An association of cognitive function with mobile metrics of community walking in older cancer survivors: a pilot study

Abstract

Introduction:

Older cancer survivors have an elevated risk for mobility dysfunction compared to their cancer-free peers. Despite the established link between cognitive function and community walking in older cancer-free adults, little is known about this relationship in older cancer survivors. This pilot study aimed to evaluate the association of performance-based and self-reported cognitive function with mobile metrics of community walking collected by a wearable Fitbit device.

Materials and Methods:

This study enrolled older cancer survivors (mean age 73 years old, range 65–83; 98% White; 50% female) within five years of completing primary treatment. Cognitive function, specifically executive function and processing speed was collected with the digit symbol substitution test (DSST) and self-reported cognition was evaluated by the Patient Reported Outcomes Measurement Information System- Cognitive (PROMIS-Cog). Continuous walking data from Fitbit wearable devices were collected passively over four weeks. To examine associations between DSST and PROMIS-Cog with mobile measures of walking, we conducted bivariate correlation and multivariable linear regression analyses adjusting for age, education, and number of comorbidities.

Results:

In bivariate analyses, higher DSST scores were correlated with higher step count and peak cadence and lower fragmentation of walking in daily life (r= 0.48–0.51, p<.01). Higher PROMIS-Cog scores were correlated with higher peak cadence (r=0.32, p<0.05), trended towards correlation with step count (r=0.30, p=0.06), and were not correlated with fragmentation of walking (r= −0.24, p=0.13). In multivariable models adjusting for age, presence of graduate level education, and number of comorbidities, higher DSST scores were independently associated with higher peak cadence, step count, and demonstrated a trend towards lower fragmentation of walking in daily life, but PROMIS-Cog was not independently associated with any mobility metrics. Similar results for association of DSST with walking when models included adjustment for PROMIS-depression scale, receipt of chemotherapy treatment, or when education was defined by presence of a bachelor’s degree.

Discussion:

This study suggests an association between cognitive functions of executive function and processing speed with mobile metrics of community walking in older cancer survivors. Understanding how cognitive function affects walking may help identify new rehabilitation targets for older cancer survivors.

INTRODUCTION

By 2040, it is anticipated that 19 million Americans over 65 years of age will survive cancer, constituting 25% of the older adult population.¹ Although there is excitement regarding the rapid expansion in the number of older adults surviving cancer, cancer and its treatments often result in long-term impairments.^2,3 Specifically, walking performance may be impaired compared to cancer-free peers, increasing the risk of falls, fractures, and missed cancer treatments.^4–6 However, the underlying causes of impaired walking control in older cancer survivors are understudied.^4,7,8 Therefore, fall risk assessments and mobility interventions for older cancer survivors would benefit from targeting the underlying contributors to walking control.

Although walking appears as a simple task, in fact it is very complex, requiring effective functioning of locomotor factors including the cardiopulmonary, musculoskeletal, and the central and peripheral nervous systems. Emerging evidence in older adults demonstrates that as a consequence of aging on locomotor factors, walking control relies more heavily on the central nervous system, and specifically on prefrontal cortical substrates responsible for cognition.⁹ Older adults with higher cognitive processing speed and healthier prefrontal cortical grey and white matter demonstrate resilience to aging-related walking declines, including in the presence of peripheral locomotor impairment.^10,11 Moreover, assessment of walking in community environments may provide a more ecologically valid measure of walking as it requires tasks such as talking while walking or navigating sloping or uneven terrains, making it an optimal outcome to evaluate the cognitive-walking control relationship. Despite knowledge of the simultaneously accentuated and accelerated aging in older cancer survivors, the relationship between cognitive function—assessed either through neuropsychological testing or self-report—and community walking has not been established in this population.^12,13

