The number of breast cancer survivors in the United States is expected to rise dramatically with its aging population and increased rates of breast cancer survival. Cognitive impairment is a disruptive and persistent condition that is common among breast cancer survivors [1]. Breast cancer survivors often show deficits on objective neurocognitive measures and self-reported measures of cognition, which assess different, yet important, aspects of cognition [2]. Cognitive difficulties can impact quality of life and ability to return to work [3]. Therefore, identifying effective interventions to improve cognition is a research priority in cancer survivorship [4].

A potential intervention is physical activity, recommended by the National Comprehensive Cancer Network for cancer-related cognitive dysfunction [4]. Breast cancer survivors often reduce physical activity after treatment and have low levels of physical activity [5]. The low activity levels in breast cancer survivors and the benefits of physical activity for cognition among noncancer populations [6] suggest that physical activity could be an important target for breast cancer survivors. Recent trials, including our 12-week intervention in 87 breast cancer survivors [7], have shown that increasing physical activity improves objective and self-reported cognition in cancer survivors [7-9]. As the intervention literature continues to grow, there is a need to understand the mechanisms driving the effects of physical activity on cognition [9].

A hypothesized mechanism linking physical activity with cognition is improvements in physical and psychological function. Problems with self-reported cognition have consistently been associated with poorer physical function, anxiety, depression, and fatigue, which are often elevated or impaired in cancer survivors [10-12]. Physical activity in cancer survivors can improve all these aspects of physical and psychological function [13-15]; therefore, physical and psychological function may mediate the relationship between physical activity and self-reported cognition. A longitudinal observational study in 1477 breast cancer survivors found that distress and fatigue were on the pathway between physical activity and self-reported cognition [16]. Physical and psychological function factors may also be associated with objective neurocognitive function. Prospective studies in healthy older adults have shown that poor physical function can predict future cognitive impairment [17-19]. A cross-sectional study in 299 breast cancer survivors found that physical activity was related to improved executive function and working memory and that the effect of physical activity on cognition was partially explained by physical activity’s influence on fatigue [20]. To our knowledge, no randomized controlled trials have explored aspects of physical and psychological function as mediators of the effect of physical activity on cognition in cancer survivors. Given the high prevalence of impaired physical and psychological function in breast cancer survivors and the strong connection between cognition and physical activity, improvements in physical and psychological function are plausible mechanisms through which physical activity may improve cognition [21].

Although cognitive and physical function have been shown to be related among cancer survivors [22,23], the direction of this relationship has not been studied. Evidence from cohort studies and randomized controlled trials in older adults suggests that cognitive impairments may precede physical decline [24,25]. Therefore, improvements in cognition may mediate the effects of physical activity on improved physical function.

Physical activity is thought to positively influence cognition via several biological mechanisms including increased brain-derived neurotrophic factor (BDNF), improved metabolic function, and reduced systemic inflammation. BDNF is a biomarker of brain health. BDNF levels are correlated with processing speed [26] and are significantly elevated after aerobic physical activity [27]. In noncancer populations, BDNF mediates the effects of physical activity on neurocognitive outcomes [28] and is positively related to objective [29] and self-reported [30] cognition in cancer patients. BDNF is regulated by energy balance, insulin, and inflammatory cytokines; thus, it is likely part of a central mechanism through which physical activity integrates with elements of metabolism to impact cognition. In fact, measures of metabolic dysfunction, for example, insulin resistance measured by the homeostatic model assessment of insulin resistance (homeostatic model assessment of insulin resistance [HOMA-IR] or homeostatic model assessment 2 of insulin resistance [HOMA2-IR], calculated using fasting plasma insulin and glucose) [31,32], and systemic inflammation, for example, levels of C-reactive protein (CRP), are inversely associated with physical activity [33,34] and cognitive performance [33,35]. Insulin resistance/HOMA-IR is associated with cognitive decline in diabetic [36] and nondiabetic populations [37,38]. Routine physical activity is anti-inflammatory [10] and improves insulin sensitivity [39]. Physical activity interventions in cancer survivors result in beneficial effects on inflammation and insulin pathway biomarkers [39]. Interestingly, exercise frequency and an anti-inflammatory genotype each predicted better cognitive performance in a sample of breast cancer survivors [40]. Other accumulating evidence supports a direct link between inflammation and cognition in breast cancer survivors. In breast cancer patients studied before and up to 2 years post treatment, CRP levels were inversely correlated with cognition [41]. Breast cancer survivors have elevated systemic inflammation that is prevalent for decades post treatment and that is associated with cognitive decline [42,43]. In noncancer populations, CRP and insulin resistance have been associated with neurocognitive impairment [44]. Therefore, BDNF, systematic inflammation, and insulin resistance are potential putative mechanisms that could causally link physical activity with cognition in cancer survivors.

