An Examination of the Longitudinal Relationship Between Cognitive Function and Physical Activity Among Older Breast Cancer Survivors in the Thinking and Living With Cancer Study

Abstract

Background

Older cancer survivors are at risk for cognitive decline. Physical activity can improve cognition, and better cognitive function may facilitate greater physical activity.

Purpose

We examined the potential bidirectional relationship between cognitive function and physical activity in older breast cancer survivors and controls.

Methods

The sample included women with newly diagnosed, nonmetastatic breast cancer (n = 395) and women without cancer (n = 374) ages 60–98. Participants were recruited as part of a larger multisite study, assessed prior to systemic therapy, and followed yearly for 36 months. Attention, processing speed, and executive function was measured using six neuropsychological tests, self-reported cognitive function using the Perceived Cognitive Impairments subscale of the Functional Assessment of Cancer Therapy—Cognitive Function , and physical activity using the International Physical Activity Questionnaire-Short Form. Separate random intercepts cross-lagged panel models were used to examine the between- and within-person effects for survivors and controls, controlling for age, education, and study site.

Results

Survivors reported significantly less physical activity than controls at baseline (1,284.92 vs. 2,085.98 MET min/week, p < .05). When survivors reported higher activity, they simultaneously had better objective cognition at 12 months (β = 0.24, p = .04) and reported better perceived cognition at 12 and 24 months (β = 0.25, p = .03), but this relationship was not seen in controls. Cognition did not predict subsequent physical activity or vice versa in either group.

Conclusions

Cognition and physical activity are cross-sectionally associated in survivors, but the expected prospective relationships were not found.

Higher physical activity was associated with better objective and perceived cognitive function among older breast cancer survivors. However, physical activity did not predict cognitive function, and cognitive function did not predict physical activity at yearly follow-ups.

Introduction

In the USA, women diagnosed with breast cancer aged 65 and older are among the largest groups of cancer survivors, comprising an estimated 2.5 million women [1], and 75% of survivors are expected to be 65 and older by the year 2030 [2]. Breast cancer and its treatment affect several cognitive domains for some survivors, including higher-order cognitive functions such as attention and executive function [3]. Breast cancer survivors also have changes in the structure and function of prefrontal brain regions associated with executive function [4, 5]. Furthermore, more than 50% of breast cancer survivors report perceived cognitive impairment [6, 7]. Evidence of compensatory hyperactivation in frontal brain regions indicates that breast cancer survivors may require greater effort to achieve similar cognitive performance to noncancer controls, which may contribute to cases with reported impairment despite normal objective cognitive performance [8]. Cancer-related cognitive impairment is associated with a variety of treatment modalities (e.g., chemotherapy, hormone therapy, radiation) [3, 7] and has even been found before systemic cancer therapy [9].

Older cancer survivors may be especially vulnerable to cancer-related cognitive impairment due to underlying aging and potential for accelerated aging from cancer treatments [9, 10]. Biological mechanisms for accelerated aging due to cancer treatments include DNA damage, inflammation, and cellular senescence [11]. Accelerated aging among older adults with cancer is currently under investigation [12], yet older cancer survivors may have greater vulnerability to accelerated aging than younger survivors and older adults without cancer due to the combined effects of aging and cancer treatments. For example, both aging and cancer treatments are associated with increased sleep disturbance [13, 14], which has been linked to increased fatigue, lower cognitive function, and biological evidence of aging including cellular senescence [15–17].

Randomized clinical trials have shown that physical activity can improve objective and perceived cognitive function as well as other symptoms (e.g., sleep disturbance, fatigue) and functional status in cancer survivors [18–21] and older adults [22]. Despite the cognitive and health benefits of physical activity, only 33% of older cancer survivors meet national guidelines of 150 min of moderate or 75 min of vigorous intensity physical activity (i.e., at least 450 metabolic equivalents of task [MET] minutes) per week [23, 24], and 40% are sedentary (i.e., report no leisure time physical activity) [25]. Although older cancer survivors and older adults in general show comparable levels of nonadherence to physical activity guidelines [23, 25], older cancer survivors are understudied in physical activity research. Physical activity also tends to decline during systemic cancer treatments such as chemotherapy and hormone therapy [26, 27]. Furthermore, older survivors are potentially vulnerable to cancer-related cognitive and accelerated aging [9, 10], which suggests that relationships between cognition and physical activity observed in older adults without cancer histories may not generalize to older cancer survivors, especially as they progress through systemic cancer treatment and into survivorship.

