• physical activity;
  • breast cancer;
  • mortality;
  • recurrence;
  • prognosis


  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References

Evidence is emerging that physical activity (PA) may improve overall survival after breast cancer diagnosis. However, the effect of PA on breast cancer recurrence and on cause-specific mortality is less investigated. We assessed the association of pre-diagnosis PA with recurrence, overall and cause-specific survival in a prospective cohort study in Germany including 3,393 non-metastatic breast cancer patients aged 50–74 years. Cox proportional hazards models were calculated adjusted for relevant prognostic factors. During a median follow-up of 5.6 years, 367 patients deceased. Overall mortality was significantly inversely associated with pre-diagnosis recreational PA. However, this effect was mainly attributed to deaths due to causes other than breast cancer. Multiple fractional polynomial analyses yielded a nonlinear association with markedly increased non-breast cancer mortality for women who did not engage in any sports or cycling in the years before the breast cancer diagnosis with a hazard ratio (HR, none vs. any) of 1.71, 95% confidence interval (1.16, 2.52). There were no further risk reductions with increasing activity levels. The association with breast cancer-specific mortality showed a similar dose-response but was far less pronounced with HR (none vs. any) = 1.22 (0.91, 1.64). In contrast, regarding cancer recurrence the dose-response was linear. However, this association was restricted to estrogen/progesterone receptor-negative (ER−/PR−) cases (pinteraction = 0.033) with HR (highest vs. no recreational PA) = 0.53 (0.24, 1.16), ptrend = 0.0045. Thus, breast cancer patients with a physically inactive lifestyle pre-diagnosis may decease prematurely irrespective of their cancer prognosis. Higher levels of exercise may reduce the risk of recurrence of ER−/PR− breast tumors.


body mass index


confidence interval


estrogen/progesterone receptor


hazard ratio: HT: hormone therapy


metabolic equivalent


physical activity


peripheral arterial obstructive disease

There is convincing evidence that physical activity (PA) is a preventive factor for postmenopausal breast cancer.[1] Several biological pathways may take effect on cancer development or progression. Studies have shown that PA reduces endogenous estrogens,[2-5] insulin resistance[6, 7] and inflammation.[8, 9] Hereby, PA acts partly by changing body composition, but effects have also been observed independently of body mass index (BMI).[3, 5, 7] Moreover, several studies suggest that PA might affect tumor progression rather than tumor initiation.[10-13]

Some of these primary-preventive effects of PA on breast cancer development might also act after a breast cancer diagnosis, inhibiting progression and improving prognosis. Indeed, evidence is emerging that PA improves survival after breast cancer. A meta-analysis published in 2010 investigating the effect of PA on survival after breast cancer (n = 12,108 patients) identified four studies on pre-diagnosis and three on post-diagnosis PA.[14] This meta-analysis resulted in a hazard ratio (HR) of 0.82 with a 95% confidence interval (CI) of (0.67, 0.99) for the association of pre-diagnosis PA with overall mortality. Post-diagnosis PA was associated with an even stronger risk reduction with HR = 0.59 (0.53, 0.65). Regarding breast cancer-specific mortality a significant association with post-diagnosis PA was observed (HR = 0.66 (0.57, 0.77)) but not with pre-diagnosis PA. Recently, more studies have been published with results tending toward beneficial effects of PA before or after diagnosis on overall or breast cancer-specific survival.[15-20] However, studies are inconclusive as to whether effects are as strong for breast cancer-specific mortality as for overall mortality. In addition, the association of PA with breast cancer recurrence or progression has been rarely investigated and existing studies yield mixed results.[15, 21]

Therefore, we evaluated the association of pre-diagnosis PA with cancer recurrence as well as with mortality, hereby differentiating between deaths related to breast cancer and deaths due to other causes. Moreover, as the amount of PA needed to improve prognosis and survival after breast cancer is still unclear, we explored dose–response relationships using multiple fractional polynomial analyses.

Material and Methods

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References

Study population

The MARIE study is a population-based case–control study including 3,813 breast cancer patients aged 50–74 years from two German study regions (Hamburg and Rhein-Neckar-Karlsruhe).[22] The patients were diagnosed with a histologically confirmed primary invasive or in situ breast tumor between January 2001 and September 2005 in Hamburg and between August 2002 and July 2005 in the Rhein-Neckar-Karlsruhe region and were identified through participating clinics and the Cancer Registry of Hamburg. Cases were followed up until the end of 2009.

For the survival analyses we excluded 259 patients with previous tumors (except in situ carcinoma, and basal and squamous cell skin carcinoma), 107 patients with Stage IV at primary diagnosis (baseline) and 54 with incomplete data on PA, TNM status, grading, or estrogen/progesterone receptor (ER/PR) status, resulting in 3,393 cases (see Fig. 1). For the investigation on recurrence, we additionally excluded patients with in situ or Stage IIIb or IIIc carcinomas, as those have different prognosis compared to Stage I–IIIa, resulting in a sample of n = 2,902 patients. Sensitivity analyses showed no change in mortality results when in situ or Stage IIIb or IIIc carcinomas were excluded also for the other endpoints.


Figure 1. Patient flow.