The secondary analysis (this study) examined physical and psychological function and biological mechanisms potentially linking physical activity with changes in objective and self-reported cognition among breast cancer survivors enrolled in a 12-week randomized controlled trial. In the primary study of 87 breast cancer survivors, the exercise arm had greater improvements in objectively measured processing speed and self-reported cognitive abilities than the control arm [7]. The primary aim of this analysis was to explore whether proposed physical and psychological function measures (depression, anxiety, fatigue, and physical functioning) and mechanistic biomarkers (BDNF, HOMA2-IR, and CRP) mediated intervention effects on cognition. For the mediation analyses, the following steps were taken: (1) intervention effects on potential mediators (physical/psychological function/biomarkers) were examined, (2) associations between potential mediators (physical /psychological function/ biomarkers) with objectively measured processing speed and self-reported cognitive abilities were examined, and (3) if changes in potential mediators (physical/psychological function/biomarkers) mediated the relationship between the intervention effect on objectively measured processing speed and self-reported cognitive abilities was tested. We a priori hypothesized that, compared with breast cancer survivors in the control arm, participants in the exercise arm would have improvements in physical and psychological function and biomarkers, and that these factors would mediate the relationship between physical activity and cognitive outcomes. To address uncertainties about the directionality of the relationship between physical function and cognition, we also conducted exploratory analyses assessing self-reported cognitive abilities and objectively measured processing speed as mediators of the relationship between physical activity and physical function.

A total of 87 breast cancer survivors were enrolled in a randomized controlled trial of a 12-week physical activity intervention [21]. Data were collected from February 2015 to July 2016. The study was approved by the University of California (UC) San Diego Institutional Review Board (protocol number 140694), and all participants provided written informed consent. The trial was registered on ClinicalTrials.gov (NCT02332876).

Eligible participants were female breast cancer survivors who were aged 21 to 85 years, who completed breast cancer surgery less than 5 years ago, and who completed chemotherapy or radiation treatment. Other inclusion criteria included the following: self-reporting less than 60 min of moderate-to-vigorous physical activity (MVPA) in 10-min bouts per week; self-reported fogginess or worsening of memory, thinking, or concentration; and internet access. Exclusion criteria included the following: any medical condition that could preclude safe participation in an unsupervised physical activity intervention and other primary or recurrent invasive cancer within the last 10 years.

Detailed information on study procedures and the intervention have been published [21]. Briefly, women were primarily recruited through cancer registries. Potential participants were phone-screened for eligibility and then scheduled for an in-person visit at UC San Diego Moores Cancer Center, during which participants completed questionnaires, a fasting blood draw, and neurocognitive tests. Height and weight were measured. Next, participants received an ActiGraph GT3X+ accelerometer (ActiGraph, LLC.) to wear for 7 days and return at the randomization visit. At the randomization visit, 87 breast cancer survivors were randomized in a 1:1 allocation ratio to receive either a 12-week physical activity intervention (exercise arm, n=43) or waitlist wellness contact control condition (control arm, n=44). All baseline measures were repeated at 12 weeks. A participant from each arm was lost to follow-up, resulting in 98% retention (exercise, 42/43 and control, 43/44) [7].

Unless otherwise specified, all analyses were conducted using SAS version 9.4 (SAS Institute, Inc). All analyses were performed using an intent-to-treat principle. Longitudinal random-effects models were developed. This method uses all available data, does not omit subjects with missing data, and provides unbiased results provided the data conform to a missing at random missing data mechanism. Group differences in baseline characteristics were assessed using t tests and chi-square tests. Mixed-effects regression models with a subject-level random intercept and an unstructured covariance structure, as determined by model Akaike Information Criterion comparisons in the main effects models, were used for all other models. Models assessing intervention effects on repeated measures (at baseline and 12 weeks) of anxiety, depression, fatigue, physical function, and biomarkers (BDNF, HOMA2-IR, and CRP) included fixed effect terms for group, time point (baseline or 12 weeks), and time-by-group interaction. Contrasts were used to calculate the difference of change based on the regression model when examining group differences. Each outcome was assessed in a separate model. All biomarkers were log transformed before analyses to correct for right-skewed residuals.

Assuming a 2-sided test with an alpha of .05 and a sample size of 80 with 1:1 randomization to treatment or control, there was 80% power to detect a main effect of 0.32.

Participants (n=87) were, on average, aged 57 (SD 10.4) years and 30 (SD 16.7) months post breast cancer surgery. Majority of the participants were diagnosed with stage 1 disease (61%, 53/87), received chemotherapy (53%, 46/87), and were taking an aromatase inhibitor or tamoxifen (70%, 61/87) at the time of study enrollment. Mean T-scores on the physical and psychological function measures ranged from a mean of 50.4 (SD 7.98) for depression to a mean of 55.2 (SD 7.72) for anxiety across the study arms. See Table 1 for baseline characteristics stratified by study arm; there were no significant differences between the exercise and control arms in baseline characteristics (P>.05).