According to self-regulation theory and prior research, difficulties with cognitive functions such as attention, processing speed, and executive function (APE) can interfere with organizing and carrying out exercise activities [28–31]. In the general population of older adults, cognitive function and physical activity appear to have a positive bidirectional relationship [30] with better executive function in particular associated with increased exercise [30, 31], and exercise resulting in improved cognition [32]. In fact, one study of older adults found that over a 6-year period, executive function predicted physical activity with 55% greater magnitude than the effect of physical activity on subsequent executive function [30]. While physical activity shows promise for improving cognitive functioning in cancer survivors [33, 34], the effect of higher-order cognitive functions such as attention and executive function on subsequent physical activity has not been examined in older cancer survivors.

We used data from the Thinking and Living with Cancer (TLC) prospective cohort [12, 35] of breast cancer survivors ages 60+ and frequency-matched noncancer controls to conduct secondary analyses of the bidirectional relationship between physical activity and both objective and perceived cognitive function. Including both objective and perceived cognitive function is consistent with expert recommendations because neuropsychological testing alone does not capture cancer-related cognitive impairment as reported by survivors [36]. The aim of this project was to test for these potential bidirectional effects within the context of natural changes in cognition and physical activity in the first years following cancer diagnosis and treatment. Results will also build the evidence base for the potential utility of physical activity to address cognitive difficulties associated with cancer and aging, while also accounting for the potential impact of cognition on physical activity.

Methods

The study protocol was approved by all institutional review boards and met HIPAA standards (clinicaltrials.gov identifier NCT03451383). All participants provided written informed consent. The current project used presystemic treatment and follow-up data at 12, 24, and 36 months from TLC participants recruited between August 2010 and December 2016. The overall aim of the TLC study is to determine how cancer treatment and aging affect the trajectory of cognitive function in older breast cancer survivors [12], and recruitment and follow-up are ongoing. Accrual sites include six academic and nine community practices in five U.S. regions.

Participants

Eligible survivors were female, aged 60 years and older, English speaking, with primary nonmetastatic breast cancer (American Joint Committee on Cancer V6 stages 0–III). Survivors were ineligible if they had ever had a stroke, moderate or severe traumatic brain injury (i.e., loss of consciousness >60 min or evidence of structural changes), major psychiatric or neurodegenerative disorder, or another cancer diagnosis or systemic cancer treatment within the last 5 years. Eligibility was assessed via participant self-report and confirmed via survivors’ medical record. Survivors were excluded at follow-up if they experienced breast cancer recurrence or developed a new cancer or condition that rendered them ineligible (e.g., stroke). Noncancer controls were eligible if they met the same criteria as the survivors apart from a breast cancer diagnosis, and they were ineligible at follow-up for the same reasons as survivors. Noncancer controls were matched to survivors on age (within 5 years), race/ethnicity, education (<high school, high school to <college, college+), and geographic locale.

Eligible, consenting participants were screened using the Mini Mental State Examination (MMSE) [37] and the Wide Range Achievement Test, 4th edition (WRAT-4), a measure of reading and a proxy for literacy. Those who scored ≤24 on the MMSE or <3rd grade level on the WRAT-4 were excluded, as this indicated that they may not have the ability to complete the study (n = 1 survivor, n = 1 control). Controls were also ineligible if their performance on baseline cognitive tests were >3 standard deviations below normative values (n = 8). Figure 1 shows the number of participants in the final analytic sample (Supplementary Material 1 by study site).

Fig. 1.

CONSORT diagram for sample of older breast cancer survivors and noncancer controls. Participants can skip an assessment and remain eligible for follow-up.

Procedures

The procedures for the TLC study have been published [12, 35]. Briefly, survivors were recruited from medical oncology clinics at participating sites and enrolled after diagnosis and prior to systemic hormonal therapy or chemotherapy. Controls consisted of friends of the survivors and older adults recruited from communities in the areas where survivors lived. Baseline assessments for survivors were completed before systemic cancer therapy and after surgery, apart from seven survivors treated with neoadjuvant therapy who completed baseline prior to systemic cancer therapy and surgery. Demographic information was collected during the baseline interview. Neuropsychological testing (approximately 55 min, completed in-person) and self-report measures (approximately 30–40 min, completed either in-person, over the telephone, or via mail) were administered at baseline and yearly follow-up assessments. All participants received a $50 gift card per assessment.