Download figure to PowerPoint

The study was approved by the ethics committees of the University of Heidelberg, the Medical Board of the Medical Association of Hamburg and the Medical Board of the State of Rheinland-Pfalz, and was conducted in agreement with the Helsinki declaration. All participants provided written informed consent.

Data assessment

At recruitment, in-person interviews were performed to collect information on demographic, anthropometric, socioeconomic and lifestyle factors including PA, as well as self-reported information on medications and on comorbidities that had ever been diagnosed by a physician. Moreover, further information on possible prognostic factors was collected from clinical and pathological records.

Assessment of physical activity

PA since age 50 years until breast cancer diagnosis was assessed using a detailed interviewer-administered questionnaire.[23] It was first asked for duration and type of occupational PA and time spent doing household tasks (including gardening and child care). Next, splitting a typical week into weekdays and weekend, the number of hours per week spent walking as well as the number of hours per week spent cycling were recorded. Finally, participants were asked to list up to three sports performed during the considered age period, including sport type, duration and frequency. A metabolic equivalent (MET) value was assigned to each reported activity according to the Compendium of Physical Activities.[24, 25] Domain specific and combined PA variables in MET-hours per week were calculated by summing the average hours per week spent walking, cycling, engaged in sports or in occupation and household weighted by the appropriate MET values. As day-to-day cycling and sports showed similar effects and both activities are physiologically similar, they were combined into a single recreational PA variable. Recreational PA was investigated as continuous variable, and in addition was categorized using meaningful cutpoints that provide roughly equally sized categories, i.e. = 0 (no recreational PA), <12 MET-hr/week (e.g., less than 2 hr cycling or moderate exercise per week), 12 to <24 MET-hr/week, 24 to <42 MET-hr/week and ≥42 MET-hr/week (e.g., more than 1 hr cycling or exercise daily).

Outcome definition

Vital status of patients was ascertained via local population registries through the end of 2009 (100% completeness of follow-up), and causes of death were verified by death certificates and coded based on ICD-10 classifications. During follow-up telephone interviews conducted from May to September 2009 patients were asked about recurrences or second cancers. In addition, medical records were checked or treating physicians were contacted to identify recurrences or second cancers of those patients who had died or could not be contacted, and to verify the information collected in the interviews (>90% self-reported events verified). The endpoints of interest were overall mortality, breast cancer-specific mortality, other deaths and local or distant recurrence. For breast cancer-specific mortality, deaths from breast cancer (coded as ICD-10 C50) were events of interest and deaths from other causes were censored at date of occurrence. Vice-versa, when other deaths were considered as endpoints, breast cancer deaths were censored. Recurrence included ipsilateral/contralateral/local/regional invasive recurrence, or distant recurrence emerging after the primary diagnosis. Participants without event of interest were censored at date of last contact or December 31, 2009, whichever came first. Median follow-up time was calculated as time between recruitment and the event of interest or censoring using the reverse Kaplan–Meier estimation.[26]

Statistical analysis

Differences in baseline characteristics between recreational PA categories were tested using the χ2-statistic for nominal variables and Kruskal–Wallis-test for ordinal variables.

We used delayed entry Cox proportional hazard models, based on time since study enrollment until event or censoring, to estimate HRs and 95% CIs for overall survival as well as cause-specific HRs for breast cancer-specific mortality, other deaths and breast cancer recurrence. As exposure variables we investigated total PA, recreational PA, household PA, occupational PA and walking. As associations were mainly observed with recreational PA, we focused on this PA variable. All analyses were stratified by study center and adjusted for covariables selected on the basis of the theory of directed acyclic graphs,[27] i.e. age at diagnosis and the established prognostic factors: tumor size (<2 cm, 2–5 cm, ≥5 cm, growth into chest wall, neoadjuvant chemotherapy-treated carcinoma, in situ carcinoma), nodal status (0, 1–3, 4–9, ≥10 affected lymph nodes, neoadjuvant chemotherapy-treated carcinoma, in situ carcinoma), histological grade (1+2, 3+4, in situ carcinoma, neoadjuvant chemotherapy-treated carcinoma), ER/PR status (in situ carcinoma, ER+/PR+, ER+/PR− or ER−/PR+, ER−/PR−, neoadjuvant chemotherapy-treated carcinoma), radiotherapy (yes/no), breast cancer detection type (physician-detected by clinical examination/mammography/ultrasound, self-detected by palpation/secretion/pain), use of menopausal hormone therapy at diagnosis (current, never/past), pre-diagnosis BMI (<18.5, 18.5 to <25, 25 to <30, ≥30 kg/m2), smoking status at diagnosis (current, ex, never) and packyears, as well as pre-existing angina pectoris. Models for overall mortality and for other deaths were adjusted in addition for pre-existing hypertension (yes/no), previous stroke (yes/no) and use of insulin (yes/no). Other adjustment sets considering the following covariates were also investigated, but did not alter the HRs for the PA variables by more than 10%: type of surgery (mastectomy, breast conserving), chemotherapy (yes/no), hormone therapy (yes/no), HER2-neu status, postmenopausal at diagnosis (yes/no), marital status (married, single, separated, divorced, widowed), alcohol use at diagnosis (0, >0 to <19, ≥19 g/day), previous myocardial infarction (yes/no), peripheral arterial obstructive disease (yes/no), osteoporosis (yes/no), rheumatoid arthritis (yes/no), venous thrombosis (yes/no), previous pulmonary embolism (yes/no), chronic lung disease (yes/no), chronic liver (yes/no), gastric or renal diseases (yes/no), psychological disorders (yes/no), migraine (yes/no), thyroid disorders (yes/no), previous myoms (yes/no) or ovarian cysts (yes/no).

