Postdiagnosis weight change is associated with poorer survival in breast cancer survivors: A prospective population‐based patient cohort study

More women are surviving after breast cancer due to early detection and modern treatment strategies. Body weight also influences survival. We aimed to characterize associations between postdiagnosis weight change and prognosis in postmenopausal long‐term breast cancer survivors. We used data from a prospective population‐based patient cohort study (MARIE) conducted in two geographical regions of Germany. Breast cancer patients diagnosed 50 to 74 years of age with an incident invasive breast cancer or in situ tumor were recruited from 2002 to 2005 and followed up until June 2015. Baseline weight was ascertained at an in‐person interview at recruitment and follow‐up weight was ascertained by telephone interview in 2009. Delayed entry Cox proportional hazards regression was used to assess associations between relative weight change and all‐cause mortality, breast cancer mortality, and recurrence‐free survival. In total, 2216 patients were included. Compared to weight maintenance (within 5%), weight loss >10% increased risk of all‐cause mortality (HR 2.50, 95% CI 1.61, 3.88), breast cancer mortality (HR 3.07, 95% CI 1.69, 5.60) and less so of recurrence‐free survival (HR 1.43, 95% CI 0.87, 2.36). Large weight gain of >10% also increased all‐cause mortality (HR 1.64, 95% CI 1.02, 2.62) and breast cancer mortality (HR 2.25, 95% CI 1.25, 4.04). Weight maintenance for up to 5 years in long‐term breast cancer survivors may help improve survival and prognosis. Postdiagnosis fluctuations in body weight of greater than 10% may lead to increased mortality. Survivors should be recommended to avoid large deviations in body weight from diagnosis onwards to maintain health and prolong life.


| INTRODUCTION
Improvements in early detection, and personalized and targeted treatments have led to more women surviving after breast cancer. 1,2 Modifiable factors such as body weight can also influence breast cancer prognosis. 3 Results that include evidence from a large systematic review and meta-analysis point toward increasing risk of total mortality, breast cancer mortality and risk of developing a second primary breast cancer with obesity or excess body weight both before and after breast cancer diagnosis. 3 Weight change is common during breast cancer. It may be a consequence of breast cancer or a combination of breast cancer and anticancer treatments such as surgery, chemo, radiation and/or hormonal therapies or combinations of these, which, while essential for improving survival, contribute considerably to therapy-associated side effects. [4][5][6] There is accumulating evidence that weight gained in different periods during the breast cancer trajectory-during adulthood, from prediagnosis to postdiagnosis breast cancer, from pretreatment to posttreatment and during treatment-may impact adversely on total mortality, although there was large heterogeneity between studies in metaanalyses. 3,7 Similar but weaker patterns of association between weight gain and breast cancer-specific mortality have also been observed, but there are fewer investigations between associations with recurrence. 3,7 Although the bulk of weight change and prognosis research has focused on weight gain, results of a systematic review and meta-analysis conducted in six studies-with significant heterogeneity-revealed prediagnosis to postdiagnosis weight loss (highest vs lowest/stable) to be strongly associated with all-cause mortality. 3 Collectively then, limited evidence suggests that weight maintenance may be optimal for health and survival following breast cancer. Sources of heterogeneity between studies can be mainly attributed to differences in study designs that affect timing and duration of weight measurements in relation to diagnosis and treatment. Indeed, time since diagnosis may influence the associations between weight change and survival. 8 Risk of breast cancer and other chronic diseases increases with age. Therefore, at diagnosis and during survivorship, many breast cancer survivors have other chronic comorbid conditions that also affect survival. 9,10 With this in mind, we aim to describe the associations between weight change after diagnosis and subsequent prognosis, including recurrence, in a population-based patient cohort of long-term breast cancer survivors in Germany. We further explore whether these associations differ by weight at recruitment, and number of comorbidities, and whether the rate at which weight changes also impacts survival. To the best of our knowledge, this is the first study to evaluate postdiagnosis weight change and prognosis within a European setting.

