Mediators of Exercise Effects on Fatigue in Cancer Survivors
Mediators of Exercise Effects on Fatigue in Cancer Survivors
Although not statistically significant, the direction and magnitude of the effect sizes related to our exercise intervention varied depending on the fatigue measure used and aspect assessed. A nonsignificant small to medium effect size increase was noted for fatigue intensity, with a nonsignificant small to medium effect size decrease noted for fatigue interference and PROMIS® fatigue. Only PROMIS® fatigue showed a within-group statistically significant decline in the intervention group. Study power was limited by our small sample size, which resulted from the budgetary and logistical constraints of a pilot study. Nevertheless, our effect size reductions in fatigue interference and general fatigue are consistent with that of previous studies, with our results suggesting an important finding relative to a possible increase in intermittent fatigue intensity. It is noteworthy that our data suggest important mediating relationships warranting further study. Specifically, inflammation, sleep quality, psychosocial factors (i.e., exercise social support and enjoyment), and minutes of weekly physical activity mediated the effects of our exercise intervention on fatigue in breast cancer survivors. Complex relationships that include both mediation and strengthening of the intervention–fatigue relationship exist. These relationships vary among the fatigue measures used, and further study is needed to improve our understanding of these relationships. The inflammatory mediators of fatigue response to exercise may be due to beneficial reductions in the pro–anti-inflammatory ratios resulting from the dynamic response of the cytokine system to pro-inflammatory aspects of exercise training.
The National Comprehensive Cancer Network (NCCN) guidelines (a national resource used by U.S. medical oncologists as the "standard of care" for all cancer patients; http://www.nccn.org/index.asp) include exercise in its treatment algorithm for cancer-related fatigue. However, our study demonstrates that some cancer survivors will not experience reductions in their fatigue with an exercise intervention. Identifying reasons for this lack of response can be used to tailor future exercise interventions for improved effectiveness. To our knowledge, we are the first study to report predictors of fatigue response to an exercise intervention. Although preliminary in nature, our data suggest that breast cancer survivors with higher baseline anxiety, depression, and barriers interference may be more likely to experience a beneficial fatigue response to our intervention. Given the moderately complex nature of our intervention with respect to cost and staff time, our data suggest that targeting individuals with predictors of greater response could be used to better allocate financial and staff resources. It is also noteworthy that the factors that contribute to developing fatigue after cancer diagnosis have been well studied but may not be the same factors that predict response to exercise. For example, affective state may increase fatigue prevalence but, in contrast, predict a better response to our exercise intervention. Further research is needed to better understand predictors of fatigue response to exercise.
Our data further support the fact that exercise intervention effects on fatigue may vary when different fatigue measures are used and/or aspects assessed. The four Fatigue Symptom Inventory items used to measure fatigue intensity were fatigue on the day felt most fatigued, fatigue on day felt least fatigued, average fatigue, and current fatigue. In our study, it is conceivable that exercise may transiently increase fatigue intensity after an exercise bout, which might increase the perception of fatigue by participants on the day they felt most fatigued and on average. However, the exercise training may have reduced fatigue interference because of improved participants' physical ability to engage in their daily activities regardless of the transient increases in fatigue after exercise bouts. Also related, the seven PROMIS® fatigue items include three items asking about fatigue, which interferes with work, thinking clearly, and bathing/showering; therefore, it is not surprising that the intervention effects were similar to that noted with fatigue interference. Consistent with this, the change in PROMIS® fatigue was significantly correlated with the change in fatigue interference in the intervention group (r = 0.62, P < 0.01). However, the PROMIS® fatigue scale combines other aspects of fatigue with fatigue interference for a more general fatigue assessment, which may explain the additional mediators related to minutes of physical activity and enjoyment. Future exercise and postcancer fatigue research should measure multiple fatigue aspects (or dimensions), assess the pattern of fatigue intensity pre- and postexercise bout, and compare how effects may differ for vigorous versus moderate-intensity exercise training. Further study is needed to better understand and define the meaning of fatigue (a subjective patient-specific measure) while also determining fatigue aspects responsive to exercise that are most important for improving patient quality of life.
