The impact of body size changes on recurrence risk depends on age and estrogen receptor status in primary breast cancer

To investigate the prognostic impact of body size changes during the first postoperative year in breast cancer. A cohort of 1,317 primary breast cancer patients included in Sweden (2002–2014) underwent body size measurements at the preoperative and 1-year visits (n = 1,178). Landmark survival analyses were used to investigate how postoperative weight gain or loss (> 5%) or change in waist–hip ratio (WHR) categories (≤ 0.85 or > 0.85) impact prognosis. Median age at inclusion was 61 years and body mass index 25.1 kg/m2. After a median follow-up of 5.0 years from inclusion, 165 recurrences and 77 deaths occurred. Weight gain (17.0%) conferred over twofold recurrence risk only in patients < 50 years (Pinteraction = 0.033). Weight loss (8.6%) was only associated with a poor prognosis in patients ≥ 70 years, but not after restriction analysis. Weight change did not impact prognosis in patients 50 to < 70 years. Changes between WHR categories were associated with differential recurrence risk depending on estrogen receptor (ER) status (Pinteraction = 0.007), with higher recurrence risk in patients with ER+ tumors and lower recurrence risk with ER− tumors. Both changes in terms of weight and WHR category yielded independent prognostic information. Further research is imperative before recommending weight loss for all overweight breast cancer patients.

Weight change following breast cancer diagnosis and treatment is common but might differ according to patients' age and treatments [6,[20][21][22]. A possible U-shaped relationship with increased risk for both weight gain and loss has been suggested [7,9,12,15,16,18]. Adipose tissue is biologically active, and multiple pro-carcinogenic processes associated with weight gain have been identified [3]. Less is known about the mechanisms behind the poor breast cancer prognosis associated with weight loss in observational studies [6,7,9,11,12,15,16,18].
Central obesity is associated with metabolic syndrome [23], altered estrogen/testosterone ratio [24], overall mortality [25], and cancer survival [26]. Central obesity in breast cancer patients has been linked to all-cause and breast cancer-specific mortality [27], but associations might differ between ethnic groups [28].
Waist-hip ratio (WHR) can be used as surrogate measure of central obesity [29]. Recently, the anthropometric index 'a body shape index' (ABSI) was introduced, which is based 1 3 on waist circumference (WC) adjusted for body mass index (BMI) and height [30]. Higher ABSI was associated with mortality risk in the general population [30][31][32][33], but breast cancer recurrence and death due to other causes involve different pathways. The prognostic impact of ABSI in breast cancer is unknown.
We hypothesized that a stable body constitution during the first postoperative year after diagnosis would be associated with a favorable prognosis, but that associations might differ between subgroups of patients. The study aimed to investigate the prognostic impact of anthropometric changes from the preoperative visit to the 1-year postoperative visit in breast cancer patients.

Methods
Primary breast cancer patients assessed preoperatively since October 2002 at the Skåne University Hospital in Lund were asked to join a prospective cohort study regarding lifestyle factors, as previously described [5]. The preoperative visit typically took place a few weeks after diagnosis. The study was approved by the Lund University Ethics Committee (LU75-02, LU37-08, LU658-09, LU58-12, LU379-12, LU227-13, LU227-15, and LU458-15), and written informed consents were collected from all participants. By 31 December 2014, 1,578 patients were included in the cohort of whom 1,317 were included in the present study, after exclusion of patients without recorded weight or height, with carcinoma in situ, any preoperative treatment, or early metastases < 0.3 years. The associations between changes in anthropometric measures and prognosis were analyzed in 1,178 patients who were recurrence-free and had a 1-year visit with measured weight 0.9-1.5 years from inclusion ( Fig. 1).
Patient characteristics, including weight, height, and waist and hip circumference, were measured by trained research nurses at the preoperative visit. These variables, with the exception for height, were measured once more at the 1-year visit. WC was measured at the umbilicus, and hip circumference was measured at the widest part between the hip and the major trochanter. The patients were also asked to complete a questionnaire, as previously described [34].
Tumor characteristics were obtained from each patient's pathology report, regarding invasive tumor size, axillary nodal involvement, histological grade, estrogen receptor (ER), and progesterone receptor (PR) status. All tumors were analyzed at the Department of Pathology at Skåne University Hospital in Lund as previously described [34,35]. Tumors with > 10% positive nuclear staining were considered receptor positive.
Adjuvant treatments, including chemotherapy, radiotherapy, tamoxifen, and aromatase inhibitors (AIs), were prescribed according to standard of care. Treatment and additional clinical data, recurrences, and new breast cancers were collected from patient charts and follow-up questionnaires. Only breast cancer treatment before any breast cancer event was considered in the analyses. Date of death was obtained from the Swedish population registry. Data on events and new cancers were also collected from the regional tumor registry, which is close to 100% complete.