To assess nonlinear trends for the continuous PA variable in the multivariable Cox models, we used the method of fractional polynomials.[28] The continuous PA variable was entered in the multivariable Cox models via a set of defined transformations {x−2, x−1, x−0.5, x0.5, x, x2, x3, log(x)}, allowing a maximum of two terms in the model. The best-fitting transformation based on the −2 log likelihood of the respective model was selected.

The proportional hazards assumption was tested by including a time-dependent covariate representing the interaction of the PA variable and follow-up time.[29] No indications for a violation of the proportional hazards assumption was found for any of the analyses.

Sensitivity analyses were performed for all three mortality endpoints by excluding patients with in situ or with late-stage breast cancer (Stage IIIb and IIIc). Possible effect modification by tumor characteristics (in situ/invasive; low + moderate/high grade; ER+PR+/ER−PR−/other), pre-diagnosis BMI (<25/≥25), smoking status (ever/never) or pre-existing cardiovascular diseases (yes/no) was assessed by including an interaction term in the regression models.

For all analyses, two-sided p-values <0.05 were considered significant. No adjustment for multiple testing was performed because of the exploratory character of this study. All analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC).


  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References

The median age of the 3,393 patients included in the analyses was 62.7 years (interquartile range: 57.6–66.7) at diagnosis. The majority of patients (73.9%) had a normal adult BMI (18–25 kg/m2) before cancer diagnosis, were non-smokers (52.8%) or ex-smokers (27.8%) and had a basic educational level (57.5%) as it is typical for German women of this generation. Regarding the PA behavior, 675 patients (19.9%) did not engage in any sports or cycling since the age of 50 years until cancer diagnosis, while 597 (17.6%) were highly physically active in this period (≥42 MET-hr/week sports or cycling). The median recreational PA was 14.3 MET-hr/week (interquartile range: 2.8–32.4). Table 1 presents the characteristics of the study population by recreational PA level. Among those who did not engage in any sports or cycling before diagnosis, there were more smokers, less women with higher education, less users of menopausal hormone therapy and more patients with large breast tumors. Less active women also had more often pre-existing disorders, especially hypertension, angina pectoris, previous stroke or diabetes.

Table 1. Characteristics of the study population
 Pre-diagnosis recreational PA (MET-hr/week)
Median (Q1, Q3) or column (%)None>0 to <1212 to <2424 to <42≥42p2
(n = 675)(n = 833)(n = 687)(n = 619)(n = 579)
  1. BMI, usual adult body mass index during age 25–50; CT, chemotherapy; ER, estrogen receptor; HT, hormone therapy; MET, metabolic equivalent; PA, physical activity; PAOD, Peripheral arterial obstructive disease; PR, progesterone receptor; Q1, first quartile; Q3 third quartile.

  2. a

    Tumor classification is not possible for patients who received neoadjuvant chemotherapy or had in situ carcinoma.

  3. b

    χ2 test for nominal variables and Kruskal–Wallis test for ordinal variables.

  4. c

    Only patients with stage I–IIIa at baseline are included in this analysis.

Age (years), median (Q1, Q3)62 (57, 67)63 (58, 68)63 (58, 67)62 (57, 66)63 (58, 66)0.0030
BMI (kg/mb), median (Q1, Q3)23.3 (21.3, 25.5)22.8 (21.0, 24.8)22.7 (21.2, 24.5)22.6 (20.8, 24.5)22.5 (20.8, 24.5)<0.0001
Smoker (at diagnosis, %)29.318.918.513.715.7<0.0001
Menopausal HT (%)39.948.045.950.650.40.0044
Education (%)     <0.0001
Tumor size (%)      
≤2 cm44.653.457.955.152.20.0004
>2–5 cm38.731.629.329.633.2 
>5 cm4. 
Into chest wall/skin2. 
Neoadjuvant CTa5. 
In situ carcinoma4. 
Nodal status (%)     0.19 
Neoadjuvant CTa5. 
In situ carcinoma4. 
Tumor grade (%)     0.18
Neoadjuvant CTa5. 
In situ carcinoma4. 
ER/PR status (%)     0.22
Neoadjuvant CTa5. 
In situ carcinoma4. 
Mode of detection (%)     0.067
Type of surgery (%)     0.044
Breast conserving65.668.372.572.772.0 
Chemotherapy (%)     0.0071
Radiotherapy (%)     0.19
Hormonal therapy (%)     0.22
Diabetes (%)
Use of insulin (%)
Arthritis (%)49.653.648.944.744.00.025
Osteoporosis (%)11.37.810.
Hypertension (%)48.344.537.335.935.6<0.0001
Angina pectoris (%)
PAOD (%)
Stroke (%)
Myocardial infarction (%)
Thrombosis (%)
Chronic lung disease (%)12.311.
Number of events (n)      
Overall mortality10981656052<0.0001
Breast cancer mortality64554343380.13
Other deaths45262217140.0002

Baseline assessments were conducted at a median time of 104 days after diagnosis. Overall, 367 deaths occurred, 243 (66.2%) due to breast cancer. Further causes of death were other cancers (n = 58, 15.8%), cardiovascular disease (n = 26, 7.1%), cerebrovascular disorders (n = 8, 2.4%), and other causes (n = 32, 9.4%). Of the 2,902 patients with Stage I–IIIa disease and available data on recurrence status, 317 (10.9%) had a breast cancer recurrence.