| Study population
We used data from the MARIE (Mammary Carcinoma Risk Factor Investigation) study, 11  The primary exposure of interest was weight change, so women who completed baseline and follow-up interviews (n = 2542) were included. After exclusion of patients who were premenopausal (n = 148), had metastases at diagnosis (n = 22), tumors other than breast cancer or nonmelanoma skin cancer before diagnosis (n = 138), missing baseline (n = 2) or follow-up weight (n = 16), there were 2216 patients available for analyses of all-cause and breast cancer-specific mortality ( Figure 1). For recurrence-free survival, women who experienced a recurrence prior to the follow-up were additionally excluded leaving 2068 women for recurrence-free survival analyses.

| Assessment of weight and other exposures
At baseline interview (median 3.9 months after diagnosis), selfreported current body weight in kg and height in cm were recalled in an in-person interview. In 2009 (median 5.8 years after diagnosis), self-reported current height and body weight were ascertained at the follow-up telephone interview. Clinical and pathological characteristics were obtained from hospital and pathology records. Information on lifestyle, socioeconomic and demographic, comorbidities and other pertinent protective and risk factors were collected from baseline and follow-up interviews.

What's new?
Weight change, while common among breast cancer survivors, may not be optimal for survival. Here, the authors offer the first evaluation of the associations between post-diagnosis weight change and subsequent prognosis in postmenopausal long-term breast cancer survivors within a European setting. The results show that weight maintenance for up to 5 years after diagnosis is associated with better survival and prognosis, while fluctuations of more than 10% of body weight are associated with increased mortality. Encouraging breast cancer survivors to maintain weight after diagnosis may possibly help them to maintain a healthy and prolonged life.

| Outcome assessment
Participant vital status was determined through population registries of the study regions up until the end of June 2015, and all deaths were verified by death certificates. Self-reported recurrences of the primary breast cancer, second cancers and metastatic events were identified during 2009 and 2015 telephone interviews with patients and verified by clinical records or with treating physicians. For patients who died, medical records were checked or treating physicians were contacted. Primary outcomes were all-cause mortality (death from any cause) and breast cancer-specific mortality (death from breast cancer). Recurrence-free survival (including ipsilateral/ local/regional invasive recurrence, distant recurrence and metastases occurring after primary diagnosis, death) 13 was a secondary outcome.
Participants without events of interest were censored at date of last contact or June 30, 2015, whichever came first.