Previous investigations have associated cognitive function capacities with laboratory evaluations of balance, gait speed, physical performance and reported falls.^14,15 Herein, this study will build on this work by examining the association of cognitive function with mobile metrics of community walking in older cancer survivors. This study operationalizes cognitive function by assessing (1) the Digit Symbol Substitution Test (DSST), which measures cognitive processing speed and executive functioning, and (2) self-report of cognitive function through Patient Reported Outcomes Measure Information System Item Bank v2.0 Cognitive Function (PROMIS-Cog). This study employs Fitbit devices to capture gait cadence and walking patterns in real-world environments. This device records metrics such as peak gait cadence, which is closely linked to gait speed, and walking patterns, including daily step count and the average lengths of walking periods over a day known as walking fragmentation. The primary goal of this study was to determine if DSST correlates with mobile metrics of community walking in older cancer survivors over a four-week period. The study also explored the relationship of PROMIS-Cog to community walking.

METHODS

Participants

Participants (n = 40) were recruited from two community research registries as well as oncology clinics. Eligibility criteria included: (1) had completed primary treatment (including surgery, chemotherapy, and/or radiation) within the previous five years for cancer of any type and stage except basal cell skin carcinoma; (2) aged 65 years or older; (3) were independently ambulatory (use of walking aids was acceptable); and (4) owned a smartphone capable of syncing the consumer wearable device and executing the study applications. All participants provided informed consent and the study was approved by the University of Pittsburgh and Duquesne University Institutional Review Boards.

Design

The study included an in-person assessment in which cognitive and physical function tests were performed by a physical therapist or cancer rehabilitation nurse with assistance from trained physical therapy students. After the in-person assessment, the Fitbit app was downloaded to participants’ Android or iOS smartphones. They were provided a Fitbit Inspire 3 and asked to wear it continuously for four weeks except while charging.

Independent Variables

The primary independent variable selected a priori was cognitive processing speed and executive functioning (attention, working memory, cognitive flexibility) measured by the DSST.¹⁶ The DSST is a paper-and-pencil task that requires copying as many novel symbols corresponding to numbers as possible in 90 seconds.¹⁶ DSST is a validated cognitive test and has been extensively used to evaluate processing speed and executive functioning in older adults including its relationship to walking.¹⁰

A secondary independent variable was the PROMIS Item Bank v2.0 - Cognitive Function (PROMIS-Cog) t-score, a measure of self-reported cognition, where a higher score indicates higher levels of cognition. PROMIS-Cog is a universal assessment of patient-perceived cognitive difficulties over four weeks and has been validated in cancer survivors.¹⁷ Concepts include mental acuity, concentration, verbal and nonverbal memory, verbal fluency, perceived change in cognitive functions, extent of cognitive impairment interference on daily functioning, other’s observation of cognitive impairments, and impact of cognitive dysfunction on quality of life.¹⁷

Dependent Variables

Mobile Measures of Community Walking

Fitbit was used to collect physical activity data, and minute-level step count data was accessed using the Fitbit Application Programming Interface (API). We used our Reproducible Analysis Pipeline for Data Streams (RAPIDS) to compute day-level (24 h from midnight to midnight) behavioral features from Fitbit data.¹⁸ These behavioral features were then averaged over all days on which sensor data was collected for at least eight hours, resulting in one summary feature per participant. For Fitbit data, we used the presence of heart rate data to infer continuous wear time and excluded days with fewer than eight hours of wear time.

For the current analyses we focused on the following Fitbit sensor features selected a priori to capture aspects of community walking. Specifically, we captured peak gait cadence (maximum steps/min), daily step count, and walking fragmentation (activity-to-sedentary transition probability, computed as the reciprocal of the mean active bout length, multiplied by 100). For walking fragmentation, a lower value indicates greater lengths of walking bouts. Peak gait cadence is strongly related to gait speed and, due to gait speed’s established relationship to mobility outcomes, was the primary dependent variable for community walking.¹⁹ Secondary dependent variables were daily step count, which captures amount of walking, and walking fragmentation, which is associated with diminished stamina for activity.