The overall effects of the intervention on physical activity and cognitive outcomes have been published [7]. Briefly, the exercise arm had significantly greater increases in accelerometer-measured MVPA (mean min/day increase: 14.2 vs −0.7; beta=7.24; P<.001) than the control arm from baseline to 12 weeks. Participants in the exercise arm showed significantly greater improvements in objectively measured processing speed than those in the control arm (mean increase score 7.0, SD 10.2 vs 3.0, SD 8.2; beta=2.01; P<.05). The between-group difference in self-reported cognitive abilities (beta=.92; P=.09) was not statistically significant, but the magnitude of the change within the exercise arm was suggestive of clinically meaningful improvement (average 2.7-point improvement from baseline to 12 weeks) [46].

Changes in physical and psychological function measures from baseline to 12 weeks in the exercise versus control arms are presented in Figure 1. The exercise arm had significantly greater improvements in physical functioning than the control arm (beta=1.23; 95% CI −0.42 to 0.03; P=.049). The exercise arm also showed significantly greater reductions in anxiety than the control arm (beta=−1.50; 95% CI −0.07 to −2.94; P=.04). Furthermore, changes in each of physical functioning and anxiety were significantly associated with the change in cognition (Figures 2 and 3 and Multimedia Appendix 1). Therefore, conditions were met to test for the potential mediation effects of anxiety and physical functioning. Results indicate that anxiety significantly mediated the intervention effect on cognitive abilities (see Figure 2). Differences between the exercise and control arms in changes in cognitive abilities were, in part, because of greater decreases in anxiety among intervention participants compared with those in the control group (bootstrap 95% CI −1.96 to −0.06). Physical functioning did not mediate the intervention effect on cognitive abilities (bootstrap 95% CI −1.12 to 0.10). Neither anxiety (bootstrap 95% CI −1.18 to 0.74) nor physical functioning (bootstrap 95% CI −2.34 to 0.15) mediated the effect of the intervention on processing speed (see Figure 3). Exploratory analyses found that neither cognitive abilities (bootstrap 95% CI −0.06 to 0.68) nor processing speed (bootstrap 95% CI −0.15 to 0.63) mediated the intervention effect on physical function.

The exercise arm had greater reductions in CRP than the control arm (beta=.253; 95% CI −0.04 to 0.57; P=.09; see path a in Figures 2 and 3 and Multimedia Appendix 1). Change in CRP was not associated with either measure of cognition (see path b in Figures 2 and 3 and Multimedia Appendix 1); therefore, mediation could not be tested. There were no between-group differences in changes in BDNF (beta=.092; 95% CI −0.25 to 0.43; P=.59) or HOMA2-IR (beta=.05; 95% CI −0.12 to 0.22; P=.55); therefore, conditions were not met to test for mediation, and no further analyses were conducted.

This is the first known published randomized trial with breast cancer survivors to explore potential physical and psychological function and biological mechanisms underlying the effect of a physical activity intervention on objective and self-reported cognition. Results indicated that reductions in anxiety may have contributed to improvements in self-reported cognitive abilities, but there was no evidence of mediation for objectively measured processing speed or other measures of physical or psychological function. In addition, there was no evidence that the proposed biological mechanisms mediated the intervention effect on objective or self-reported cognition.

Findings from this pilot study support the importance of anxiety for self-reported cognition and the potential for physical activity interventions to reduce anxiety in breast cancer survivors. Although previous physical activity intervention trials with cancer survivors that have examined physical or psychological function factors as possible mechanisms by which physical activity impacts cognition could not be identified, this finding is consistent with the published literature suggesting that self-reported cognition is likely a psychosocial aspect of cancer and indicator of psychological distress [60,61]. Self-reported cognition is consistently associated with other dimensions of psychological functioning such as depression and anxiety [2,61,62] that are elevated in breast cancer survivors and have been shown to improve with physical activity [14,63]. These results are also consistent with the notion that self-reported cognition is a distinct construct from objective neurocognitive function [2,64] and that there may be different mechanisms of the effects of physical activity on self-reported and objectively measured cognition.