Assessments were performed by trained research staff who completed certification for neuropsychological test administration every 6 months. Medical records were reviewed for survivors’ clinical data (e.g., diagnosis date, chemotherapy, hormonal therapy) at baseline and annually to determine recurrence and hormonal and other therapies.

Measures

Objective cognitive function

The APE domain was measured by six neuropsychological tests [9]. The six tests were (a) The Neuropsychological Assessment Battery (NAB) Digits Forward and (b) Digits Backward, (c) The Trail Making Test Part A (TMT-A) and (d) Part B (TMT-B) from the Halstead-Reitan Neuropsychological Battery [38], (e) The Digit Symbol-Coding subtest of the Wechsler Adult Intelligence Test-III [39, 40], and (f) The Controlled Oral Word Association Test (COWAT) [41]. The factor structure for the APE domain demonstrated acceptable internal consistency for survivors (domain Cronbach’s α = 0.74–0.77 at baseline, 12, and 24 months of follow-up) and controls (α = 0.66–0.71) [9, 12]. Neuropsychological tests that were used to assess APE in this study have established reliability and validity in diverse older adult populations and have been recommended by the International Cognition and Cancer Task Force [42]. Practice effects with multiple assessments were minimized by using alternate forms of the NAB [43]. The APE domain score was used to remain consistent with reporting procedures for the TLC study [12], and to avoid reliance on one specific cognitive assessment [44].

Perceived cognition

Perceived cognitive function was measured via the 18-item Perceived Cognitive Impairments (PCI) subscale of the Functional Assessment of Cancer Therapy—Cognitive Function (FACT-Cog) questionnaire [45]. Participants were asked to respond to statements about the frequency of cognitive impairments over the past 7 days on a 5-point Likert scale (0 = “never”, 4 = “several times a day”). The PCI scores ranged from 0 to 72, where higher scores indicated better perceived cognitive function.

Physical activity

Physical activity was measured with the International Physical Activity Questionnaire-Short Form (IPAQ-SF) that assessed vigorous and moderate activities and walking (i.e., light activity) during leisure and work over the past week (e.g., “During the last 7 days, on how many days did you do vigorous physical activities like heavy lifting, digging, aerobics, or fast bicycling?” and “How much time did you usually spend doing vigorous physical activities on one of those days?”). Days were multiplied by time (in minutes), and activities were weighted by METs (vigorous = 8, moderate = 4, walking = 3.3) to estimate the continuous number of MET minutes per week [46]. IPAQ data have been moderately correlated with measures of physical fitness independent of body mass index, suggesting that the IPAQ captures physiologically relevant activity [47–49].

Covariates

Age, education, and recruitment site were considered covariates. Demographic covariates were determined a priori based on prior research indicating that these variables are associated with cognitive function and/or physical activity in older adults with and without breast cancer [9, 21].

Data Analysis

Frequencies, means, and standard deviations were computed in SPSS version 27.0 [50] to characterize all study variables. Continuous variables were examined for normality (i.e., skewness <3.0 and/or kurtosis <8.0) and outliers [51]. The Log_e transformation was employed for continuous MET minutes due to excessive skewness. APE domain scores were the mean of z scores for the neuropsychological tests scored such that a higher score reflected better performance, and z scores were calculated from the baseline mean and standard deviations for age and education strata-matched controls [52]. Perceived cognitive function was examined as continuous PCI scores.

Preliminary analyses

Paired-samples t-tests were used to compare survivors and noncancer controls on demographics (i.e., age, education) and outcomes to characterize the sample. Participants who at least partially completed follow-up time points were compared with those who refused or were lost to follow-up on demographic and clinical characteristics and outcomes.

