completing the telephone interview, eligible individuals were scheduled for two 3-hour in-person visits at the research center. During the visits, participants received comprehensive neuropsychological, cognitive, psychological, and mobility assessments as well as a structured neurological examination. The neuropsychological assessment included the Repeatable Battery for the Assessment of Neuropsychological Status (21) and other standardized tests covering several cognitive domains including literacy, language, visual spatial abilities, attention, memory, and EF (22). Diagnoses of dementia were assigned according to the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (23) at consensus diagnostic case conferences as previously described (24). Clinical gait abnormalities were formally assessed during the neurological examination (25). Central Control of Mobility in Aging participants are followed longitudinally at yearly intervals. Written informed consents were obtained in person according to study protocols and approved by the institutional review board.

Measures

Gait speed (cm/s) served as the outcome and was measured using an instrumented walkway with embedded pressure sensors (GAITRite, CIR Systems, Havertown, PA). The GAITRite system is widely used in clinical and research settings and has excellent psychometric properties. The walkway measures 8.5 m × 0.9 m × 0.01 m (L × W × H) with an active recording area of 6.1 m × 0.61 m (L × W). Gait speed and the mean gait characteristics demonstrated excellent test–retest reliability in our (reliability of two repeated trials, r = .96) (4) and in other research settings (Intraclass Correlation Coefficient ≥ .80) (26). Using gait speed as a single dependent measure is justified because of the statistical robustness of this measure and the literature implicating it in critical clinical outcomes such as falls, disability, and mortality (27,28). Further, we wanted to minimize concerns regarding multiple comparisons given that gait and cognitive outcomes were assessed in single- and dual-task conditions.

Walking protocol.

Participants were asked to walk on the instrumented walkway at their usual pace in a quiet well-lit room wearing comfortable footwear and without any attached monitors. The participants walked for one trial under two separate task conditions: (1) normal walk (NW) and (2) walking while counting backward by 7s (WWC). In WWC, the participants were asked to pay equal attention to both the walking and cognitive interference task (ie, equal priority) as previously described (1,2). Single task cognitive performance was assessed by asking participants to perform serial subtractions (counting backward by 7s) while standing for 10 seconds. In this single-task condition, responses generated within the 10-second interval were recorded. In WWC, however, responses generated during the duration of the task were recorded. Therefore, the percent of correct responses was used to directly compare performance on the counting task in single- and dual-task conditions. Significant positive correlations, in the expected direction, between the total correct responses and percent correct in single-task (r = .455, p < .001) and dual-task (r = .504, p < .001) conditions alleviated concerns that a low number of responses might inflate estimates of cognitive accuracy. The order of the three tasks was counterbalanced to reduce practice effects.

Cognitive predictors

Flanker task.

All sessions were conducted in the same quiet and well-lit room. A research assistant supervised the training and testing procedures of each participant. The Flanker task that was used in this study consisted of a six-trial training block, which was followed by three blocks of 12 trials. The reliability of this abbreviated task was excellent (Guttman split-half coefficient; r = .901; p < .001). The trials were randomly presented using Eprime 2.0 software (Psychology Software Tools, Inc., Pittsburgh, PA). The participant’s task was to report the direction of the target (ie, central) arrow by pressing the left mouse button for leftward pointing arrows and the right mouse button for rightward pointing arrows. Participants were required to fixate on a central fixation point (a black cross) visible on the monitor and respond to each stimulus as quickly as possible without making errors. Reaction time (RT) and accuracy data were recorded. The stimuli employed were rows of five black lines with arrowheads that point leftward or rightward and appear directly to left or right of a central fixation cross. The target stimulus is the central arrow and is always surrounded by two flankers on each side. There are two flanker types: congruent flankers (two arrows on each side that point in the same direction as the central target stimulus) and incongruent flankers (two arrows on each side that point in the opposite direction of the central target stimulus). The height of the arrows (ie, target and flanker stimuli) was 0.64 cm and the width of the arrows was 3.81 cm as previously reported (20). The flanker effect was used to operationalize conflict resolution. It requires individuals to resolve conflicting visual information. The mean RT of all congruent trials was subtracted from the mean RT of all incongruent trials to calculate the flanker effect (18). Larger flanker scores are indicative of worse performance (ie, longer RTs required for conflict resolution). The coefficient of variation ([SD/mean] × 100) was used to calculate an intraindividual variability measure for each participant. Speed of processing was assessed by obtaining an overall individual mean reaction (RT) based on all valid trials. Participants were required to attain 75% accuracy to participate in the current study and only correct responses were used for statistical analysis as commonly done in other studies (18,20).