| Statistical analysis
Relative weight change was calculated using [(follow-up weightbaseline weight)/baseline weight] × 100. To assess weight change, five categories were created: weight stable (weight change within 5% from baseline to follow-up), moderate gain/loss (weight gain/loss of ≥5% to ≤10%), large gain/loss (weight gain/loss of >10%), respectively. 7 These categories were selected for comparisons with other studies 7 and considered to be clinically meaningful. 14 To estimate hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) for associations between weight change with allcause mortality, breast cancer mortality and recurrence-free survival, delayed-entry Cox proportional hazards regression models from follow-up interview until event of interest/censoring were constructed. Time-to-event started from date of diagnosis, and time-atrisk started from the date of follow-up interview. Weight stable women served as reference. The proportional hazards assumption was examined using a weighted least-squares line fitted to the plots of scaled Schoenfeld residuals. 15 No indication for violation of the proportional hazards assumption was found. Follow-up time was calculated using reverse Kaplan-Meier. 16 Subgroup analyses were defined a priori, and effect modification was tested applying the likelihood ratio test to a model with the interaction term of the main exposure and potential modifier and a model without the interaction term. For this purpose, five weight change categories were collapsed into three (weight stable (weight change within 5%), weight loss (weight change ≥5%), and weight gain (weight change ≥5%)). Associations between weight change and outcomes by baseline BMI (18.5 to <25.0 kg/m 2 /≥25 kg/m 2 ), and comorbidities using the Charlson Comorbidity Index (CCI 0-1/≥2) were evaluated. The CCI was built from 17 conditions 17,18 and adapted. The CCI at follow-up was used, as new comorbidities may have been acquired after baseline.
The rate of weight change (% weight change per 1 year) was also assessed to facilitate comparisons between our results and those of other studies, where there was large variation in the duration or window in which weight was gained or lost. 3,7 To calculate rate of weight change, percent weight change was divided by time between baseline and follow-up in years. Five categories for rate of weight change were constructed: weight maintenance (within 0.5% per year), slow weight gain/loss (≥0.5% to ≤1.0% per year) and fast gain/loss (>1.0% per year). Relative weight change was also modeled continuously using absolute values in all women and also allowing for different slopes in those who gained or lost ≥5% body weight. A model was fit using an F I G U R E 1 Flow chart of inclusion and exclusion criteria for participants of the MARIE study for analyses relating to changes in weight and all-cause mortality, breast cancer mortality and recurrence-free survival T A B L E 1 Characteristics of 2216 postmenopausal women diagnosed with a first primary breast cancer according to relative weight change   interaction term between weight change as a continuous variable and a dummy for ≥5% weight gained or lost.
In addition, associations between baseline weight (per 5 kg increase) as well as associations between follow-up weight (per 5 kg increase) and the three endpoints were investigated separately. Associations between follow-up weight and cancer outcomes were stratified by baseline BMI (normal/overweight and obese).
All models included the prognostic factors tumor size, nodal status, tumor grade, hormone receptor (HR) status, as well as mode of tumor detection and were stratified by study center and age at diagnosis in 5-year categories to allow for variation in baseline hazard. In sensitivity analyses, all analyses were repeated for all three outcomes, excluding (1) women who developed a recurrence (ipsilateral, local/regional, distant and metastatic recurrence or second tumor) by the first follow-up interview (n = 117), and (2) women with in situ tumors (n = 127).
For all analyses, complete-case analysis was performed, as the proportion of missing was less than 5% for all variables. All tests of statistical significance were two-sided and significance level was set to 0.05. Analyses were conducted using the SAS statistical software package (version 9.4).  (HR (95% CI): 1.83 (1.27, 2.63)) and breast cancer-specific (HR (95% CI): 1.82 (1.07, 3.10)) mortality. There were also possible suggestions that fast weight gain were associated with all-cause (HR (95% CI):
Higher baseline weight was associated with increasing all-cause To better understand how various therapies may affect baseline weight, we compared baseline weight and BMI in the whole population against different subsets of the study population: in women who never/before or during baseline/after baseline received aromatase inhibitor therapy, tamoxifen, either aromatase or tamoxifen, chemotherapy, radiation therapy, mastectomy and breast-conserving therapy. We did not find meaningful differences in weight or BMI at recruitment between any of these subsets (data not shown).