Statistical analyses
Statistical analyses were performed with the software SPSS 24.0 (IBM). Changes in weight were expressed as percentage differences between inclusion at the preoperative visit and the 1-year visit and categorized as 'weight loss' (> 5% loss), 'stable weight' (≤ 5% change), or 'weight gain' (> 5% gain). Similarly, changes between inclusion and the 2-year visit were calculated. The 5% cut-off was chosen according to standard recommendations to reduce the risk of weightrelated diseases and cancer [36,37]. WHR was calculated as the ratio between the WC and the hip circumference. WHO recommends that a woman's WHR is ≤ 0.85 [38]. Patients were stratified into four groups according to whether they remained or changed category (≤ 0.85 or > 0.85). Weight change and WHR categories were analyzed in relation to preoperative patient characteristics with non-parametric tests (Kruskal-Wallis) since several variables were not normally distributed. The median and interquartile ranges (IQR) are presented. Chi-square tests were used to analyze categorical variables in relation to weight change and WHR categories.
ABSI was calculated using the definition WC/ (BMI 2/3 × height 1/2 ) with WC and height expressed in meters [30]. Patients' ABSI values were converted to Z-scores and divided into quintiles. The difference between ABSI at inclusion and ABSI at the 1-year visit was then calculated and converted into Z-scores. Patients with an ABSI difference Z-score in the 2nd to 4th quintile were considered stable, while patients with a difference in the 1st quintile were considered as having downward change in ABSI, and patients with a difference in the 5th quintile of upward change.
Breast cancer events included ipsilateral, contralateral, and regional recurrences, as well as distant metastasis. Breast cancer-free interval was censored at the time of a non-breast cancer-related death or at the time of the last follow-up prior to 1 July 2016 for patients who were alive and event-free prior to this date. Among the 1,317 patients, a total of 165 patients had any recurrence, of which 104 were distant metastasis, and 77 patients with a prior recurrence died. Of the 1,152 patients without recurrence, 59 died. In the subgroup of 1,178 patients analyzed for anthropometric changes, 138 recurrences occurred (87 distant), and 62 died.

3
Of the 1,040 patients without breast cancer events, 47 died. Overall survival was calculated until death due to any cause.
For all variables indicating changes, including weight (gain or loss versus stable weight), WHR category changes (over the > 0.85 cut-off versus remaining in the same category), or ABSI change (upwards or downwards versus 2-4 quintile), landmark analyses were performed where the event-free interval was calculated from the date of the 1-year follow-up visit. Univariable survival analyses (Kaplan-Meier) were performed. Cox regressions were used to calculate adjusted Hazard Ratios (HR) with 95% confidence intervals (CI) and adjusted p values. All models were adjusted for invasive tumor size (> 20 mm or skin/muscular involvement), any axillary nodal involvement, histological grade III, ER status, age at diagnosis (continuous), and the natural logarithm (ln) of preoperative weight. Adjustment for the preoperative ABSI Z-score was included in models investigating impact of ABSI changes. Further adjustments were made for any alcohol consumption and smoking or treatments, including a history of chemotherapy, radiotherapy, tamoxifen, and AI prior to any event.
To examine whether there were effect modifications for preoperative BMI ≥ 25 kg/m 2 , age ≥ 50 or ≥ 70 years at inclusion, tumor characteristics (tumor size > 20 mm or skin/muscular involvement, any axillary nodal involvement, histological grade III, or ER status), and treatments (chemotherapy, for changes in weight and body composition in relation to prognosis radiotherapy, tamoxifen, or AI) on the associations between weight change categories and prognosis, multiplicative interaction variables between these factors and the categories 'weight loss' and 'weight gain' were calculated. To examine whether there were effect modifications of tumor characteristics or treatments, multiplicative interaction variables between these factors and the WHR categories were calculated. All p values were two-sided, and each p value should be interpreted as the level of evidence against each null hypothesis. Since this is an exploratory study, nominal p values are presented without adjustments for multiple testing.