Covariates

Sociodemographic and medical history variables recorded included age, sex, race, highest level of education completed, cancer history and treatment, and medical comorbidities; see Table 1. Additionally, the Eastern Cooperative Oncology Group (ECOG) performance status was obtained from the participants at the time of testing through questions related to gross physical functioning, rather than being rated by a clinician.²⁰ Self-report of depression was recorded by PROMIS-Ca Bank 1.0, where a higher score indicates higher levels of depression.²¹

Table 1:

Baseline characteristics of the full sample

Characteristics with SD or n (%)	Full Sample (n = 40)
Mean age in years	73.4 (4.9)
Female	20 (50%)
White race	39 (98%)*
Education
High school or less	2 (5%)
Some college/2-year degree	9 (23%)
Bachelor’s degree	9 (23%)
Graduate degree	19 (48%)
Cancer type
Prostate	17 (43%)
Breast	9 (23%)
Leukemia/lymphoma	5 (13%)
Other	9 (23%)
Cancer treatment
Surgery	28 (70%)
Chemotherapy	15 (38%)
Radiation	21 (53%)
Immunotherapy	11 (28%)
Hormonal Therapy	8 (20%)
Other	3 (8%)
Mean months since cancer treatment	24.95 (22.02)
Presence of comorbidity
Asthma, emphysema, or bronchitis	4 (10%)
Arthritis or rheumatism	25 (63%)
Diabetes	6 (15%)
Digestive problems (e.g., ulcer, colitis)	3 (8%)
Heart trouble (e.g., angina)	6 (15%)
Kidney disease	2 (5%)
Liver problems (e.g., cirrhosis)	2 (5%)
ECOG score of 0	35 (88%)
Falls and Near Falls over 4 weeks	14 (35%)
Mean Short Physical Performance Battery score	9.57 (2.29)
Mean peak cadence (step/min)	100.35 (17.50)
Mean daily step count	6428 (3282)
Mean walking fragmentation	0.457 (0.108)
PROMIS-depression score	46.5 (6.5)
PROMIS-cog score	51.1 (6.6)
DSST	46.7 (10.9)

^*n=1, Black

SD- standard deviation, ECOG- Eastern Cooperative Oncology Group, PROMIS- Patient Reported Outcomes Measurement Information System, DSST- Digit Symbol Substitution Test

Statistical Analysis

Descriptive statistics for demographics, medical and cancer history, and cognitive assessment were performed for all participants (n = 40). Results are displayed as percentages or means with standard deviations depending on the type of variable. To examine associations between measures of cognitive function and community walking metrics obtained from the Fitbit, bivariate Pearson correlations were assessed between DSST and PROMIS-Cog with peak cadence, step count, and walking fragmentation. Next, multivariable linear regression models were conducted for the cognitive function measures (DSST and PROMIS-Cog) to assess if they were independently associated with walking metrics when adjusting for age, number of comorbidities, and education (defined by the presence or absence of a graduate level degree). Of note, number of comorbidities was collapsed into three groups: 0 = 23%, 1 = 55%, 2 or more = 23%. Exploratory models adjusted for education defined by presence or absence of bachelor’s degree as well as self-report of depression symptoms by PROMIS. Additionally post-hoc t-tests and bivariate correlations were performed to assess the impact of chemotherapy treatment status on cognition and mobile metrics of community walking. A plot of the residuals of the predictors was obtained separately for each dependent variable and each demonstrated a nearly linear relationship, signifying normal distributions. All tests were two-tailed, with an alpha level of 0.05, and analysis was completed with Stata 17.0.

RESULTS

The baseline characteristics for the full cohort (n=40) except education (n=39), as education status was missing in one participant, are depicted in Table 1. Overall, the participants were White (98%), older adults (65–83 years old), 50% female, and most had bachelor’s degree or higher (71%). The three most common cancer types were prostate, breast, and leukemia/lymphoma and many received treatments such as surgery, radiation, chemotherapy, and immunotherapy. The mean time from cancer treatment was 25 months with a standard deviation of 22 months. The most frequent comorbidity reported was arthritis or rheumatism (63%) followed by diabetes (15%) and heart trouble (15%). The participants were of a high physical functional status evidenced by 88% reporting an ECOG score of 0. On average, participants had 26 ± 3.6 days of valid Fitbit data with a range of 12–29 days. The range of DSST and Promis-Cog in the study’s participants were 23–66 and 37–64, respectively.