| DISCUSSION
Postdiagnosis weight change in relation to prognosis was evaluated in 2216 postmenopausal long-term breast cancer survivors in Germany.
Weight loss >10% of body weight was associated with poorer prognosis compared to weight maintenance. The increased mortality associated with weight loss was independent of baseline BMI, and more pronounced in those with severe comorbidities, who were more likely to be negatively impacted by weight loss. Per percent increments of postdiagnosis weight from baseline to the follow-up was likewise associated with poorer subsequent prognosis. That we see stronger associations and a dose-response relationship with weight loss could indicate that "reserves" are necessary for health and to possibly withstand metabolic challenges from breast cancer and its sequelae. In further support that "reserves" may be necessary after breast cancer, we To date, three systematic reviews 3,7,22 and two meta-analyses 3,7 have examined weight change and prognosis after breast cancer. One of these evaluated weight gain and found that only those who gained >10% were at increased risk of all-cause mortality, 7  Furthermore, large weight losses of >10% were associated with poorer recurrence-free survival, which was similar to results from one study. 27 These results were not supported by another study that assessed BMI change stratified by smoking status, 26 which may possibly account for the heterogeneous findings. Recurrence rather than recurrence-free survival was the endpoint of interest in these two studies. 26,27 Results from a meta-analysis of three studies did not indicate an association between weight gain and recurrence. 7 Similarly, we observed no association between weight gain and recurrence-free survival. An additional challenge of summarizing associations with recurrence is inconsistent definitions of recurrence; we have used those defined by Hudis. 13 Of all studies identified in the systematic literature reviews and included in the meta-analyses on postdiagnosis weight change and prognosis, none were European. Yet we are aware of three studies from Europe, two of which assessed weight gain during adulthood and total and breast cancer-specific mortality 28,29 and one that examined weight change during chemotherapy and total mortality and disease-free survival. 30 We are not aware of any studies  31,32 In this analysis, however, relationships between weight change and cancer endpoints were evaluated, so systematic underestimation or overestimation of weight at both time points in the same direction may be less likely to bias results than if evaluating weight alone.
Weight underestimation and overestimation by overweight and underweight women, respectively, to normal weight would also dilute observed associations, thus true associations could likely be stronger.
Also, though comorbidities used to generate the CCI were selfreported and unverified, use of patient questionnaires to ascertain comorbidities has been shown to be reliable. 33 Reverse causation is possible if women loss or gain weight because they are sicker than those who are weight stable. Although tumor size, nodal status and grade were similar between weight change groups, women severely impacted by comorbidities were apparently more likely to have poorer prognosis from weight loss than women who had no or mild comorbidities.
Study strengths include longitudinal follow-up data from breast cancer survivors over more than 10 years. Most studies on weight gain or loss have examined weight change before (as prediagnosis or usual weight) and 12 months or more after diagnosis or treatment 3,7 with time between measurements a median 1.5 years, 7 during which time patients may still be undergoing treatment or may still be experiencing the effects of treatment. Duration between the two weight measurements in our study was median 5.1 years, so weight change due to acute effects of different treatment regimens could possibly have been circumvented. Weight at follow-up may better indicate long-term weight, when acute sequelae of breast cancer or behavioral changes related to initial diagnosis or treatment may have abated or plateaued. 34 Information on numerous important potential confounding factors were collected and tested in our comparison models. In the current analysis, we only included women for whom we had both weight measures-at baseline (between 2002 and 2005) and follow-up (2009)-so these women may reflect a healthier subset of women, as they survived at least 5 years after the initial breast cancer. To the best of our knowledge, this is the largest analysis to elucidate associations between postdiagnosis weight change and prognosis in a European population. Specifically, our results further contribute to the literature by demonstrating evidence for the first time in our study sample that weight loss, particularly weight that is lost quickly, may be associated with higher risk of recurrence in breast cancer survivors.

| CONCLUSIONS
In conclusion, our results are the first within a European setting to underline the importance of weight maintenance for up to 5 years after diagnosis in long-term breast cancer survivors for the benefit of survival and prognosis after diagnosis. Survivors should strive to avoid large changes in weight, especially in a short period of time, from diagnosis onwards to maintain health and prolong life.

ACKNOWLEDGEMENTS
We are grateful to all the MARIE study participants. We thank U.
Eilber for her most valuable technical assistance and data management. We express our deep gratitude to Dieter Flesch-Janys for his invaluable contributions to the MARIE projects throughout all these years as PI of the Hamburg study region and close collaborator.

CONFLICT OF INTEREST
Authors declare that they have no conflicts of interest.

DATA AVAILABILITY STATEMENT
The data can be made available upon reasonable request to the principal investigator(s) (J. C.-C. and H. B.).

ETHICS STATEMENT
The study was approved by the ethics committees of the University of Heidelberg, the State of Rhineland-Palatinate and the Hamburg Medical Council, and was conducted in accordance with the