Also related to the definition of fatigue, previous research by other investigators (primarily in noncancer populations) has differentiated between central and peripheral fatigue. We did not conceptualize fatigue in this manner because our goal was to examine fatigue as it might be reported by a patient in a clinical setting (i.e., patients usually do not differentiate central from peripheral fatigue). Nevertheless, interpretation of our data from the perspective of central versus peripheral fatigue warrants discussion and suggests future research directions. Peripheral fatigue is caused by neuromuscular abnormalities (e.g., excitation–contraction coupling, impaired calcium reuptake, etc.) that can be assessed with objective measures. Central fatigue results from a variety of possible central nervous system abnormalities, which include but are not limited to the lack of self-motivation influenced by psychosocial factors such as depression or catastrophe. Although further study is needed, central fatigue has been reported to be the primary cause of fatigue in cancer patients and survivors. Exercise without psychosocial support is expected to improve peripheral more than central fatigue, which may have contributed to the decline in fatigue interference and not fatigue intensity seen in our study. However, it is not possible to differentiate the effects of exercise alone from the effects of exercise plus the additional staff attention and group support in our study (e.g., support may have improved psychosocial factors that influence central fatigue). Therefore, it is possible that our intervention influenced both peripheral and central fatigue but to a different extent for each participant, thus explaining the variable rates of fatigue improvement and the greater effect on the more general measure of PROMIS® fatigue. The inclusion of measures that differentiate peripheral from central fatigue in future studies is warranted for improving our understanding of fatigue and its response to exercise interventions in cancer survivors.
Importantly, using consistent measures across studies would improve our ability to compare study results. PROMIS® is sponsored by the National Institutes of Health (NIH) and aims to develop a system of tools for patient-reported health status that can be used in multiple research and clinical populations and settings (http://www.nihpromis.org/about/abouthome). Our results indicate that this scale shows change over time with the intervention and can be used to examine fatigue mediators. Our study is the first exercise and cancer trial to report the use of this scale and supports the use of the PROMIS® fatigue scale in future trials to facilitate collection of data that are comparable not only across cancer types but also with various chronic disease populations.
Identifying factors mediating the largest proportions of exercise intervention effects on fatigue will help prioritize and focus future interventions (exercise and otherwise) to treat cancer-related fatigue. This is particularly important with regard to the role of inflammation in fatigue due to the inconsistent associations related to cytokines and fatigue after cancer diagnosis reported in the literature. Our data suggest that the pro–anti-inflammatory balance plays a more consistent role in mediating exercise effects on fatigue when compared with individual cytokines alone. It is possible that individual variation in fatigue response to exercise may be due, in part, to genomic differences in inflammatory response. However, larger trials are needed to confirm our results while also examining moderators of the inflammatory response and the complex interaction between changes in inflammatory markers and intervention effects on fatigue beyond mediation alone.
In an effort to better understand the complex relationships among the individual cytokines and related ratios, post hoc analyses examined the Pearson correlations among the raw difference scores for the intervention participants. Changes in IL-6 and IL-10 were significantly associated with TNF-α (r = 0.55, P < 0.05 and r = 0.49, P < 0.05, respectively). None of the individual cytokines were significantly associated with the ratios with the highest correlation noted for IL-10 and each of the ratios (r = 0.20 to −0.23, nonsignificant). Although our small sample size precludes definitive conclusions related to specific mechanistic pathways, the direction of effect size changes and the correlations among the difference scores support the theorized increase in anti-inflammatory cytokines (e.g., IL-10) as a result of higher levels of pro-inflammatory cytokines released during exercise.
Our data also provide additional support for the close association between sleep quality and fatigue previously reported in the literature. Mediation by social support and enjoyment support continued investigation of the theorized biobehavioral models of fatigue. It is noteworthy that cross-sectional associations have suggested a relationship between self-efficacy and fatigue but our study (the first to look at these relationships in a prospective design) did not detect a significant mediation effect by self-efficacy. Taken as a whole, our data suggest that combining exercise with interventions including sleep hygiene counseling, exercise social support, and exercise enjoyment has the potential to improve intervention effects on fatigue.