Pearson correlation coefficients for cognitive function with mobile community walking metrics are reported in Table 2. Higher DSST scores were moderately and significantly correlated with higher peak cadence, higher step count, and lower walking fragmentation. Higher reported cognitive function via PROMIS-Cog was weakly and significantly correlated with peak cadence but not step count or walking fragmentation.

Table 2:

Pearson correlations of measures of cognitive function with mobile metrics of community walking in older cancer survivors (n=40).

	DSST	PROMIS-Cog
Peak Cadence	0.485***	0.324*
Step Count	0.513**	0.301
Walking Fragmentation	−0.483**	−0.244

DSST: Digit Symbol Substitution Test, PROMIS-Cog: Patient Reported Outcomes Measurement Information System

^*p<0.05

^**p<0.01

^***p<0.001

Multivariable linear regression models were adjusted for age, education, and number of comorbidities with measures of cognitive function, either DSST and PROMIS-Cog as primary determinants for peak cadence, step count, and walking fragmentation. Table 3 presents the multivariable models using DSST and Table 4 presents the models using PROMIS-Cog.

Table 3:

Linear regression models with digit symbol substitution test as primary independent variable for mobile metrics of community walking in older cancer survivors (n=39). Unstandardized (B) and standardized

	Model for Peak Cadence			Model for Step Count			Model for Walking Fragmentation
Model R2	0.487			0.426			0.418
	B (SE)	B	p	B (SE)	B	p	B (SE)	B	p
Age	−1.20 (0.51)	−0.33	0.02	−234 (99.6)	−0.35	0.03	0.008 (0.003)	0.37	0.02
Education	−3.88 (5.10)	−0.11	0.45	−634 (1000)	−0.10	0.53	0.003 (0.03)	0.02	0.92
Comorbidities
0	Base			Base			Base
1	−9.87 (5.50)	−0.28	0.08	−2400 (1080)	−0.37	0.03	0.066 (0.04)	0.31	0.08
2+	−18.6 (6.58)	−0.45	0.008	−1190 (1290)	−0.15	0.36	0.078 (0.04)	0.30	0.08
DSST	0.55 (0.25)	0.34	0.03	125 (49.0)	0.42	0.02	−0.003 (0.002)	−0.31	0.07

(β) estimates from linear regression analyses relating predictors to community walking metrics.

DSST: Digit Symbol Substitution Test

SE: standard error

Table 4:

Linear regression models with PROMIS-Cog as primary independent variable for mobile metrics of community walking in older cancer survivors (n=39). Unstandardized (B) and standardized (β) estimates from linear regression analyses relating predictors to community walking metrics

	Model for Peak Cadence			Model for Step Count			Model for Walking Fragmentation
Model R2	0.460			0.366			0.377
	B (SE)	B	p	B (SE)	B	p	B (SE)	B	p
Age	−1.62 (0.50)	−0.45	0.003	−326 (99.9)	−0.49	0.002	0.010 (0.003)	0.47	0.003
Education	−0.53 (4.79)	−0.02	0.91	180 (964)	0.03	0.85	−0.018 (0.032)	−0.082	0.58
Comorbidities
0	Base			Base			Base
1	−10.2 (5.64)	−0.29	0.08	−2506 (1135)	−0.39	0.03	0.070 (0.027)	0.32	0.07
2+	−16.8 (7.04)	−0.41	0.02	−928 (1416)	−0.12	0.52	0.074 (0.047)	0.29	0.12
PROMIS-Cog	0.68 (0.39)	0.24	0.09	130 (78.1)	0.25	0.10	−0.0027 (0.003)	−0.15	0.31