The Kaplan–Meier analysis presented in Figure 2 shows that patients who did not engage in any sports or cycling had a significantly poorer overall survival (log-rank p < 0.0001) compared to all other PA categories where the survival curves did not differ substantially. Among physically inactive women the 5-year survival rate was 88.2% while among the others it was 93.3%.


Figure 2. Kaplan–Meier distribution functions for overall survival, stratified by pre-diagnosis recreational PA categories.

Download figure to PowerPoint

The Cox regression adjusting for potential confounders showed significant reductions in overall mortality and deaths due to causes other than breast cancer for physically active women regardless of their activity level compared to the inactive women (Table 2). HRs for overall mortality ranged between 0.66 and 0.77. HRs for non-breast cancer mortality were even lower ranging between 0.50 and 0.71. In contrast, HRs for breast cancer-specific deaths were less pronounced and not statistically significant. For breast cancer recurrence the adjusted HRs decreased with increasing PA category from 0.96 to 0.65, reaching statistical significance only for the highest PA category. These associations with recreational PA were independent of other prognostic factors such as tumor characteristics or treatment, pre-existing diseases, pre-diagnosis BMI or smoking. To investigate the dose–response relationship, recreational PA was included in the models in different polynomial forms following the algorithm for multiple fractional polynomials (Fig. 3). For breast cancer recurrence, a linear dose-response showed up as the best fitting term (ptrend = 0.067). In contrast, the best fitting models for overall mortality, breast cancer-specific mortality and mortality due to other causes included PA in the form PA−0.5, i.e. there was no linear dose–response relationship but rather a higher HR for inactivity compared to any activity (Fig. 3a). Accordingly, we then dichotomized recreational PA and calculated the mortality risks for inactivity vs. any activity yielding HR = 1.41 (1.12, 1.78) for overall mortality, HR = 1.22 (0.91, 1.64) for breast-cancer-specific mortality and HR = 1.71 (1.16, 2.52) for other deaths (Table 3).

Table 2. Adjusteda hazard ratios and 95% confidence intervals for pre-diagnosis recreational physical activity regarding the endpoints overall mortality, breast cancer-specific death, death due to other causes and cancer recurrence
 Pre-diagnosis recreational PA (Met-hr/week)
 None>0 to <1212 to <2424 to <42≥ 42
  1. BMI, usual adult body mass index during age 25–50; ER, estrogen receptor; HT, hormone therapy; MET, metabolic equivalent; PA, physical activity; PR, progesterone receptor.

  2. a

    All models were adjusted for tumor size, nodal status, tumor grading, ER/PR status, radiotherapy, screening-detected tumor, HT use at diagnosis, age at diagnosis, BMI pre-diagnosis, smoking status and packyears and pre-existing angina pectoris. In addition, models for overall mortality and for other deaths were adjusted for pre-existing hypertension, previous stroke and use of insulin.

  3. b

    Endpoint includes ipsilateral/contralateral/local/regional invasive recurrence, or distant recurrence emerging after the primary diagnosis. Only patients with stage I–IIIa at baseline are included in this analysis.

Overall mortality1.000.67 (0.50, 0.90)0.76 (0.55, 1.04)0.77 (0.56, 1.07)0.66 (0.47, 0.92)
Breast cancer specific deaths1.000.74 (0.51, 1.08)0.82 (0.55, 1.22)0.97 (0.65, 1.44)0.80 (0.53, 1.21)
Other deaths1.000.58 (0.35, 0.96)0.71 (0.41, 1.21)0.55 (0.31, 0.98)0.50 (0.27, 0.93)
Cancer recurrenceb1.000.96 (0.70, 1.32)0.93 (0.66, 1.32)0.97 (0.61, 1.25)0.65 (0.44, 0.97)

Figure 3. Estimated best fitting functions for the adjusted HRs of recreational PA in MET-hr/week versus no recreational activity (Reference = 0 MET-hr/week) according to the method of fractional polynomials. p-values refer to the recreational PA variable in the best fitting functional form in the adjusted Cox models, (a) regarding the four different endpoints, (b) regarding recurrence stratified by ER/PR status.

Download figure to PowerPoint

Table 3. Adjusteda hazard ratios and 95% confidence intervals for none versus any pre-diagnosis recreational physical activityb stratified by pre-diagnosis BMIc and smoking status
 Overall mortalityBreast cancer-specific deathsOther deaths
 Events (n)HR (none vs. any)pinteractionEvents (n)HR (none vs. any)pinteractionEvents (n)HR (none vs. any)pinteraction
  1. a

    Models were adjusted as described in Table 2.