We originally published a pilot study evaluating a physical activity behavior change intervention effects on inflammatory markers of inflammation. Because of the small effects on cytokines, we attempted to reduce variability, which would increase effect sizes by narrowing our study inclusion/exclusion criteria and by prescribing a more specific exercise dose (i.e., 40 min of aerobic exercise on 4 d·wk with no more than 2 d lapsing between exercise sessions and resistance training on two nonconsecutive days of the week). We also included participants with fatigue and/or sleep dysfunction in an effort to prevent the potential "floor effect" occurring when nonfatigued individuals are included. When the two studies are compared, fatigue effect sizes are higher in the study reported here. Also, the directions of cytokine-related effects were similar with the exception of TNF-α, but the magnitude of the effects remained small to medium in size. Importantly, effect sizes in both studies, although small, suggest beneficial changes in the pro–anti-inflammatory ratios with chronic exercise participation.
Although extensor leg strength was assessed using a back and leg dynamometer, no significant change in this outcome was noted for the intervention compared with the control group (d = −0.07, P = 0.831). This differs from prior reports, indicating an increase in strength with this measurement in response to a similar walking intervention. Several possible explanations exist. Prior study assessments were done by individuals who were not blinded to group allocation. Because of the lack of blinding, assessors may have inadvertently provided increased encouragement during the testing for participants in the intervention group. Also, our intervention focused on general muscle strength and aerobic fitness rather than being specific to those muscle groups tested with the back/leg dynamometer. Lastly, our resistance protocol may not have been rigorous enough to result in significant improvements in muscle strength using the simple back/leg dynamometer. Future studies should assess mediation of fatigue using a strength measure sensitive to change with the intervention and/or a more intensive resistance training protocol before muscle strength is excluded as a possible mediator of fatigue response to exercise.
Also related to the adequacy of the exercise dose, the baseline mean weekly minutes of ≥moderate-intensity exercise exceeded the intervention goal of 160 weekly minutes, which could conceivably threaten sufficient increases in exercise minutes due to a ceiling effect. The self-report of leisure-time exercise was used when determining study eligibility during the screening process, but only the objective measure is reported here because it is generally considered to be a more accurate assessment of exercise behavior when compared with self-report. Nevertheless, accelerometers do not differentiate between leisure activity (i.e., volitional behavior that is more apt to change with an intervention) from nonleisure activities (e.g., occupation). Therefore, it is possible that some participants may have had greater amounts of physical activity when they wore the accelerometer (compared with self-report) because of nonleisure activities. The magnitude of the between-group differences were similar for self-report and accelerometer (i.e., 110.5 min for self-report and 98.8 min for the accelerometer), and the baseline mean of the self-report was 17 ± 39 min for all participants combined. This suggests that the volitional (or leisure) exercise, which would be anticipated to change the most during an exercise intervention, was sufficiently low at baseline to limit a ceiling effect. Furthermore, the standardized effect size for aerobic fitness (i.e., 0.37) is comparable with that reported by other studies, including the weighted mean standardized effect size in a meta-analysis of exercise studies in cancer survivors (i.e., 0.32 for posttreatment cancer survivors). In addition, the paired t-test of the within group change demonstrated a significant improvement in fitness for the intervention group participants from baseline to postintervention. Therefore, limited study power due to the relatively large standard deviation of the between-group difference is a more likely explanation of the lack of statistical significance for aerobic fitness rather than an aerobic intervention that was "too mild" to cause improvement.