PROMIS-Cog: Patient Reported Outcomes Measurement Information System

SE: standard error

Overall, after adjusting for age, education, and number of comorbidities, higher DSST was significantly associated with higher peak cadence (ΔR²= 0.07) and higher step count (ΔR²= 0.11); higher DSST scores were associated with less walking fragmentation, but this was not statistically significant (ΔR²= 0.06). In contrast, PROMIS-Cog was not significantly associated with peak cadence (ΔR²= 0.05, p=0.09), step count (ΔR²= 0.05, p=0.10), or walking fragmentation (ΔR²= 0.02, p=0.31) after covariate adjustment. Additional models were constructed to adjust for education defined by the presence or absence of bachelor’s degree and for self-reported depression symptoms using PROMIS-Dep and these models demonstrated similar results (data not shown).

Post-hoc exploratory analyses were conducted to further examine the influence of chemotherapy on cognition and mobility. From t-test comparisons, older cancer survivors treated with chemotherapy had statistically significantly worse peak cadence and walking fragmentation (p<0.05) than did those who had not received chemotherapy, as well as an association with lower step counts but this was not statistically significant (p=0.07); there were no differences in cognitive function, Supplementary Table 1. Bivariate Pearson correlations for DSST with peak gait cadence, step count, and walking fragmentation, performed separately in older cancer survivors treated with and without chemotherapy, were significant in both groups (p<0.05), but the correlation coefficients ranged 0.63–0.71 and 0.28–42, respectively, Supplementary Table 2. Multivariable linear regression models adjusting for receipt of chemotherapy when evaluating DSST as a predictor for peak cadence, step count, and walking fragmentation demonstrated similar results for DSST’s contribution as the models reported in Table 3 and receipt of chemotherapy was not a significant predictor.

DISCUSSION

In this pilot study of older cancer survivors, cognitive function was correlated with mobile metrics of community walking including higher peak cadence, higher total step count, and lower walking fragmentation in daily life. Specifically, cognitive function assessed by DSST, but not self-reported cognitive function from PROMIS-Cog, continued to be associated with mobile community walking metrics after adjustment for age, education, and number of comorbidities in multivariable models. Collectively, these findings suggest that cognitive function may provide unique insight into community walking performance of older cancer survivors.

The association of cognitive function, specifically DSST, with walking in older cancer survivors is supported by the recent literature. Previously, executive function capacities of inhibition and set-shifting were strongly associated with sit to stand function, single leg stance time, and falls in older cancer survivors treated with chemotherapy.¹⁵ Additionally, set-shifting performance has been linked to mobility evaluations in older cancer survivors including walking speed and Short Physical Performance Battery scores.^14,22 Although there is no universal consensus, it is contended that DSST measures both executive functions and cognitive processing speed.^16,23 Despite the widely recognized bidirectional relationship of cognitive function, specifically with DSST, with walking in cancer-free older adults this has been minimally assessed in older cancer survivors.^10,24,25 In cancer survivors, physical activity, including walking, is undergoing active investigation as an intervention to improve cancer-related cognitive impairment.^26,27 This study extends the current literature by demonstrating that cognitive function is associated with walking, suggesting that this relationship may be bidirectional, similar to their cancer-free peers. A bidirectional relationship would be clinically advantageous, as walking performance could help identify cancer survivors with possible cognitive impairment, and improving cognitive function may help rehabilitate walking and mobility. While higher scores on the DSST, a measure of executive function, correlated with higher function in two mobility measures in older cancer survivors—cadence, step count, and walking fragmentation, after adjusting for age, education, and number of comorbidities—it is crucial to note that these associations do not confirm a causative link to prior cancer treatments or chemotherapy-related cognitive impairment.