  2. b

    Any sports or cycling between age 50 years and diagnosis.

  3. c

    Usual adult body mass index during age 25–50.

All3671.41 (1.12, 1.78) 2431.22 (0.91, 1.64) 1241.71 (1.16, 2.52) 
BMI < 252601.63 (1.23, 2.17)0.0631801.38 (0.97, 1.95)0.26802.10 (1.29, 3.41)0.10
BMI ≥ 251071.04 (0.67, 1.61) 630.90 (0.51, 1.62) 441.10 (0.55, 2.21) 
Never smoker1911.43 (1.03, 2.00)0.861351.55 (1.06, 2.28)0.050561.02 (0.53, 1.95)0.051
Ever smoker1761.33 (0.95, 1.87) 1080.81 (0.50, 1.30) 682.44 (1.46, 4.08) 

Regarding overall mortality, effect modification of the association with recreational PA (considering the best fitting functional form of PA, i.e. the dichotomized inactivity variable) by pre-diagnosis BMI (<25, ≥25 kg/m2) was observed (pinteraction = 0.063), showing a significant association with inactivity (HR = 1.63 (1.23, 2.17)) only among women with BMI < 25 (Table 3). Results for cause-specific deaths were similar (HR = 1.38 (0.97, 1.95) for breast cancer-specific mortality, HR = 2.10 (1.29, 3.41) for other deaths), but the interactions with BMI were not statistically significant (pinteraction = 0.26 and 0.10). Smoking status showed no effect modification on the association of PA with overall mortality. However, associations with cause-specific mortality tended to be modified (pinteraction = 0.050 and 0.051). While among never-smokers, physical inactivity increased breast cancer-specific mortality (HR = 1.55 (1.06, 2.28)) but showed no association with other deaths (HR = 1.02 (0.53, 1.95)), among current or ex-smokers inactivity was associated with increased non-breast cancer mortality (HR = 2.44 (1.46, 4.08)) rather than with breast cancer deaths (HR = 0.81 (0.50, 1.30)).

Regarding recurrence, there was significant effect modification of the association with recreational PA (considering the linear term as best fitting functional form) by ER/PR status of the primary tumor (pinteraction = 0.033). Stratified analyses yielded among ER−/PR− cases a HR (most active vs. inactive) of 0.53 (0.24, 1.16) and a significant linear dose-response (ptrend = 0.0045) compared to no clear association among ER+/PR+ cases with a HR (most active vs. inactive) of 0.77 (0.44, 1.35) and ptrend = 0.76 (Fig. 3b). There were no interactions with other tumor characteristics, BMI or smoking.

With regard to the other PA variables, no clear associations were found for total PA, and for walking, occupational or household activities, neither when all PA types were simultaneously included in the adjusted regression models nor when single PA variables were investigated. There were no significant trends, even when best functional forms were considered with the method of fractional polynomials. Regarding those PA variables categorized into quintiles, solely one significant HR was observed for overall mortality (HR = 0.62 (0.43, 0.91) for Q4 vs. Q1 of walking), but with HR = 0.97 (0.70, 1.34) for Q5 vs. Q1 there was no clear tendency. The results for recreational PA were not altered substantially when walking, occupational and household PA were additionally included in the models.


  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References

This large study following 3,393 non-metastatic breast cancer patients for a median of 5.6 years found markedly increased mortality for women who did not engage in any sports or cycling in the years before their breast cancer diagnosis compared to women who engaged in any sports or cycling. This detrimental effect was mainly attributed to death due to causes other than breast cancer, and there was no further risk reduction with increased activity levels. Regarding breast cancer-specific mortality the association with recreational PA showed a similar functional form but was far less pronounced. However, the associations between inactivity and overall or cause-specific mortality were observed only among patients who had a low or normal BMI in the years before diagnosis.

There was an inverse linear dose-response with recreational PA for cancer recurrence, which was significantly heterogeneous by ER/PR status. The data suggest that among breast cancer patients with ER−/PR− tumors, the hazard for cancer recurrence might be reduced by about 50% with higher amounts of pre-diagnosis PA (≥ 42 MET-hr/week).

Generally, it has to be discussed in which time period PA is assessed. Some studies considered PA before diagnosis,[15, 16, 18, 19, 30-32] others analyzed PA after diagnosis.[18, 20, 21, 32-34] Both approaches have strengths and limitations. One might expect that the time period after diagnosis may be more relevant for breast cancer progression and survival than the time before diagnosis with regard to PA. However, the activity behavior after diagnosis is likely influenced by the severity of disease and by cancer treatment which in-turn are associated with survival. Hence, inferences about causality in analyses with post-diagnosis PA on survival may be hampered by reverse causality.