Discussion groups were included in our intervention to improve adherence to the exercise protocol. These groups may have inadvertently affected fatigue because they encouraged cognitive reframing (which may have influenced enjoyment) and social support relative to exercise. Therefore, conclusions about the effects of exercise independent of the group sessions cannot be made especially given the mediation by social support. However, the significant mediation of PROMIS® fatigue by the increase in physical activity minutes suggests that exercise independent of the groups plays a role. Also, our results suggest the importance of interventions that focus on multiple potential mediators. We also acknowledge the limited study power due to the small sample size. Nevertheless, strong study design (e.g., randomized controlled trial), use of multiple fatigue measures, excellent retention, and report of mediators in a prospective study design significantly improve the usefulness of these study data. Also, we documented with accelerometer that time spent in sedentary behavior did not change for the intervention compared with control group. This is important because of the health risks associated with sedentary behavior, associations between sedentary behavior and fatigue in breast cancer survivors, and concerns about exercise training causing individuals to be less active during other times of day.
Our data suggest several important clinical and research implications. The correlation between the difference scores for change in fatigue intensity and interference in the intervention group was 0.44 (P < 0.10), suggesting that these constructs are different (account for only 19% of the variance of the other construct) yet overlap. It is possible that exercise that is too rigorous for an individual might increase fatigue intensity, which worsens interference and could potentially act as an exercise barrier. Therefore, exercise recommendations for survivors with higher fatigue intensity should focus on adapting the exercise program that monitors for and avoids increases in fatigue intensity. In contrast, an individual with higher levels of anxiety and depressive symptoms at baseline can be advised to adapt an exercise training protocol similar to our intervention. Further research is needed to identify strategies for and usefulness of tailoring exercise counseling to the nature of the cancer survivor's fatigue.
Inflammation, sleep quality, and psychosocial factors may mediate or influence exercise intervention effects on fatigue in breast cancer survivors. Larger trials are needed to confirm our results and better understand the complex mediator and moderator relationships between biobehavioral factors and fatigue response to exercise. The inclusion of additional possible mediators such as catastrophe, pain perception, neuropeptides, and catecholamines should be considered. Future studies should use several fatigue measures, including but not limited to the PROMIS® fatigue scale, to allow comparison with other studies and measurement of different fatigue aspects. If the biobehavioral mechanisms suggested in this study continue to be observed, interventions targeting these mechanisms can be developed to reduce fatigue in breast cancer survivors.
Discussion
Although not statistically significant, the direction and magnitude of the effect sizes related to our exercise intervention varied depending on the fatigue measure used and aspect assessed. A nonsignificant small to medium effect size increase was noted for fatigue intensity, with a nonsignificant small to medium effect size decrease noted for fatigue interference and PROMIS® fatigue. Only PROMIS® fatigue showed a within-group statistically significant decline in the intervention group. Study power was limited by our small sample size, which resulted from the budgetary and logistical constraints of a pilot study. Nevertheless, our effect size reductions in fatigue interference and general fatigue are consistent with that of previous studies, with our results suggesting an important finding relative to a possible increase in intermittent fatigue intensity. It is noteworthy that our data suggest important mediating relationships warranting further study. Specifically, inflammation, sleep quality, psychosocial factors (i.e., exercise social support and enjoyment), and minutes of weekly physical activity mediated the effects of our exercise intervention on fatigue in breast cancer survivors. Complex relationships that include both mediation and strengthening of the intervention–fatigue relationship exist. These relationships vary among the fatigue measures used, and further study is needed to improve our understanding of these relationships. The inflammatory mediators of fatigue response to exercise may be due to beneficial reductions in the pro–anti-inflammatory ratios resulting from the dynamic response of the cytokine system to pro-inflammatory aspects of exercise training.
The National Comprehensive Cancer Network (NCCN) guidelines (a national resource used by U.S. medical oncologists as the "standard of care" for all cancer patients; http://www.nccn.org/index.asp) include exercise in its treatment algorithm for cancer-related fatigue. However, our study demonstrates that some cancer survivors will not experience reductions in their fatigue with an exercise intervention. Identifying reasons for this lack of response can be used to tailor future exercise interventions for improved effectiveness. To our knowledge, we are the first study to report predictors of fatigue response to an exercise intervention. Although preliminary in nature, our data suggest that breast cancer survivors with higher baseline anxiety, depression, and barriers interference may be more likely to experience a beneficial fatigue response to our intervention. Given the moderately complex nature of our intervention with respect to cost and staff time, our data suggest that targeting individuals with predictors of greater response could be used to better allocate financial and staff resources. It is also noteworthy that the factors that contribute to developing fatigue after cancer diagnosis have been well studied but may not be the same factors that predict response to exercise. For example, affective state may increase fatigue prevalence but, in contrast, predict a better response to our exercise intervention. Further research is needed to better understand predictors of fatigue response to exercise.