Importantly, chemotherapy has been associated with declines in balance and walking function among cancer survivors and is a significant contributor to falls.⁸ In this study, post-hoc exploratory analyses were conducted to explore the relationship between chemotherapy and mobile metrics of community walking and cognition. The findings suggest that older cancer survivors treated with chemotherapy exhibit worse mobile metrics of community walking compared to those treated without chemotherapy, which aligns with the existing literature of laboratory mobility assessments.⁸ However, contrary to previous studies, we did not find significant differences in self-reported cognition or DSST scores between cancer survivors treated with and without chemotherapy.^28,29 This discrepancy may be attributed to our small sample size, possibly resulting in a type II error. The bivariate correlation analysis revealed a significant relationship between DSST scores and community walking metrics in older cancer survivors, regardless of chemotherapy treatment status. Interestingly, the correlation coefficients for DSST were larger for those treated with chemotherapy. These analyses, while exploratory in nature, may suggest that the relationship between DSST scores and community walking metrics may vary depending on chemotherapy status. In summary, the impact of chemotherapy on cognitive walking control warrants further investigation in future studies to better understand this relationship.

PROMIS-Cog is a validated measure that offers unique insights into the cognitive status of cancer survivors but in the present study it was not associated with mobile metrics of community walking in multivariable models.³⁰ Considering the significant role of cognitive motor control in aging adults, it was hypothesized that PROMIS-Cog might correlate with community walking metrics. Despite existing research supporting a strong correlation between PROMIS physical function and mobility, the specific relationship of PROMIS-Cog with walking has been rarely explored in older adults, regardless of cancer status.³¹ In adults with multiple sclerosis, PROMIS-Cog was correlated with walking fatigability and symptoms of depression but not as strongly with neuropsychological testing.^32–35 Similarly, among older head and neck cancer survivors, PROMIS-Cog was not significantly associated with cognitive performance as measured by the NIH cognitive toolbox neuropsychological battery, but it was correlated with symptoms of depression, anxiety, fatigue, and pain.³⁶ These findings suggest that self-reported and neuropsychological assessments of cognitive function may contribute differently, yet complementarily, to our understanding of walking performance. Given the critical role of walking ability in predicting both overall survival and quality of life, future research should focus on elucidating the specific impacts of various cognitive function assessments on walking performance in cancer survivors.^37,38

This pilot study has limitations associated with a cross-sectional design, single site setting, small sample size and a homogenous participant group of high physical function status. Given these limitations, the generalizability of our findings to broader clinical populations requires verification through additional research. Therefore, future research should aim for a broader demographic scope, including varied physical functional status, educational and racial backgrounds, and ideally, assess cognitive function longitudinally and using a neuropsychological battery. Despite the need for varying the factors mentioned, it may be important to focus on a single cancer type due to the known heterogeneity from multiple cancer types and treatments to better generalize to these specific populations of cancer survivors.

These limitations notwithstanding, the findings may have both biologic plausibility and clinical implications. Growing evidence in older adults indicates that cognition, and in particular executive function and processing speed play an important role in the control of walking. Specifically, in older age the control of walking shifts from being automated, relying primarily on peripheral locomotor factors and motor networks, to requiring cognitive processes (e.g., more attention) and more activation of the prefrontal cortex; this is especially evident in the community environment where attention may be divided (e.g., talking while walking or navigating street crossings).^9,11 Therefore, given that advanced aging is a known process within cancer survivors it is biologically plausible that cognitive processing speed and executive functioning would be significantly associated to community walking metrics including gait cadence, a proxy for gait speed, and total step count, a measurement of total walking distance.^12,13 This finding has important clinical implications as cognitive function and its neural correlates could potentially mediate mobility resilience in older cancer survivors who are experiencing advanced aging and have concomitant peripheral locomotor risk factors (e.g., decreased proprioception from chemotherapy induced peripheral neuropathy).^11,15,42 Therefore, it is reasonable that cognitive-motor control may be an important rehabilitation target to improve mobility resilience in older cancer survivors.

The collective findings of this pilot study suggest that cognitive processing speed and executive functioning are associated with gait cadence, total step count, and walking fragmentation among older cancer survivors in their community settings. Moving forward, future research is required to understand the mechanisms on how different aspects of cognitive function mediate community walking performance, especially in the context of locomotor risk factors. Lastly, this study supports the developing literature for investigating cognitive-motor control in older cancer survivors to improve mobility resilience.