Primary analyses

Random intercepts cross-lagged panel models (RI-CLPM) were used to test multidirectional relationships between physical activity and cognitive function over time using Mplus version 8.0 [53, 54]. Separate RI-CLPM for survivors and controls were employed to test the relationships between APE or PCI and physical activity. RI-CLPM are structural equation models which estimate the stable between-person effects and time-varying within-person effects [54]. The RI-CLPM simultaneously estimated multiple dimensions of the cognition–physical activity relationships. The models examined the overall association between cognition and physical activity at the between-person level, where relationships evaluated cognition/physical activity when participants have higher or lower levels compared with others over time. The models also evaluated within-person effects, capturing how physical activity and cognition vary when a participant has higher levels than her average over time. Within-person effects included the relationships between cognition and physical activity at each time point and the lagged relationships—physical activity at one time point predicting cognition at a subsequent time point—as well as the reciprocal relationship of cognition predicting subsequent physical activity. Full information maximum likelihood estimation was employed to reduce bias due to missing data. The final models control for age, education, and recruitment site. Model fit was assessed holistically by a combination of the χ² test, root mean square error of approximation (RMSEA ≤0.05 is good fit, ≤0.10 is adequate fit), comparative fit index (CFI ≥0.95 is acceptable fit), and Tucker Lewis index (TLI ≥0.95 is acceptable fit) [51].

Secondary analyses

Linear mixed models were used to examine physical activity and cognition over time and by treatment type (i.e., chemotherapy, hormone therapy alone) compared with controls. Additional RI-CLPM were also used to examine the relationship between physical activity and individual APE tests.

Results

Sample Characteristics

Survivors were on average 68.0 years old (SD 5.9, range 60–98), controls were 67.9 years old (SD 7.0, range 60–91), the average participant in each group had completed some college education, and most participants were non-Hispanic and White (Table 1). There were no significant differences between survivors and controls on age, education, race, or ethnicity. There were significant differences between survivors and controls on insurance status, with more controls than survivors covered through Medicare. Most survivors had stage 0 or 1 disease (67.3%) and received systemic hormone therapy without chemotherapy (68.4%). Compared with controls, survivors had significantly lower unadjusted baseline physical activity and perceived cognition at 36 months (survivors: 1,284.92 MET min/week at baseline, 59.05 PCI score at 36 months; controls: 2,085.98 MET min/week at baseline, 62.22 PCI score at 36 months; p < .05; Supplementary Material 2).

Table 1.

Characteristics of Older Breast Cancer Survivors Prior to Systemic Therapy and Contemporaneously Assessed, Frequency-Matched Noncancer Controls

Characteristic	Survivors (N = 395)	Controls (N = 374)	t or χ2 (df)	p
	n (%)a	n (%)a
Age, mean (SD)	68.0 (5.9)	67.9 (7.0)	−0.20 (767)	.84
Years of education, mean (SD)	15.2 (2.2)	15.4 (2.3)	1.76 (765)	.24
Race
Black or African American	32 (8.1)	33 (8.8)	3.79 (5)	.58
White	326 (82.5)	315 (84.2)
American Indian or Alaska Native	4 (1.0)	4 (1.1)
Asian	19 (4.8)	10 (2.7)
Multiracial or other	14 (3.5)	11 (2.9)
Ethnicity
Hispanic	23 (7.2)	26 (8.0)	0.59 (1)	.44
Non-Hispanic	297 (92.8)	297 (92.0)
Insuranceb
Private insurance	324 (82.0)	276 (73.8)	30.98 (4)	<.01
Medicare	16 (4.1)	49 (13.1)
Medicaid	7 (1.8)	10 (2.7)
Uninsured	16 (4.1)	26 (7.0)
AJCCc V6 stage
0 (DCISd)–I	266 (67.3)	—
II–III	129 (32.7)
Systemic treatment
Chemotherapy ± hormone therapy	110 (27.8)	—
Hormone only	270 (68.4)
None	15 (3.8)
Local treatment
Lumpectomy with radiation	175 (44.3)	—
Lumpectomy only	51 (12.9)
Mastectomy	166 (42.0)
None	3 (0.76)
Hormone treatmentb
Aromatase inhibitor	264 (66.8)	—
Tamoxifen	26 (6.6)
None	62 (15.7)
Chemotherapy regimenb
ACe	6 (1.5)	—
ACe + Tf	43 (10.9)
CMFg	12 (3.0)
Tf	33 (8.4)
None	285 (72.2)

^aUnless otherwise specified.

^bDoes not add to 100% due to missing data.

^cAmerican Joint Commission on Cancer.

^dDuctal Carcinoma In Situ.

^eAdriamycin and cyclophosphamide.

^fTaxol.

^gCyclophosphamide, methotrexate, and fluorouracil.

Preliminary Analyses

Participants who completed one or more follow-up assessments were compared with those who refused or were lost to follow-up (Supplementary Materials 3 and 4). Survivors who refused or were lost to follow-up at 12 months had lower baseline APE than survivors who at least partially completed follow-up at 12 months (p = .048). Compared with completers, lower education was also found for survivors who refused or were lost to follow-up at 24 (p = .01) and 36 months (p = .02) and controls who refused or were lost to follow-up at 12 (p = .003) and 24 months (p = .04).