Covariates

Structured clinical interviews were used to identify self-reported medical diagnoses. Consistent with our previous studies (1,2), dichotomous rating (presence or absence) of diabetes, chronic heart failure, arthritis, hypertension, depression, stroke, Parkinson’s disease, chronic obstructive lung disease, angina, and myocardial infarction was used to calculate a disease comorbidity summary score (range 0–10). Additional covariates included age, education, gender, and the presence of clinical gait abnormalities. The study clinician conducted a structured clinical assessment of gait by examining walking patterns as previously described (25). Clinical gait abnormalities, if present, were graded on severity levels and subtyped to either neurological or non-neurological causes.

Statistical Analysis

Linear mixed effects model was used to examine the effect of dual-task interference on gait speed. Specifically, in this model, the effect of dual-task interference is operationalized as the difference in gait speed in the two-level repeated measure walking condition (ie, contrast of single vs dual task) adjusting for demographic and cognitive covariates. Correlations examined the relationship between the cognitive predictors and gait speed stratified by task condition (single vs dual task). Separate multivariate linear regressions were used to examine the relations of the cognitive predictors and covariates, entered simultaneously, with gait speed as assessed in single- and dual-task conditions. Data were inspected descriptively and graphically and model assumptions were formally tested. Fully adjusted models controlled for gender, age, education, disease comorbidity, and clinical gait abnormalities. Statistical analyses were performed using IBM SPSS version 20 (IBM, Somers, NY).

Results

Participants

Recruitment to Central Control of Mobility in Aging for baseline and follow-up annual visits is ongoing. For the current investigation, 247 participants had complete mobility and cognitive baseline data. Of this sample, 13 individuals were excluded due to poor flanker task performance that was deemed unreliable as per previous reports (overall accuracy < 75% correct) (20). Exclusion of the 13 participants did not change the sample characteristics in terms of key demographic variables nor were the main results materially different. The study sample had a mean age (76 years; ±6.1), education (14.6 years; ±3.1), and gender distribution (%female = 55) that is broadly representative of the demographic characteristics of individuals in this age group who reside in the study catchment area. The mean Repeatable Battery for the Assessment of Neuropsychological Status standardized total score (92.3 ± 11.6) was in the average range of cognitive function, and the disease comorbidity summary score (1.2 ± 1.0) was indicative of relatively good health. Baseline cohort characteristics are presented in Table 1.

Table 1.

Summary of Sample Characteristics, Cognitive Functions, and Gait Speed at Baseline

	Mean (SD)	Median	Range
Age (years)	76.48 (±7.15)	76.00	65.00 to 95.00
Education (years)	14.58 (±3.08)	14.00	7.00 to 28.00
Disease comorbidity index	1.16 (±1.0)	1.00	0.00 to 4.00
RBANS (standard total score)	92.47 (±11.66)	92.00	65.00 to 137.00
Flanker overall RT (ms)	669.20 (±131.48)	640.10	442.20 to 1230.80
Flanker effect (ms)	90.11 (±67.78)	80.15	−36.00 to 549.47
Flanker COV	17.10 (±5.17)	16.17	8.04 to 47.24
Gait speed NW (cm/s)	100.61 (±22.89)	98.65	26.70 to 170.19
Gait speed WWC (cm/s)	64.83 (±25.36)	61.70	10.00 to 156.30
Serial 7 (% correct)	88.24 (±19.21)	100	25.00 to 100.00
Serial 7 WWC (% correct)	80.71 (±23.77)	88.88	0.00 to 100.00
Total sample (n)	234
Women: number (%)	128 (55)

Notes: Flanker COV = coefficient of variation of the Flanker task using all valid trials; Flanker effect = conflict resolution defined as the difference between congruent vs incongruent trials; Flanker overall RT = average reaction time based on all valid trials; NW = normal pace walk; RBANS = Repeatable Battery for the Assessment of Neuropsychological Status; Serial 7 (% correct) = percent of correct responses on the Serial 7 subtraction task when performed as a single task; Serial 7 WWC (% correct) = percent of correct responses on the Serial 7 subtraction task when performed while walking; WWC = walking while counting.

Linear mixed effects model was used to assess the effect of dual tasking on gait speed adjusting for the cognitive predictors and covariates (Table 2). Mean gait speed under single-task condition was 100.61 (±22.89) cm/s. The dual-task interference resulted in a significant decrement in gait speed (estimate = 35.99; 95% CI = 33.19–38.80). Of the three cognitive predictors, only intraindividual variability was associated with walking speed (Table 2).

Table 2.