On the other hand, PA may change due to the cancer diagnosis and treatment. Thus PA pre-diagnosis might not truly reflect PA post-diagnosis. Therefore, associations between PA and survival might be attenuated when considering pre-diagnosis PA. However, while PA typically drops extremely during chemo- and radiotherapy, activity levels seem to quickly recover, being similar to pre-diagnosis levels for many patients.[35-37] Although the breast cancer diagnosis motivate some patients to adopt a healthier lifestyle, women with a physically inactive lifestyle before diagnosis frequently do not manage to become more active after cancer diagnosis and treatment. Therefore, pre-diagnosis PA behavior can be considered as an acceptable indicator for PA in the cancer continuum. In particular, our own data collected at follow-up interview indicate that patients with higher pre-diagnosis PA were generally also more active one year after breast cancer surgery with a Spearman correlation of r = 0.50 (p < 0.0001).[37, 38] The median of pre-diagnosis recreational PA (sports and cycling) was 15.0 MET-hr/week (interquartile range 3.5–33.0) and was only slightly higher than the post-diagnosis median of 12.2 MET-hr/week (0.0–35.6). Among patients who were physically inactive before diagnosis (no sports or cycling), 68% remained inactive one year post-surgery and additional 15% became only modestly active (< 12 MET-hr/week). Only 9% became substantially active with ≥ 24 MET-hr/week of sports or cycling. However, these considerations include solely patients who completed the long-term follow-up interview, hence excluding patients who have died or were too ill to respond. Therefore, the true proportion of women who remained inactive after diagnosis is likely to be even larger.

No clear associations were observed with walking, occupational or household PA. Regarding walking, one reason might be, that many patients increased their time spent walking during and after cancer treatment,[38] thus the pre-diagnosis walking variable might not reflect walking behavior after diagnosis well. Likewise, occupational PA pre-diagnosis might have changed after diagnosis, because most patients were at least temporarily released from their jobs during cancer treatment. Household activity is very difficult to assess, thus potential moderate associations with mortality might be obscured by measurement error.

Our finding of a beneficial effect of recreational PA on overall survival after breast cancer is in line with previous studies.[15-21, 32, 34] In several of these studies, HRs also seem to be reduced for any activity, including the low PA levels, compared to no activity.[15, 17-21, 34] However, none of these studies statistically tested the functional form of the dose–response relationship. Generally, a linear trend was calculated if dose-response was investigated at all. Yet, our fractional polynomial analysis does not support a linear trend but rather increased mortality for inactivity vs. any activity.

Previous studies evaluating breast cancer-specific mortality yielded mixed results: some studies found no significant effect of PA,[30-32, 39, 40] while others observed a significant improvement of breast cancer survival by pre- or post-diagnosis PA.[15, 16, 18-21, 34] The reasons for the inconsistent results are unclear. In those studies with significant associations, the reported HRs indicate a detrimental effect of physical inactivity vs. any activity rather than a linear dose-response, which fits to our results based on the fractional polynomial analyses.

To our knowledge, no previous study in breast cancer patients reported the association between PA and deaths related to causes other than breast cancer, which comprise one-third of deaths in our study. Of those deaths 46.8% were due to neoplasms not related to the breast tumor and 27.4% were attributed to the circulatory system. A physically inactive lifestyle usually results in poor cardiorespiratory fitness. PA as well as cardiorespiratory fitness has been found to be inversely associated with all-cause mortality and cardiovascular events in large meta-analyses.[41, 42] In contrast to our observation of a nonlinear dose-response in breast cancer patients, those meta-analyses found inverse linear relationships between cardiorespiratory fitness or PA and overall mortality in the general population. It is unclear why neither our data nor other breast cancer studies show a linear dose-response. One potential explanation is that the cardio-toxic effect of many cancer therapies may influence the relationship between PA and mortality. Nevertheless, our data on pre-diagnosis activity levels indicated that even with a normal BMI and with a good cancer prognosis breast cancer patients may have a poorer survival when they are physically inactive or unfit compared to the other breast cancer patients. Our results strengthen the recommendations of the American College of Sports Medicine and the American Cancer Society, which stated that avoiding inactivity is likely helpful.[43, 44] The lack of a significant association between PA and mortality among overweight or obese women could be a consequence of the lower sample size in this subgroup, but could also be due to different hormone and cytokine levels in obese women. As adipose tissue can secrete a variety of inflammatory proteins (adipokines) and is the major source of steroid hormones in postmenopausal women, these risk factors might outrun the potential risk of physical inactivity.

Current or ex-smokers were at substantially increased risk for non-breast cancer mortality if they were not engaged in any sports or cycling in the years before diagnosis. Among smokers the proportion of non-breast cancer deaths on overall mortality were higher than among never-smokers (39% vs. 29%), and further physical deconditioning due to inactivity may have a more severe effect among smokers who are already at higher risk for many chronic diseases. However, the modification by smoking status of the associations between PA and cause-specific mortality needs further verification by other studies.

To our knowledge, only one study on pre-diagnosis life-time PA[15] and one study and a pooled analysis using data of four studies on post-diagnosis PA[20, 21] have investigated the effect on recurrence in breast cancer patients. While the pooled analysis found no association, the other two studies found a significant reduction, yet did not investigate the functional form of the dose–response relationship. We found a linear dose–response relationship of PA with relapse or progression only for ER−/PR− breast tumors. Biological mechanisms by which PA may affect cancer progression, and likewise explain the observed effect modification by ER/PR status, are largely unclear. In hormone receptor positive tumors hormonal parameters related to body fat distribution, diet or cancer medication might interfere or mask smaller effects of PA.