Cognition and Physical Activity: Survivors

Survivors had better APE performance when they also reported higher concurrent physical activity, especially at 12 months (Fig. 2a; within-person association at 12 months: β = 0.24, p = .04; within-person association at 36 months: β = 0.24, p = .07). However, having higher physical activity did not predict later APE domain scores, and APE scores did not predict later physical activity (i.e., cross-lagged within-person effects). There also was not an overall association between APE and physical activity at the between-person level. When survivors had better perceived cognition (Fig. 2b), they reported more concurrent physical activity at several time points (within-person cross-sectional associations at 12 months [β = 0.25, p = .03] and 24 months [β = 0.25, p = .002]; nonsignificant trend at 36 months [β = 0.21, p = .06]), but having higher physical activity did not predict later perceived cognition, and perceived cognition did not predict later physical activity (i.e., cross-lagged within-person effects). There was not an overall association between physical activity and perceived cognition at the between-person level.

Fig. 2.

Results from random intercepts cross-lagged panel model among survivors, controlling for age, years of education, and study site. cAPE latent centered attention, processing speed, and executive function; cPCI latent centered perceived cognitive impairment; cPhysical activity latent centered physical activity; RI random intercepts. Parameter estimates are standardized. **p < .01, *p < .05, ^†p < .10. (a) Longitudinal relationship between APE cognitive performance and physical activity among older breast cancer survivors. Model fit indices: χ² = 33.12, p = .65, RMSEA = 0.00, CFI = 1.00, TLI = 1.01. (b) Longitudinal relationship between PCI (where higher scores indicate better perceived cognitive function) and physical activity among older breast cancer survivors. Model fit indices: χ² = 53.23, p = .04, RMSEA = 0.03, CFI = 0.97, TLI = 0.94. APE attention, processing speed, and executive function; CFI comparative fit index; PCI Perceived Cognitive Impairments; RMSEA root mean square error of approximation; TLI Tucker Lewis index.

Survivors that were more physically active than their previous average tended to remain active over time (β = 0.17, p = .09, from baseline to 12 months in model with APE and β = −0.18, p = .07 from baseline to 12 months in model with PCI, β = 0.27, p = .02 from 24 to 36 months in model with APE and β = 0.38, p < .001 in model with PCI) and those with higher perceived cognition at 12 months maintained higher perceived cognition at 24 months (β = 0.27, p = .01).

Cognition and Physical Activity: Noncancer Controls

Physical activity and objective or perceived cognition levels in noncancer controls were not related to each other when examining cross-sectional and lagged relationships at the within-person level (Fig. 3a and b). Physical activity and objective cognition were also not associated at the between-person level. However, perceived cognition and physical activity were associated at the between-person level (β = 0.25, p = .01), suggesting that controls with better perceived cognition than the average control also had higher physical activity across time points.

Fig. 3.

Results from random intercepts cross-lagged panel model among controls, controlling for age, years of education, and study site. Parameter estimates are standardized. **p < .01, *p < .05, ^†p < .10. (a) Longitudinal relationship between APE cognitive performance and physical activity among noncancer controls. Model fit indices: χ² = 74.80, p < .001, RMSEA = 0.06, CFI = 0.96, TLI = 0.92. (b) Longitudinal relationship between PCI (where higher scores indicate better perceived cognitive function) and physical activity among noncancer controls. Model fit indices: χ² = 72.66, p < .001, RMSEA = 0.06, CFI = 0.95, TLI = 0.88. APE attention, processing speed, and executive function; CFI comparative fit index; PCI Perceived Cognitive Impairments; RMSEA root mean square error of approximation; TLI Tucker Lewis index.

Controls that were more physically active than their previous average tended to remain active over time (β = 0.29, p = .02 from 12 to 24 months in model with APE and β = 0.26, p = .06 in model with PCI, β = 0.56–0.57, p < .001 from 24 to 36 months in both models), and controls with higher perceived cognition than their average at 24 months maintained higher perceived cognition at 36 months (β = 0.53, p < .001).