Linear Mixed Effect Model: Cognitive Predictors of Gait Speed

	Estimate	t	95% CI	p
Variable
NW vs WWC	35.99	25.28	33.19 to 38.80	<.001
Age	−.80	−4.48	−1.15 to −.45	<.001
Gender	.24	.095	−4.53 to 5.34	.924
Education	.30	.76	−.48 to 1.10	.323
Comorbidity score	−2.04	−1.65	−4.48 to .40	.101
Gait abnormalities	−12.72	−4.91	−17.82 to −7.62	<.001
Flanker RT (ms)	−.007	−.724	−.02 to .010	.470
Flanker effect (ms)	−.01	−.552	−.05 to .029	.602
Flanker COV	−.606	−2.360	−1.11 to −.10	.019

Notes: Flanker COV = coefficient of variation of the Flanker task using all valid trials; Flanker effect = conflict resolution defined as the difference between congruent vs incongruent trials; Flanker overall RT = average reaction time based on all valid trials; NW = normal pace walk; WWC = walking while counting.

Unadjusted analyses revealed that the three cognitive predictors were significantly correlated with gait speed in single- and dual-task conditions in the expected direction (Table 3).

Table 3.

Bivariate Correlations: Cognitive Predictors of Gait Speed in Single- and Dual-Task Conditions

	Gait Speed: NW		Gait Speed: WWC		Serial 7		Serial 7 WWC		RBANS
	r	p	r	p	r	p	r	p	r	p
Flanker task variables
Processing speed	−.163	.013	−.142	.030	−.198	.002	−.092	.163	−.316	<.001
Flanker effect	−.159	.015	−.152	.020	−.149	.023	−.106	0.107	−.107	.103
Flanker COV	−.189	.004	−.242	<.001	−.222	.001	−.091	0.165	−.275	<.001

Notes: Flanker COV = coefficient of variation of the Flanker task using all valid trials; Flanker effect = conflict resolution defined as the difference between congruent vs incongruent trials; Flanker overall RT = average reaction time based on all valid trials; NW = normal pace walk; p = p value; r = Pearson correlation; RBANS = Repeatable Battery for the Assessment of Neuropsychological Status; Serial 7 = percent of correct responses on the Serial 7 subtraction task when performed as a single task; Serial 7 WWC = percent of correct responses on the Serial 7 subtraction task when performed while walking; WWC = walking while counting.

When the cognitive predictors were entered simultaneously in multivariate linear regression models, intraindividual variability was marginally related to gait speed in normal walk but significantly predicted walking speed in WWC (Table 4). Speed of processing and conflict resolution were not related to gait speed in either walking condition. As expected, the correlations of speed of processing with intraindividual variability (r = .364; p < .001) and conflict resolution (r = .431; p < .001) as well as the correlation between intraindividual variability and conflict resolution (r = .299; p < .001) were all significant. However, the magnitude of these correlations and formal testing of model assumptions for the multivariate linear regressions were not indicative of collinearity.

Table 4.

Multivariate Linear Regressions Predicting Gait Speed in Single- and Dual-Task Conditions

	Gait Speed NW: R = .457; R 2 = .299; p = <.001			Gait Speed WWC: R = .391; R 2 =. 153; p < .001
Variables	B	95% CI	p	B	95% CI	p
Age	−.889	−1.261 to −.517	<.001	−.597	−1.052 to −.141	.010
Gender	−.756	−6.017 to 4.505	.777	2.578	−3.852 to 9.008	.430
Education	.488	−.345 to 1.329	.255	−.119	−1.148 to .910	.820
Comorbidity score	−2.50	−5.09 to .097	.059	−.986	−4.160 to 2.188	.541
Gait abnormalities	−14.069	−19.475 to −8.665	<.001	−9.561	−16.165 to −2.956	.005
Flanker RT (ms)	−.009	−.32 to .013	.426	−.005	−.032 to .023	.734
Flanker effect (ms)	−.009	−.052 to .033	.665	−.013	−.065 to .038	.608
Flanker COV	−.481	−1.17 to .056	.079	−.901	−1.557 to −.245	.007

Notes: Flanker COV = coefficient of variation of the Flanker task using all valid trials; Flanker effect = conflict resolution defined as the difference between congruent vs incongruent trials; Flanker overall RT = average reaction time based on all valid trials; NW = normal pace walk; WWC = walking while counting.

Discussion

We examined the associations of processing speed, intraindividual variability, and conflict resolution with gait speed in single- and dual-task conditions. These three measures were extracted from a single cognitive paradigm. Hence, confounds that hinder direct comparisons among measures of EF that often differ in terms of stimulus type, sensory modality through which stimuli are administered or the required output were addressed in the current study. When assessed individually increased intraindividual variability, slower processing speed and longer time to resolve visually conflicting information were all associated with slower gait speed. However, in fully adjusted models, intraindividual variability in EF emerged as the only potent predictor of gait speed, specifically under dual-task conditions where attention demands were maximized.