Strengths of our study are the large population-based patient sample with complete follow-up, verification of causes of death using death certificates and the consideration of cause-specific mortality differentiating between breast-cancer-specific deaths and other causes. A further strength is the investigation of the functional form of the dose-response using fractional polynomial analysis, as well as the consideration of all relevant prognostic factors.

In summary, our study suggests that breast cancer patients who were physically deconditioned before diagnosis have a significantly higher mortality than other breast cancer patients. This effect on mortality appeared to be mainly related to causes other than breast cancer. During a median follow-up of 5.6 years the hazard to die due to causes other than breast cancer was about twice as high in physically inactive women than in women who had been engaged in any recreational activity. However, there was no mortality reduction with increasing PA levels. In contrast, regarding breast cancer recurrence there was a linear inverse dose-response in patients with ER−/PR− tumors. Risk for recurrence was about 50% reduced in patients with ER−/PR− tumors who had been physically active pre-diagnosis, e.g. cycled or exercised at least 1 hr/day, compared to physically inactive women.

Thus, while breast cancer-specific mortality has decreased over the last decades, patients with a physically inactive lifestyle pre-diagnosis are at increased risk of premature death due to causes other than their breast cancer. Among patients with ER−/PR− breast tumors, higher levels of exercise may even reduce their risk of recurrence.


  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References

Authors thank U. Eilber, C. Krieg, S. Behrens, R. Birr, and T. Olchers for data collection and management.