Secondary Analyses

Results from linear mixed models (see Supplementary Materials 5–8) showed the only group differences (survivors who received chemotherapy or hormone treatment alone compared with controls) were for baseline physical activity, with both survivor groups reporting lower physical activity than controls (chemo group: B = −0.81, p < .001; hormone only group: B = −0.88, p < .001). The only significant time effects were for APE, with improvement at 12 (B = 0.08, p = .001), 24 (B = 0.12, p < .001), and 36 months (B = 0.09, p = .01) compared with baseline. The only significant group by time interactions were for PCI. Survivors who received chemotherapy had a greater decrease in perceived cognition from baseline to 12 months than controls (B = −2.94, p = .01). Survivors who received hormone treatment alone had a greater decrease in perceived cognition from baseline to 24 months than controls (B = −1.67, p = .049). Both survivor groups showed greater decline in perceived cognition from baseline to 36 months than controls (chemo: B = −2.97, p = .045, hormone: B = −3.21, p = .004).

RI-CLPM were also completed for each of the six APE tests (see Supplementary Materials 9–21 for results and interpretation). Briefly, the only models with significant associations between cognition and physical activity among survivors were those for Digits Backward and Trails B. There were positive cross-sectional and cross-lagged associations (within-person effects) between Digits Backward and physical activity (β = 0.21, p = .01 at baseline; β = 0.28, p = .03 at 12 months; β = 0.28, p = .01 from baseline Digits Backward to 12 months physical activity; β = 0.20, p = .05 from 12 months physical activity to 24 months Digits Backward); however, there was a negative overall association (between-person effect; β = −0.35, p = .002). There were negative cross-sectional and cross-lagged associations (within-person effects) between Trails B and physical activity (β = −0.24, p = .02 at 36 months; β = −0.32, p = .04 from 24 months physical activity to 36 months Trails B); however, there was a trend for a positive overall association (between-person effect; β = 0.25, p = .08). Among controls, the only models with significant associations between cognition and physical activity were those for Digits Backward, Digit Symbol, and COWAT. Among controls, all associations were in expected directions.

Discussion

This study is among the first to examine the cross-lagged relationships between physical activity and cognition in older breast cancer survivors and concurrently assessed matched noncancer controls. We did not find evidence for lagged relationships between cognitive function and physical activity over time. However, we did find evidence for cross-sectional relationships with small effect sizes between cognitive function and physical activity such that when survivors were more active than their usual, they had better perceived and objective cognition than at times when they were less physically active. This relationship was not seen in controls. In contrast, there was an overall relationship with a small effect size between cognition and physical activity in controls, but only for perceived cognition; controls with better perceived cognition than the average control also had higher concurrent physical activity than the average control across time points.

Few studies with older adults or cancer survivors have prospectively examined the relationship between cognitive function and physical activity up to 36 months without testing an intervention [30, 55]. This descriptive research is important for understanding how cognition and physical activity are associated within the context of the natural changes in the years following cancer diagnosis and treatment. One recent study in middle-aged breast cancer patients from pre- to post-treatment and 6 months post-treatment found that greater moderate-to-vigorous physical activity was associated with better perceived cognition and performance on tests of attention and visual memory over time [56]; however, this study did not test for potential effects of cognition on physical activity. To our knowledge, no prior studies have conducted this descriptive research in older cancer survivors prospectively beginning at pretreatment.

The lack of evidence for a lagged effect of cognitive function on physical activity is counter to our hypothesis based on self-regulation theory and prior research in older adults [29–31]. According to self-regulation theories, deficits in higher-order cognitive functions may interfere with self-regulatory behaviors (e.g., goal setting, planning) that facilitate physical activity behavior [29]. Inconsistencies between present findings and theory might be due to differences in definitions and measurement of higher-order cognitive functions. In the TLC study, we included APE in one domain, while others have included a composite executive function score from tests of verbal fluency and selective attention [30] or have used structural equation models with multiple tests of executive function [31]. Survivors often had high self-reported activity levels and experienced subtle cognitive deficits, which may have made it difficult to observe differences. Additionally, participants were not asked to engage in the effortful process of changing their physical activity, and cognitive function may be most important for making a change to their usual habits.

Interestingly, there were more consistent small cross-sectional effects in the relationship between perceived versus objective cognition and physical activity among survivors. Discrepancy in the association between cognitive function and physical activity based on the measurement method might be due to insufficient sensitivity of neuropsychological tests to detect subtle deficits in cognitive function. Some have argued that self-report is a more appropriate measurement of subtle—yet still distressing—cancer-related cognitive impairment [36].