Contrary to expectations, anxiety was the only physical or psychological function factor that was a significant mediator of the intervention effects on cognition. This finding may be because of the general lack of problems with physical and psychological function in our sample at baseline. It must be noted that, among all physical and psychological function factors examined, anxiety had the highest average scores at baseline (indicating higher levels of anxiety) and was the only variable to achieve a clinically meaningful change, which has been defined as a 3-point difference in scores for these measures [46]. Exploratory analyses to better understand the direction of the relationship between physical function and cognition were also nonsignificant. Research from the literature on aging suggests that both cognitive impairments precede physical declines [24,25], and physical declines precede cognitive declines [17,19]. However, this sample did not report impaired physical function at baseline, which might have limited the potential for physical functioning to mediate the relationship between physical activity and cognition or for cognition to mediate improvements in physical function. Fully powered trials, and trials targeting women with lower physical functioning, should be conducted to determine whether physical activity–associated improvements in day-to-day physical functioning are part of the causal pathway through which exercise improves cognition. Overall, this study provides preliminary evidence that in breast cancer survivors, physical activity interventions may improve self-reported cognition by reducing anxiety. Future studies with longer time frames and more distressed participants may be able to elicit clinically meaningful benefits to other aspects of physical and psychological functioning. Future research should also explore other aspects of physical and psychological function as potential mechanisms underlying the impact of physical activity interventions on cognition.

Although the intervention showed some reduction in systemic inflammation, as measured by CRP, this was not associated with cognition. Contrary to expectations, the intervention did not improve BDNF or HOMA2-IR. These findings may be because of having a sample with a normal to overweight BMI and, on average, low HOMA2-IR, small sample size, and the short intervention duration. Previous larger physical activity randomized controlled trials with breast cancer survivors have found that the effects of physical activity on the insulin pathway are stronger among obese survivors [39]. BDNF has been understudied in breast cancer survivors; however, a review of randomized controlled trials of physical activity interventions in noncancer populations found improvements in BDNF in 3 out of 5 trials, with 2 of the trials being longer than 6 months [27]. Findings from this study suggest that among breast cancer survivors with a generally healthy BMI, neurotrophic factors, inflammation, or insulin pathways may not be the mechanisms for changes in cognition in a physical activity intervention. However, fully powered trials are needed to confirm these results.

As the research testing the impact of physical activity on cancer-related cognitive impairments grows, it is important to continue exploring the mechanisms that causally link physical activity and cognition. Future studies should consider factors related to biological and cellular aging. It has been hypothesized that MVPA can slow, or even reverse, cellular aging and thereby improve cognitive performance at the cellular level. This association is particularly relevant for cancer survivors who may experience accelerated aging caused by chemotherapy and psychological stress [1,65]. Other potential mechanisms may be related to the impact of physical activity on brain structure and function. Neuroimaging studies in breast cancer survivors have found structural changes including reduction in gray matter volume following chemotherapy, which has been associated with impairments in processing speed [66,67]. Although the direct impact of increasing physical activity on brain volume has not been tested, there is some evidence that hippocampal volumes are larger in breast cancer survivors with high cardiovascular fitness levels [68]. Physical activity also has the potential to increase cerebral blood flow, which may slow neurodegeneration [69]. Exploring whether the rate of cellular aging can be slowed by physical activity and whether physical activity can improve brain structure and function are important directions for future research with cancer survivors.

The small sample size may not have provided sufficient power to detect significant mediation, and only bivariate mediation possibilities were considered. Subsequent studies with larger samples should consider multiple mediating mechanisms. Nonetheless, information gathered from these analyses can help generate new hypotheses about mechanisms and guide future studies to improve cognitive function in breast cancer survivors. The short duration and relative physical and mental health of the study population may also have limited our ability to detect mediation in the chosen mechanisms. Future studies with longer durations and more diverse populations that incorporate novel measures of cellular aging and neuroimaging are needed to continue advancing the field. Owing to the pilot nature of this study with limited power, we also did not adjust for multiple comparisons. In addition, these results are limited to a single measure of processing speed, and the findings cannot be generalized to other processing speed measures or aspects of cognition, such as memory, attention, and executive function, which are often impacted by cancer and its treatments [70]. Another limitation is that the PROMIS measures may not have been sufficiently sensitive to the magnitude of change achieved in this study. A longitudinal study of almost 3000 cancer survivors demonstrated that effect sizes of 0.24 to 0.71 were needed to detect declines and improvements across different PROMIS measures [47].

Despite these limitations, the study had many strengths including the randomized design and objective measurement of physical activity. In addition, this study uniquely focused on physical and psychological function and biological mechanisms of action across both objective and self-report measures of cognition. Furthermore, bootstrap methods to assess mediation were used, which is recommended in small to moderate samples [71]. This research extends what is known about the impact of physical activity on cognitive function to examine mechanisms of change. Identifying underlying mechanisms is critical for determining modifiable intervention targets and enhancing intervention efficacy.

Results of this novel study provide preliminary evidence that an intervention that increases physical activity may improve self-reported cognition by decreasing anxiety. If supported by larger studies, recommending increasing physical activity may be an effective strategy for reducing anxiety and improving self-reported cognition among cancer survivors. Continued research in this area is needed to determine mechanisms of change for objectively measured cognition.