  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References
  • 1
    Friedenreich CM. Physical activity and breast cancer: review of the epidemiologic evidence and biologic mechanisms. Recent Results Cancer Res 2011;188:12539.
  • 2
    McTiernan A, Tworoger SS, Ulrich CM, et al. Effect of exercise on serum estrogens in postmenopausal women: a 12-month randomized clinical trial. Cancer Res 2004;64:292328.
  • 3
    Monninkhof EM, Velthuis MJ, Peeters PH, et al. Effect of exercise on postmenopausal sex hormone levels and role of body fat: a randomized controlled trial. J Clin Oncol 2009;27:449299.
  • 4
    Friedenreich CM, Woolcott CG, McTiernan A, et al. Alberta physical activity and breast cancer prevention trial: sex hormone changes in a year-long exercise intervention among postmenopausal women. J Clin Oncol 2010;28:145866.
  • 5
    Liedtke S, Schmidt ME, Becker S, et al. Physical activity and endogenous sex hormones in postmenopausal women: to what extent are observed associations confounded or modified by BMI? Cancer Causes Control 2011;22:8189.
  • 6
    Mason C, Foster-Schubert KE, Imayama I, et al. Dietary weight loss and exercise effects on insulin resistance in postmenopausal women. Am J Prev Med 2011;41:36675.
  • 7
    Friedenreich CM, Neilson HK, Woolcott CG, et al. Changes in insulin resistance indicators, insulin-like growth factors, and adipokines in a year-long trial of aerobic exercise in postmenopausal women. Endocr Relat Cancer 2011;18:35769.
  • 8
    Friedenreich CM, Neilson HK, Woolcott CG, et al. Inflammatory marker changes in a year-long randomized exercise intervention trial among postmenopausal women. Cancer Prev Res (Phila) 2012;5:98108.
  • 9
    Campbell PT, Campbell KL, Wener MH, et al. A yearlong exercise intervention decreases CRP among obese postmenopausal women. Med Sci Sports Exerc 2009;41:153339.
  • 10
    Schmidt ME, Steindorf K, Mutschelknauss E, et al. Physical activity and postmenopausal breast cancer: effective periods in life and effect modification by different breast cancer characteristics. Cancer Epidemiol Biomarkers Prev 2008;17:340210.
  • 11
    Friedenreich CM, Courneya KS, Bryant HE. Influence of physical activity in different age and life periods on the risk of breast cancer. Epidemiology 2001;12:60412.
  • 12
    Moradi T, Nyren O, Zack M, et al. Breast cancer risk and lifetime leisure-time and occupational physical activity (Sweden). Cancer Causes Control 2000;11:52331.
  • 13
    Steindorf K, Ritte R, Tjonneland A, et al. Prospective Study on Physical Activity and Risk of In Situ Breast Cancer. Cancer Epidemiol Biomarkers Prev, 2012;21:2209–19.
  • 14
    Ibrahim EM, Al-Homaidh A. Physical activity and survival after breast cancer diagnosis: meta-analysis of published studies. Med Oncol 2011;28:75365.
  • 15
    Friedenreich CM, Gregory J, Kopciuk KA, et al. Prospective cohort study of lifetime physical activity and breast cancer survival. Int J Cancer 2009;124:195462.
  • 16
    West-Wright CN, Henderson KD, Sullivan-Halley J, et al. Long-term and recent recreational physical activity and survival after breast cancer: the California Teachers Study. Cancer Epidemiol Biomarkers Prev 2009;18:285159.
  • 17
    Keegan TH, Milne RL, Andrulis IL, et al. Past recreational physical activity, body size, and all-cause mortality following breast cancer diagnosis: results from the Breast Cancer Family Registry. Breast Cancer Res Treat 2010;123:53142.
  • 18
    Irwin ML, McTiernan A, Manson JE, et al. Physical activity and survival in postmenopausal women with breast cancer: results from the women's health initiative. Cancer Prev Res (Phila) 2011;4:52229.
  • 19
    Cleveland RJ, Eng SM, Stevens J, et al. Influence of prediagnostic recreational physical activity on survival from breast cancer. Eur J Cancer Prev 2012;21:4654.
  • 20
    Beasley JM, Kwan ML, Chen WY, et al. Meeting the physical activity guidelines and survival after breast cancer: findings from the after breast cancer pooling project. Breast Cancer Res Treat 2012;131:63743.
  • 21
    Holmes MD, Chen WY, Feskanich D, et al. Physical activity and survival after breast cancer diagnosis. JAMA 2005;293:247986.
  • 22
    Flesch-Janys D, Slanger T, Mutschelknauss E, et al. Risk of different histological types of postmenopausal breast cancer by type and regimen of menopausal hormone therapy. Int J Cancer 2008;123:93341.
  • 23
    Schmidt ME, Slanger T, Chang-Claude J, et al. Evaluation of a short retrospective questionnaire for physical activity in women. Eur J Epidemiol 2006;21:57585.
  • 24
    Ainsworth BE, Haskell WL, Whitt MC, et al. Compendium of physical activities: an update of activity codes and MET intensities. Med Sci Sports Exerc 2000;32:S498S504.
  • 25
    Ainsworth BE, Haskell WL, Leon AS, et al. Compendium of physical activities: classification of energy costs of human physical activities. Med Sci Sports Exerc 1993;25:7180.
  • 26
    Schemper M, Smith TL. A note on quantifying follow-up in studies of failure time. Control Clin Trials 1996;17:34346.
  • 27
    Textor J, Hardt J, Knuppel S. DAGitty: a graphical tool for analyzing causal diagrams. Epidemiology 2011;22:745.
  • 28
    Royston P, Ambler G, Sauerbrei W. The use of fractional polynomials to model continuous risk variables in epidemiology. Int J Epidemiol 1999;28:96474.
  • 29
    Allison PD. Survival analysis using the SAS system. Cary, NC: SAS institute Inc., 1995. 15557.
  • 30
    Rohan TE, Fu W, Hiller JE. Physical activity and survival from breast cancer. Eur J Cancer Prev 1995;4:41924.
  • 31
    Enger SM, Bernstein L. Exercise activity, body size and premenopausal breast cancer survival. Br J Cancer 2004;90:213841.
  • 32
    Irwin ML, Smith AW, McTiernan A, et al. Influence of pre- and postdiagnosis physical activity on mortality in breast cancer survivors: the health, eating, activity, and lifestyle study. J Clin Oncol 2008;26:395864.
  • 33
    Pierce JP, Stefanick ML, Flatt SW, et al. Greater survival after breast cancer in physically active women with high vegetable-fruit intake regardless of obesity. J Clin Oncol 2007;25:234551.
  • 34
    Holick CN, Newcomb PA, Trentham-Dietz A, et al. Physical activity and survival after diagnosis of invasive breast cancer. Cancer Epidemiol Biomarkers Prev 2008;17:37986.
  • 35
    Andrykowski MA, Beacham AO, Jacobsen PB. Prospective, longitudinal study of leisure-time exercise in women with early-stage breast cancer. Cancer Epidemiol Biomarkers Prev 2007;16:430438.
  • 36
    Emery CF, Yang HC, Frierson GM, et al. Determinants of physical activity among women treated for breast cancer in a 5-year longitudinal follow-up investigation. Psychooncology 2009;18:37786.
  • 37
    Huy C, Schmidt M, Flesch-Janys D, et al. Physical activity in a German breast cancer patient cohort: one-year trends and characteristics associated with change in activity level. Eur J Cancer 2012;48:297304.
  • 38
    Bock C, Schmidt ME, Vrieling A, et al. Walking, bicycling, and sports in postmenopausal breast cancer survivors-results from a German patient cohort study. Psychooncology, 2012 epub; doi:10.1002/pon.3134.
  • 39
    Borugian MJ, Sheps SB, Kim-Sing C, et al. Insulin, macronutrient intake, and physical activity: are potential indicators of insulin resistance associated with mortality from breast cancer? Cancer Epidemiol Biomarkers Prev 2004;13:116372.
  • 40
    Dal Maso L, Zucchetto A, Talamini R, et al. Effect of obesity and other lifestyle factors on mortality in women with breast cancer. Int J Cancer 2008;123:218894.
  • 41
    Kodama S, Saito K, Tanaka S, et al. Cardiorespiratory fitness as a quantitative predictor of all-cause mortality and cardiovascular events in healthy men and women: a meta-analysis. JAMA 2009;301:202435.
  • 42
    Samitz G, Egger M, Zwahlen M. Domains of physical activity and all-cause mortality: systematic review and dose-response meta-analysis of cohort studies. Int J Epidemiol 2011;40:1382400.
  • 43
    Schmitz KH, Courneya KS, Matthews C, et al. American College of Sports Medicine roundtable on exercise guidelines for cancer survivors. Med Sci Sports Exerc 2010;42:140926.
  • 44
    Rock CL, Doyle C, Demark-Wahnefried W, et al. Nutrition and physical activity guidelines for cancer survivors. CA Cancer J Clin 2012;62:24374.