Evidence for the decline in processing speed and its effect on cognitive functions in aging is ubiquitous (11). The effect of age on the decline in gait speed is also well documented. Therefore, presumptions that speed of processing accounts for the relationship between performance on neuropsychological measures of EF and gait speed appeared to have been supported by strong theoretical and empirical research. However, our results suggest otherwise. Converging evidence from linear mixed effects and regression models suggests that intraindividual variability but not speed of processing predicted differences in gait speed performance. Intraindividual variability and processing speed, although extracted from the same paradigm, were sufficiently different. Moreover, careful inspection of the data and formal testing indicated that this finding could not be attributed to violations of model assumptions. Our index for speed of processing has been used in many other studies (18,20) and also correlated in the expected direction with overall levels of cognitive function. Therefore, the lack of association between speed of processing and gait speed in the fully adjusted models cannot be attributed to methodological limitations that are inherent in our measure of processing speed.

Our findings also revealed that conflict resolution was not related to gait speed in the fully adjusted models. Variations of the flanker effect have been used in numerous studies (18). Hence, it is unlikely that the lack of association between conflict resolution and gait speed is attributed to measurement limitations of the former. The differential relationship of speed of processing, conflict resolution, and intraindividual variability with gait speed is informative insofar as it provides insight into specific cognitive processes and brain substrates that are critical in higher order control of locomotion. Intraindividual variability is considered a defined facet of EF that measures the individual’s ability to monitor and optimize the consistency of responses during acute mental effort (20). Extrapolating from our findings would suggest that the ability to monitor and sustain mental effort but not conflict resolution or processing speed plays a role in explaining variance in gait speed, especially under conditions that tax the attention system.

Structural neuroimaging studies demonstrated that reduced volume in the PFC (29) and white matter disease (30) were both associated with slower gait speed in older adults. However, knowledge concerning the functional brain correlates of mobility has been more limited because active locomotion cannot be directly assessed in traditional neuroimaging methods. A recent functional Near Infrared Spectroscopy study provided, nonetheless, evidence for the critical role the PFC has in higher order control of gait, especially under attention-demanding conditions (15). Interestingly, the extant but limited literature suggests that activation in the PFC (31) and white matter volume (32) have unique associations with response time variability. By extension, it is reasonable to postulate that the correlations reported herein between intraindividual variability and gait speed may be attributed shared gray and white matter brain substrates. This hypothesis, however, should be examined in future research.

Limitations

Causality cannot be determined based on cross-sectional associations. Longitudinal research is necessary to determine whether our measures of EF predict decline in gait speed. Longitudinal effects on both the cognitive predictors and gait outcomes would be critical to further establish their temporal relationship. Using gait speed as a single-dependent measure is justified statistically and in light of the literature implicating this measure in critical clinical outcomes such as falls, disability, and mortality (27,28). The extant literature confirms a reliable association between clinical measures of EF and gait speed irrespective of whether the latter is assessed as a single (1,2) or latent (33,34) variable. It is noteworthy that poor EF (35,36) and increased variability in RT (37) are also related to increased risk of falls. However, the relationship of EF with other quantitative measures of gait is less clear. For instance, EF were related to single (38) but not latent (33,34) measures of gait variability. Future research should examine the relationship between the refined EF predictors of the current study, intraindividual variability in particular, with single and latent measures of gait variability. Although beyond the scope of the current article, it is noteworthy that here and in previous publications (4,39) the presence of clinical gait abnormalities was a potent predictor of gait speed. In a recent study, we also found that the presence of clinical gait abnormalities moderated dual-task decrements in gait speed but not in cognitive performance (40). Future work should determine whether clinical gait abnormalities and measures of EF have independent or synergistic effects on gait performance. Finally, it is important to note that a significant portion of the variance in gait speed performance here and in other studies remained unexplained. Hence, future studies should aim to identify new factors as well as existing variables that may mediate or moderate the effects of EF and other predictors on gait performance.

Clinical Implications

Among measures of EF, intraindividual variation and allocating of attention to competing task demands are key predictors of gait speed. In previous work, we have demonstrated that dual tasking predicts important clinical outcomes (7). Hence, incorporating these specific EF measures into risk assessment procedures of mobility decline and disability should be considered. In addition, we have previously shown that cognitive remediation enhanced gait speed (41). Identifying specific EF measures that are relevant to gait performance may also be used to further influence the design and implementation of cognitive remediation for the purpose of improving mobility.

Funding

This research was supported by the National Institutes on Aging (R01AG036921; PI: R.H.).