Patient characteristics
Patient characteristics and clinical data are presented in Table 1. In the group of patients who lost weight, the median weight loss was − 7.8%, while the median weight gain was + 7.7% in the group that gained weight. As there were only 35 patients (3.0%) who lost > 10% and only 44 patients (3.7%) who gained > 10% of their preoperative weight, no further subdivision of weight change categories was performed. The median age at inclusion was 61 years (range 24-87 years). Preoperatively, patients who lost weight were older, weighed more, had greater WHR at inclusion, while patients who gained weight were younger, weighed less, and had lower WHR (all P s ≤ 0.043). History of chemotherapy was less common among patients who had a stable weight (p < 0.0001) and use of AI but no tamoxifen by the 1-year visit was less common among patients who gained weight (p = 0.006). Follow-up with body measurements was available for 1,041 patients at the 2-year visit. The majority (76.8%) remained in the same weight change category as for the 1-year visit. All further analyses are therefore based on the 1-year visit to optimize follow-up time.

Tumor characteristics
Tumor characteristics are presented in Table 2. The patients who lost weight had more axillary node involvement and a higher frequency of histological grade III tumors, thus more aggressive tumors. The patients who had stable weight had a higher proportion of ER+ and PR+ tumors (all p s ≤ 0.049).

Breast cancer-free interval
Patients were followed for up to 13 years from inclusion. The 1,317 patients aged 24-99 years were included and analyzed for preoperative weight and body composition in relation to the risk of breast cancer events; the median follow-up time was 5.0 years (IQR 3.0-7.2) for the 1,093 patients still at risk. Of the 165 breast cancer events, 33 occurred in patients with ER− disease, 132 in patients with ER+, while ER status was missing for one patient. Overall, preoperative (ln) weight or a preoperative BMI ≥ 25 kg/m 2 were associated with recurrence risk (p = 0.024 and p = 0.013). The level of evidence became somewhat weaker after adjustments for age and tumor characteristics and when only the 1,178 patients analyzed for changes in body measures were included.
Overall, there was no association between weight change categories and recurrence risk (LogRank p = 0.35). To elucidate whether any association between weight change and risk for events was present in subgroups of patients, exploratory analyses of effect modification according to preoperative BMI, age, tumor characteristics, and treatments were performed.

Age and weight change in relation to prognosis
Effect modifications between age and weight changes were detected. Patients were stratified into three age groups (< 50 years/50 to < 70 years/ ≥ 70 years; Fig. 2). Age < 50 years was used as a proxy for premenopausal status. An interaction was found between age ≥ 50 years and weight gain (adjusted P interaction = 0.033), but not with weight loss, on recurrence risk. The patients were stratified by age ≥ 50 years. Patients < 50 years who gained weight had a twofold recurrence risk compared to other patients, HR adj 2.64 (95% CI 1.28-5.45), while weight loss was not associated with recurrences (P adj = 0.72). In patients ≥ 50 years, no association was found between recurrences and weight change (both P adj ≥ 0.84).
When applying a different age cut-off (≥70 years) to test whether there were effect modifications of old age on recurrence risk, an interaction was found between weight loss (adjusted P interaction = 0.021), but not with weight gain. Patients ≥ 70 years who lost weight had a threefold recurrence risk, HR adj 3.47 (95% CI 1.06-11.33), while weight gain was not associated with recurrences. In women < 70 years, no association was found between weight change and recurrences (P adj ≥ 0.22).
No effect modifications of age were found for weight gain or loss with respect to death with either of the cut-offs (≥ 50 or ≥ 70 years). However, the oldest patients (≥ 70 years) that lost weight also had an increased risk of death, HR adj 3.56 (95% CI 1.50-8.44). Eight patients in different age groups died within one year of the 1-year visit. When these patients were excluded, there was no association between weight loss and recurrence risk or death in the oldest age group. Weight change did not impact either recurrence risk or death in patients 50-< 70 years.

Preoperative BMI, tumor characteristics, treatments, and weight changes in relation to prognosis
There was no effect modification of preoperative BMI on the association between the weight change categories 'weight gain' and 'weight loss' and risk of recurrence or death (all adjusted P interactions ≥ 0.39). Furthermore, there were no effect modifications between tumor characteristics or treatments on the associations between weight gain or loss and recurrences or death (all adjusted P interactions ≥ 0.096).