Our data further support the fact that exercise intervention effects on fatigue may vary when different fatigue measures are used and/or aspects assessed. The four Fatigue Symptom Inventory items used to measure fatigue intensity were fatigue on the day felt most fatigued, fatigue on day felt least fatigued, average fatigue, and current fatigue. In our study, it is conceivable that exercise may transiently increase fatigue intensity after an exercise bout, which might increase the perception of fatigue by participants on the day they felt most fatigued and on average. However, the exercise training may have reduced fatigue interference because of improved participants' physical ability to engage in their daily activities regardless of the transient increases in fatigue after exercise bouts. Also related, the seven PROMIS® fatigue items include three items asking about fatigue, which interferes with work, thinking clearly, and bathing/showering; therefore, it is not surprising that the intervention effects were similar to that noted with fatigue interference. Consistent with this, the change in PROMIS® fatigue was significantly correlated with the change in fatigue interference in the intervention group (r = 0.62, P < 0.01). However, the PROMIS® fatigue scale combines other aspects of fatigue with fatigue interference for a more general fatigue assessment, which may explain the additional mediators related to minutes of physical activity and enjoyment. Future exercise and postcancer fatigue research should measure multiple fatigue aspects (or dimensions), assess the pattern of fatigue intensity pre- and postexercise bout, and compare how effects may differ for vigorous versus moderate-intensity exercise training. Further study is needed to better understand and define the meaning of fatigue (a subjective patient-specific measure) while also determining fatigue aspects responsive to exercise that are most important for improving patient quality of life.
Also related to the definition of fatigue, previous research by other investigators (primarily in noncancer populations) has differentiated between central and peripheral fatigue. We did not conceptualize fatigue in this manner because our goal was to examine fatigue as it might be reported by a patient in a clinical setting (i.e., patients usually do not differentiate central from peripheral fatigue). Nevertheless, interpretation of our data from the perspective of central versus peripheral fatigue warrants discussion and suggests future research directions. Peripheral fatigue is caused by neuromuscular abnormalities (e.g., excitation–contraction coupling, impaired calcium reuptake, etc.) that can be assessed with objective measures. Central fatigue results from a variety of possible central nervous system abnormalities, which include but are not limited to the lack of self-motivation influenced by psychosocial factors such as depression or catastrophe. Although further study is needed, central fatigue has been reported to be the primary cause of fatigue in cancer patients and survivors. Exercise without psychosocial support is expected to improve peripheral more than central fatigue, which may have contributed to the decline in fatigue interference and not fatigue intensity seen in our study. However, it is not possible to differentiate the effects of exercise alone from the effects of exercise plus the additional staff attention and group support in our study (e.g., support may have improved psychosocial factors that influence central fatigue). Therefore, it is possible that our intervention influenced both peripheral and central fatigue but to a different extent for each participant, thus explaining the variable rates of fatigue improvement and the greater effect on the more general measure of PROMIS® fatigue. The inclusion of measures that differentiate peripheral from central fatigue in future studies is warranted for improving our understanding of fatigue and its response to exercise interventions in cancer survivors.
Importantly, using consistent measures across studies would improve our ability to compare study results. PROMIS® is sponsored by the National Institutes of Health (NIH) and aims to develop a system of tools for patient-reported health status that can be used in multiple research and clinical populations and settings (http://www.nihpromis.org/about/abouthome). Our results indicate that this scale shows change over time with the intervention and can be used to examine fatigue mediators. Our study is the first exercise and cancer trial to report the use of this scale and supports the use of the PROMIS® fatigue scale in future trials to facilitate collection of data that are comparable not only across cancer types but also with various chronic disease populations.