There were some differences in patterns for survivors and controls, where survivors showed within- but not between-person level associations of cognitive function and physical activity, and controls showed a between-but not within-person level cross-sectional association of perceived cognition and physical activity. This suggests that survivors experienced fluctuations in their activity and cognition over time while controls experienced stable individual differences. This pattern is likely related to having cancer and undergoing treatment. The first year after a cancer diagnosis involves several adjustments for survivors that have the potential to impact their physical activity. For example, receiving a cancer diagnosis and undergoing surgery and chemotherapy all appear to reduce physical activity [56–58].

Secondary analyses with separate models for each objective cognition measure found some support for a bidirectional relationship between cognition and physical activity, but this was complicated by conflicting negative associations among survivors. The individual tests where significant associations were found include Digits Backward (both models for survivors and controls), Trails B (survivors only), Digit Symbol (controls only), and COWAT (controls only). These tests require greater executive functioning demands (i.e., working memory, task switching, incidental learning, and cognitive flexibility respectively); thus, they may be more sensitive for detecting executive cognitive performance. Conflicting between- and within-person associations were found in the models using Digits Backward and Trails B performance among survivors. These findings may be explained by differences among the associations between physical activity and cognitive functions of working memory (measured by Digits Backward) and task switching (measured by Trails B), or differential associations depending on time elapsed since baseline (i.e., presystemic treatment). However, these secondary findings should be interpreted with caution because relying on individual cognitive tests does not capture the diversity of executive functions [44]. The parent study was also not designed to examine effects of specific tests and examining individual tests may lead to false discovery due to multiple testing. Furthermore, the International Cognition and Cancer Task Force recommends a battery of cognitive assessments to ensure coverage of the cognitive domains most affected by cancer and its treatment [42].

Our results have implications for future research and survivorship care. Future research might examine lagged relationships between cognition and physical activity within a shorter timeframe than 1 year. Physical activity interventions may examine the potential bidirectional effect between cognition and physical activity when survivors are undergoing the effortful process of attempting to change their physical activity. Future descriptive observational or interventional research could also benefit from intensive repeated measures such as ecological momentary assessment and accelerometry. This would provide an avenue for investigating the temporal relationships between shorter-term variability in cognitive function and physical activity. Biopsychosocial mechanisms involved in the relationship between cognition and physical activity should also be examined. Proposed physiological mechanisms for the effect of physical activity on survivors’ cognitive function include reduced systemic inflammation and greater neurogenesis facilitated by increased expression of neurotrophic and neuroprotective proteins (e.g., brain-derived neurotrophic factor, vascular endothelial growth factor, and insulin like growth factor) [59]. Our evidence for a cross-sectional relationship—when used in combination with evidence from physical activity interventions [18–21]—suggests that maintaining physical activity is an important component of survivorship care for supporting cognitive and physical health for older breast cancer survivors.

A key strength of our project was using data from the largest prospective controlled study of cognitive performance in older survivors starting before systemic cancer treatment. Focusing on survivors aged 60 years and older was also a strength because this large, understudied group may be at risk for cognitive symptoms due to potential accelerated aging from cancer and treatment [12]. The sample was recruited from diverse regions of the USA and its racial and ethnic diversity is largely representative of older breast cancer survivors in the USA [60]. Additionally, measures included objective cognitive performance and validated self-report questionnaires. Several limitations should also be noted. First, self-reported physical activity is susceptible to overestimation from social desirability and inaccurate recall, and the IPAQ-SF does not correlate well with objective measures of physical activity [61]. Potential bias from self-reported physical activity may have reduced power to detect effects, particularly if there were differences in the accuracy of self-report by level of cognitive functioning. Use of self-reported and objective measures are recommended for future research, although persons experiencing cognitive difficulties may have difficulty remembering to wear a device to monitor activity. Second, the study sample was generally well-educated, insured (87.9% of survivors and 89.6% of controls were covered through private insurance, Medicare, or Medicaid), high functioning, and cognitively intact at baseline and primarily recruited from academic medical centers. Adherence to national guidelines for physical activity was also greater in the current sample (56.6% in survivors and 76.8% in controls at baseline) than in prior research with older adults and older cancer survivors [23, 25]. Our well-resourced, high functioning, and highly active sample may limit generalizability of our study findings and contribute to a possible ceiling effect for physical activity. However, it is notable that we found significant associations between physical activity and cognitive function in a sample that may underestimate population detriments to both cognitive function and physical activity. Third, study attrition occurred, and those who completed follow-ups tended to have higher education than those who refused or were lost to follow-up; however, years of education was a covariate, and the statistical approach used all available data. Fourth, unmeasured variables such as the number of chemotherapy cycles and treatment length have the potential to impact both physical activity and cognition; thus, these may be important covariates in future research. Finally, the models did not have sufficient power to detect small effects, and interactions with factors such as treatment type, disease stage, or comorbidities were not tested.