Preoperative WHR and changes during the first postoperative year in relation to prognosis
For the 1,291 patients with complete preoperative body measurements, recurrence risk was not associated with Overall, there was a tendency towards lower recurrence risk in the two stable WHR groups compared to the two groups that changed the WHR category (Fig. 3a). The association was somewhat weaker if 'weight gain' and 'weight loss' were also adjusted for.
Since both the curves for patients who remained in the same WHR category between preoperative and 1-year visit (≤ 0.85 or > 0.85) and both the curves for patients who changed category (increased or decreased) crossed over, these groups were combined into two groups of patients who either remained in the same or changed WHR category  (Fig. 3b). When comparing stable versus changed WHR category groups, a substantially lower recurrence risk in the group of patients who remained in the same category was found compared to those who changed category, HR adj 0.62 (95% CI 0.43-0.90). Results were essentially the same if 'weight gain' and 'weight loss' were also adjusted for. Regarding overall survival, when comparing the four groups of patients who either remained in the same or changed WHR category, a WHR > 0.85 at both visits conferred the highest risk of death (LogRank p = 0.014; HR adj 1.60 (95% CI 0.98-2.59)) when the group with ≤ 0.85 at both visits was used as a reference. When comparing the two groups of patients who remained in the same WHR category with those who changed category, no difference was found. The results were essentially the same after further adjustment for 'weight gain' and 'weight loss.'

ER status and WHR category changes versus stable WHR in relation to prognosis
A strong effect modification was found between ER status and changes in WHR category on the recurrence risk (adjusted P interaction = 0.007). Patients with ER+ tumors who remained in the same WHR category had a lower recurrence risk than those who changed category, HR adj 0.44 (95% CI 0.30-0.66). In contrast, patients with ER− tumors that remained in the same WHR category had a higher recurrence risk than those who changed category, HR adj 7.30 (95% CI 0.98-54.08; Fig. 3c-f). No interaction was found between ER status and changes in WHR on overall survival (P interaction = 0.22). Further adjustments for smoking and alcohol abstinence or treatments revealed essentially the same results.

Tumor characteristics, treatment, and WHR category changes versus stable WHR in relation to prognosis
Except for ER status, no effect modifications were found between tumor characteristics and WHR category change on recurrence risk or death (all P interactions ≥ 0.086). Similarly, there were no effect modifications between radiotherapy or chemotherapy and WHR category change on recurrence risk or death. Among patients with ER+ tumors, no effect modification was found between ever use of AIs and WHR category change (all adjusted P interactions ≥ 0.23). There was weak evidence of an effect modification between ever use of tamoxifen and WHR category change on the recurrence risk (adjusted P interaction = 0.052) but not death (adjusted P interaction = 0.32).

Preoperative ABSI and postoperative changes in relation to prognosis
ABSI was calculated for 1,291 patients with complete preoperative body measurements. Overall survival was strongly associated with preoperative ABSI (HR adj 1.26 (95% CI 1.10-1.43) per quintile. Preoperative ABSI was not associated with recurrence risk (LogRank P trend = 0.84). The results were essentially the same for the 1,153 patients analyzed for changes in ABSI.
Changes in ABSI, either downwards (n = 230) or upwards (n = 230), were not associated with recurrence risk or death when compared to patients with stable ABSI (n = 693). No further analysis of ABSI was conducted.