Identifying factors mediating the largest proportions of exercise intervention effects on fatigue will help prioritize and focus future interventions (exercise and otherwise) to treat cancer-related fatigue. This is particularly important with regard to the role of inflammation in fatigue due to the inconsistent associations related to cytokines and fatigue after cancer diagnosis reported in the literature. Our data suggest that the pro–anti-inflammatory balance plays a more consistent role in mediating exercise effects on fatigue when compared with individual cytokines alone. It is possible that individual variation in fatigue response to exercise may be due, in part, to genomic differences in inflammatory response. However, larger trials are needed to confirm our results while also examining moderators of the inflammatory response and the complex interaction between changes in inflammatory markers and intervention effects on fatigue beyond mediation alone.
In an effort to better understand the complex relationships among the individual cytokines and related ratios, post hoc analyses examined the Pearson correlations among the raw difference scores for the intervention participants. Changes in IL-6 and IL-10 were significantly associated with TNF-α (r = 0.55, P < 0.05 and r = 0.49, P < 0.05, respectively). None of the individual cytokines were significantly associated with the ratios with the highest correlation noted for IL-10 and each of the ratios (r = 0.20 to −0.23, nonsignificant). Although our small sample size precludes definitive conclusions related to specific mechanistic pathways, the direction of effect size changes and the correlations among the difference scores support the theorized increase in anti-inflammatory cytokines (e.g., IL-10) as a result of higher levels of pro-inflammatory cytokines released during exercise.
Our data also provide additional support for the close association between sleep quality and fatigue previously reported in the literature. Mediation by social support and enjoyment support continued investigation of the theorized biobehavioral models of fatigue. It is noteworthy that cross-sectional associations have suggested a relationship between self-efficacy and fatigue but our study (the first to look at these relationships in a prospective design) did not detect a significant mediation effect by self-efficacy. Taken as a whole, our data suggest that combining exercise with interventions including sleep hygiene counseling, exercise social support, and exercise enjoyment has the potential to improve intervention effects on fatigue.
We originally published a pilot study evaluating a physical activity behavior change intervention effects on inflammatory markers of inflammation. Because of the small effects on cytokines, we attempted to reduce variability, which would increase effect sizes by narrowing our study inclusion/exclusion criteria and by prescribing a more specific exercise dose (i.e., 40 min of aerobic exercise on 4 d·wk with no more than 2 d lapsing between exercise sessions and resistance training on two nonconsecutive days of the week). We also included participants with fatigue and/or sleep dysfunction in an effort to prevent the potential "floor effect" occurring when nonfatigued individuals are included. When the two studies are compared, fatigue effect sizes are higher in the study reported here. Also, the directions of cytokine-related effects were similar with the exception of TNF-α, but the magnitude of the effects remained small to medium in size. Importantly, effect sizes in both studies, although small, suggest beneficial changes in the pro–anti-inflammatory ratios with chronic exercise participation.
Although extensor leg strength was assessed using a back and leg dynamometer, no significant change in this outcome was noted for the intervention compared with the control group (d = −0.07, P = 0.831). This differs from prior reports, indicating an increase in strength with this measurement in response to a similar walking intervention. Several possible explanations exist. Prior study assessments were done by individuals who were not blinded to group allocation. Because of the lack of blinding, assessors may have inadvertently provided increased encouragement during the testing for participants in the intervention group. Also, our intervention focused on general muscle strength and aerobic fitness rather than being specific to those muscle groups tested with the back/leg dynamometer. Lastly, our resistance protocol may not have been rigorous enough to result in significant improvements in muscle strength using the simple back/leg dynamometer. Future studies should assess mediation of fatigue using a strength measure sensitive to change with the intervention and/or a more intensive resistance training protocol before muscle strength is excluded as a possible mediator of fatigue response to exercise.