In conclusion, we found evidence for cross-sectional relationships between physical activity and both objective and subjective cognitive function in the first 3 years following treatment for breast cancer. These associations were found for older breast cancer survivors at the within-person level, but only at the between-person level for matched controls. Our findings make a unique contribution to the literature by focusing on the vulnerable population of older survivors, and by examining the bidirectional relationship beginning before cancer treatment and following for 3 years. Accounting for cross-lagged associations between cognition and physical activity was a necessary step because theory and prior research in older adults indicate that there may be bidirectional effects. There are many potential directions for future research, particularly for testing the potential bidirectional relationship in the context of a physical activity intervention, more intensive repeated measures to determine the temporal relationship between cognitive function and physical activity, and community-based sampling. Considering the potential impact of cognitive function when attempts are made to change physical activity can inform future interventions to improve cognitive and physical health for older cancer survivors.

Funding

This project was supported by National Cancer Institute grant F31CA220964, and D.B.T. is also supported by the National Cancer Institute grant T32CA090314. The TLC study was supported by National Cancer Institute grants R01CA129769, R35CA197289, and R01AG068193 to J.M. This study was also supported in part by the National Cancer Institute at the National Institutes of Health grant P30CA51008 to Georgetown-Lombardi Comprehensive Cancer Center for support of the Biostatistics and Bioinformatics Resource and the Non-Therapeutic Shared Resource. The work of A.J.S. was supported in part by National Institute of Aging and National Institutes of Health grants P30AG010133, P30AG072976, R01AG019771, R01AG057739, U01AG024904, R01LM013463, R01AG068193, T32AG071444, U01AG068057, and U01AG072177. B.C.M. was supported in part by the National Institute of Aging and National Cancer Institute grants P30AG10133, R01AG19771, and R01CA244673. T.A.A. was supported in part by National Cancer Institute at the National Institutes of Health grants R01CA172119 and P30CA008748. S.K.P. was supported in part by American Cancer Society Research Scholars Grant RSG-17-023-01-CPPB.

Open Science Framework awards

Preregistered Analysis Plan The parent study was preregistered at www.clinicaltrails.gov (study identifier: NCT03451383). The analysis plan for the present project was not formally preregistered. Deidentified data from this study are not available in a public archive. Deidentified data from this study will be made available (as allowable according to institutional IRB standards) by emailing the corresponding author. Analytic code used to conduct the analyses presented in this study are not available in a public archive. They may be available by emailing the corresponding author. Materials used to conduct the study are not publicly available.

Compliance with Ethical Standards

Authors’ Statement of Conflict of Interest and Adherence to Ethical Standards H.S.L.J. is a paid consultant for RedHill BioPharma, Janssen Scientific Affairs, and Merck, and has received grant funding from Kite Pharma. A.J.S. has also received support from Avid Radiopharmaceuticals, Bayer Oncology, Siemens Medical Solutions USA, Inc., and Springer-Nature Publishing. The other authors have no conflicts of interest to disclose.

Authors’ Contributions Danielle B. Tometich served as lead for conceptualization, formal analysis, visualization, writing—original draft, and writing—review and editing. Catherine E. Mosher, Melissa Cyders, and Brenna C. McDonald contributed to supervision, conceptualization, and writing—review and editing. Brent J. Small contributed to writing—original draft. Brent J. Small, Wanting Zhai, and Xingtao Zhou contributed to data curation, formal analysis, visualization, and writing—review and editing. Heather S.L. Jim, Paul Jacobsen, Tim A. Ahles, James Root, Deena Graham, and Sunita K. Patel served in a supporting role to funding acquisition, investigation, methodology, project administration, and writing—review and editing. Jeanne Mandelblatt served as lead for funding acquisition, project administration, resources, and supervision, and contributed to writing—original draft, and writing—review and editing.

Ethical Approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Informed Consent Informed consent was obtained from all participants included in the study.