Discussion
In this study, 1-year postoperative changes in weight and WHR in breast cancer patients impacted differently on prognosis depending on age and tumor ER status. However, the association between changes in anthropometric measures and clinical outcome was not dependent on overweight. In younger patients, weight gain was associated with higher recurrence risk. Changes between WHR categories were associated with higher recurrence risk in patients with ER+ tumors and lower recurrence risk in patients with ER− tumors. High preoperative ABSI was strongly associated with shorter overall survival but not with recurrence risk.
In the present study, patients < 50 years were most likely to gain weight, which was associated with a higher recurrence risk. This is consistent with a recent scoping review, which concluded that younger chemotherapy-treated premenopausal patients were particularly at risk for weight gain [22]. One previous study showed a higher recurrence risk in premenopausal patients [10]. Another study compared the impact of BMI changes in clinical trials and found increases in BMI by 1 year to negatively impact outcomes only in pre-/ peri-/early post-menopausal trial patients [19]. However, we found no benefit of weight loss in this age group and retaining preoperative weight might be most beneficial.
The oldest patients (≥ 70 years) that lost weight had a poorer prognosis compared to other old patients. However, this was no longer evident after exclusion of patients that had died within one year after the 1-year postoperative visit, which suggests reverse causality and a symptom of an underlying disease where frailty and muscle wasting might be present. Multiple observational studies have reported poor breast cancer prognosis associated with weight loss [6,7,9,11,12,15,16,18]. However, a systematic review indicated a considerable risk of bias in some studies because of lack of information on whether the weight loss was intentional or not [39]. No association was found between weight changes and recurrences or death in the age group 50 to < 70 years, which might reflect the mix of peri-and early post-menopausal patients. Further, changes in weight and BMI do not necessarily reflect body fat distribution, which can be better measured by other methods. Individuals with a normal weight and BMI might still have an unhealthy metabolic body composition with excess body fat, especially central accumulation, referred to as sarcopenic obesity [40]. In the present study, measurements of WHR and the new index ABSI were used to capture central body fat accumulation. Patients with preoperative WHR > 0.85 had a poorer overall survival, which is in line with most [27,[41][42][43], but not all [12,44] studies. The change in WHR category conferred prognostic impact with strong effect modification with ER status independent of weight change in the present study. To the authors' knowledge, this is the first study to investigate the prognostic impact of early changes in anthropometric measures in breast cancer patients, and it is unknown whether this difference reflects a change in hormonal profile in endocrinetreated patients or impacts therapy response to one or more treatment types. The median follow-up was only five years from inclusion. Most recurrences occur within five years in patients with ER− tumors, while ER+ tumors tend to recur over a longer period of time.
Different treatments could affect both body composition and prognosis [45]. Smoking also impacts body composition in young women [46], as well as the clinical outcome after endocrine therapy [47]. Smoking and alcohol consumption or treatments were included in separate adjustment models but did not explain the results. Menopause occurred in some patients either as result of age or treatment, which might have affected body constitution [48].
We hypothesize that the different prognostic impacts of weight and WHR changes according to age and tumor ER status in part reflect shifts in the inflammatory and hormonal profile caused by changes in the central adipose tissue. The release of free fatty acids, particularly associated with abdominal visceral adipocytes [29], activates the nuclear factor kappa B (NF-κB) pathway [3]. The upregulation of inflammatory cytokines and subsequent increases in aromatase activity and estrogen synthesis might stimulate ER+ breast cancer cells [3]. The generation of crown-like structures associated with inflammation and increased aromatase activity might also play a role [49]. Increased levels of insulin and insulin-like growth factors (IGFs), particularly associated with central obesity [50], also increase the bioavailability of estrogens by downregulation of sex hormonebinding globulin (SHBG) [51]. Surprisingly, weight loss has been associated with increase in IGF-1 levels in postmenopausal women in contrast to findings in animal studies [52], reflecting the complexity of the pathways involved in changes in body measures. However, factors that promote tumor growth may not affect prognosis in patients who died from other causes and it was also observed that changes in anthropometric measures impacted recurrence risk and death somewhat differently.
Our study confirmed previous findings that higher preoperative ABSI was strongly associated with overall survival but no association with breast cancer recurrences was found. The association with overall survival was evident in patients with and without a recurrence but was stronger in patients without a previous recurrence (data not shown). A previous study reported that higher ABSI increased cancer mortality in women [32]. Raising ABSI over a 7-year period was associated with greater mortality [31]. In the present study, however, 1-year changes in ABSI were not associated with recurrence risk or death.
A significant strength of the present study is that research nurses obtained body measurements at both visits to avoid the risk of recall bias or self-presentational bias. The cohort is population-based and representative for breast cancer patients treated at this clinic [47] with excellent followup rates [53]. The patients were included preoperatively before any procedures that could affect body composition had taken place. Patients of all ages were included, which allowed for the study of the prognostic impacts of anthropometric changes in different age groups. However, there are limitations. Importantly, no information was available as to whether weight change was intentional or how fast the weight changes occurred. Since unintentional weight loss can arise due to an underlying disease, this can lead to reverse causality. Rapid weight loss might have other biologic effects than more gradual loss [54]. No information on muscle or fat mass was available for evaluation of sarcopenic obesity. Additionally, information on exercise, comorbidities, and socioeconomic status were missing, all of which might impact prognosis [55][56][57]. Only weight change by the 1-year visit was analyzed, although 76.8% remained in the same weight change category by the 2-year visit. Chemotherapy, which is administered during the first postoperative year, conferred both weight gain and loss, while late treatment effects were not captured. Few events occurred in some subgroups, resulting in lower power. Multivariable analyses were performed, including several potential confounders. Since this is an observational study, residual confounding remains possible.
In conclusion, changes in weight and anthropometric measures during the first postoperative year impacted differently on breast cancer prognosis depending on age and tumor ER status, but were independent of overweight. Whether the effect modifications of ER status results from endocrine therapy or impacts therapy response remains to be elucidated and confirmed in an independent cohort, preferably a randomized trial. Stratification of patients according to age and tumor ER status in future studies could further elucidate the impact of weight changes in subgroups of patients. Weight gain was associated with recurrences only in patients < 50 years. Weight loss did not improve prognosis in any age group. This challenges the prevailing opinion that weight loss must be beneficial in overweight patients and further research is imperative before recommending weight loss for all overweight breast cancer patients.