Also related to the adequacy of the exercise dose, the baseline mean weekly minutes of ≥moderate-intensity exercise exceeded the intervention goal of 160 weekly minutes, which could conceivably threaten sufficient increases in exercise minutes due to a ceiling effect. The self-report of leisure-time exercise was used when determining study eligibility during the screening process, but only the objective measure is reported here because it is generally considered to be a more accurate assessment of exercise behavior when compared with self-report. Nevertheless, accelerometers do not differentiate between leisure activity (i.e., volitional behavior that is more apt to change with an intervention) from nonleisure activities (e.g., occupation). Therefore, it is possible that some participants may have had greater amounts of physical activity when they wore the accelerometer (compared with self-report) because of nonleisure activities. The magnitude of the between-group differences were similar for self-report and accelerometer (i.e., 110.5 min for self-report and 98.8 min for the accelerometer), and the baseline mean of the self-report was 17 ± 39 min for all participants combined. This suggests that the volitional (or leisure) exercise, which would be anticipated to change the most during an exercise intervention, was sufficiently low at baseline to limit a ceiling effect. Furthermore, the standardized effect size for aerobic fitness (i.e., 0.37) is comparable with that reported by other studies, including the weighted mean standardized effect size in a meta-analysis of exercise studies in cancer survivors (i.e., 0.32 for posttreatment cancer survivors). In addition, the paired t-test of the within group change demonstrated a significant improvement in fitness for the intervention group participants from baseline to postintervention. Therefore, limited study power due to the relatively large standard deviation of the between-group difference is a more likely explanation of the lack of statistical significance for aerobic fitness rather than an aerobic intervention that was "too mild" to cause improvement.
Discussion groups were included in our intervention to improve adherence to the exercise protocol. These groups may have inadvertently affected fatigue because they encouraged cognitive reframing (which may have influenced enjoyment) and social support relative to exercise. Therefore, conclusions about the effects of exercise independent of the group sessions cannot be made especially given the mediation by social support. However, the significant mediation of PROMIS® fatigue by the increase in physical activity minutes suggests that exercise independent of the groups plays a role. Also, our results suggest the importance of interventions that focus on multiple potential mediators. We also acknowledge the limited study power due to the small sample size. Nevertheless, strong study design (e.g., randomized controlled trial), use of multiple fatigue measures, excellent retention, and report of mediators in a prospective study design significantly improve the usefulness of these study data. Also, we documented with accelerometer that time spent in sedentary behavior did not change for the intervention compared with control group. This is important because of the health risks associated with sedentary behavior, associations between sedentary behavior and fatigue in breast cancer survivors, and concerns about exercise training causing individuals to be less active during other times of day.
Our data suggest several important clinical and research implications. The correlation between the difference scores for change in fatigue intensity and interference in the intervention group was 0.44 (P < 0.10), suggesting that these constructs are different (account for only 19% of the variance of the other construct) yet overlap. It is possible that exercise that is too rigorous for an individual might increase fatigue intensity, which worsens interference and could potentially act as an exercise barrier. Therefore, exercise recommendations for survivors with higher fatigue intensity should focus on adapting the exercise program that monitors for and avoids increases in fatigue intensity. In contrast, an individual with higher levels of anxiety and depressive symptoms at baseline can be advised to adapt an exercise training protocol similar to our intervention. Further research is needed to identify strategies for and usefulness of tailoring exercise counseling to the nature of the cancer survivor's fatigue.
Inflammation, sleep quality, and psychosocial factors may mediate or influence exercise intervention effects on fatigue in breast cancer survivors. Larger trials are needed to confirm our results and better understand the complex mediator and moderator relationships between biobehavioral factors and fatigue response to exercise. The inclusion of additional possible mediators such as catastrophe, pain perception, neuropeptides, and catecholamines should be considered. Future studies should use several fatigue measures, including but not limited to the PROMIS® fatigue scale, to allow comparison with other studies and measurement of different fatigue aspects. If the biobehavioral mechanisms suggested in this study continue to be observed, interventions targeting these mechanisms can be developed to reduce fatigue in breast cancer survivors.
