Vasomotor symptoms in midlife women with incident breast cancer: pink SWAN

Purpose We compared trajectories of vasomotor symptoms (VMS) and their risk factors in women with breast cancer (BrCa) to those of cancer-free controls. Methods Data were from 15 nearly annual follow-up visits (1996–2017) of the multi-racial/ethnic cohort of midlife women enrolled in the Study of Women’s Health Across the Nation (SWAN). We compared women with incident BrCa to controls for patterns of VMS, controlling for risk factors identified in bivariate analyses using multivariable longitudinal analyses. Results Characteristics at study entry largely did not differ between cases (n = 151) and controls (n = 2161). Adjusted prevalence of any VMS increased significantly among cases from diagnosis to 2.75 years post diagnosis [per-year adjusted odds ratio (aOR) = 1.76, 95% confidence interval (CI) 1.39–2.24], peaking at 2.75 years post diagnosis, whereas prevalence was stable among controls in this interval [aOR = 1.04, 95% CI 0.99–1.11]. Beyond 2.75 years post diagnosis, prevalence declined significantly in cases [aOR = 0.72, 95% CI 0.61–0.84] and less in controls [aOR = 0.96, 95% CI 0.92–1.00]. Patterns were similar for frequent VMS. Adjustment for tamoxifen use slightly reduced the per-year OR for any prevalent VMS post diagnosis, partially explaining excess VMS in cases. Other treatments were unassociated with VMS. Conclusions Patterns of prevalent VMS reporting differed significantly between cases and controls, particularly post diagnosis, the latter only partially explained by tamoxifen use among cases. Risk factors for VMS largely did not differ between cases and controls.

The Australian Longitudinal Study on Women's Health showed no association of VMS with incident BrCa (aHR 1.09) [20]. In the Life and Longevity After Cancer study, post-diagnosis VMS were significantly associated with chemotherapy (aOR 1.80), adjuvant hormone therapy (aOR 2.73), prior VMS (aOR 2.20), and older age [21]. The Study of Women's Health Across the Nation (SWAN) found VMS were protective for BrCa incidence (aHR 0.63) during 11.4 years of follow-up [22]. Here, we extended follow-up to 15 SWAN visits and investigated VMS pre-and post-diagnosis to elucidate whether VMS patterns differed between women with and without BrCa or reflected BrCa treatment effects. Additionally, we compared risk factors for VMS in women with BrCa to those without cancer. We addressed the following hypotheses: (1) among women diagnosed with BrCa during followup, incidence and prevalence rates of VMS would be higher after diagnosis and cancer treatment than before; (2) prevalence rates of VMS in BrCa cases would be lower than in controls pre-diagnosis but higher post diagnosis; (3) in cases, treatment with BrCa-related selective estrogen receptor modulators (BrCa SERMS), endocrine medications, chemotherapy, or radiotherapy would be associated with greater prevalent VMS post treatment; and. (4) risk factors for VMS pre-and post-diagnosis among BrCa cases would not differ from controls and those found previously in SWAN [2].

Study participants
SWAN is a seven-site, longitudinal study of a multi-racial/ ethnic cohort, characterizing physiological and psychosocial changes during the MT and assessing their relations to subsequent health. Each site recruited community-dwelling midlife women using random-digit dialing and/or list-based sampling [23] to identify Caucasians and one minority sample (African Americans in Boston, Pittsburgh, Chicago and Detroit; Hispanics in Newark, New Jersey; Japanese in Los Angeles; and Chinese in Oakland, CA). SWAN's crosssectional survey screened women for eligibility for the longitudinal study. Eligibility for the cross-sectional study included: resident in the geographic area of one of the sites, age 40-55 years, one of the target racial/ethnic groups, and spoke English or Japanese in Los Angeles, Spanish in New Jersey, or Cantonese in Oakland. Cohort eligibility included: aged 42-52 years, at least one menstrual period in the prior three months, not pregnant or lactating and not using exogenous sex steroids. Of 16,065 cross-sectional participants, 6521 were cohort eligible, and 3302 (50.6%) enrolled. For this study (Pink SWAN), cases were women who developed BrCa since SWAN's enrollment and had no cancer (except non-melanoma skin cancer) prior to developing BrCa. Controls comprised women with no cancer (other than non-melanoma skin cancer) prior to their index visit (defined below). Data for cases were censored after any recurrence or other cancer that developed after their BrCa diagnosis and for controls after any cancer developed after their index visit. Over 15 follow-up visits, we identified 151 incident BrCa cases and 2161 controls (Fig. 1). BrCa risk status was not ascertained for controls.
Adjudication of participant medical records began at visit 12. Of 109 cases for whom medical records were received, 103 cases were confirmed (94.5% agreement with selfreport). For each case, we ascertained the date of diagnosis. To assign a corresponding date in controls, an index date was randomly assigned, using frequency matching to cases for first post-diagnosis visit, and randomly assigning a date between the last pre-diagnosis visit and the first post-diagnosis visit, such that the distributions of diagnosis/index dates were comparable in cases and controls.
Exclusions were: BrCa diagnosed prior to baseline (N = 28); other cancer diagnosed prior to baseline (N = 37) or between baseline and diagnosis/index visit (N = 104); and potential controls who missed a study visit corresponding to their randomly assigned index date (N = 885) (Fig. 1). Of these 885, 716 did not have at least one visit prior to and one visit after the index visit, so that interpolating data for a visit was not feasible. Compared with included controls, those excluded were more likely to be from the New Jersey site and Hispanic, reflecting a hiatus in data collection for visits 6-8 and 10-11 at that site. The higher proportion of Hispanics likely explains the slightly younger age, higher body mass index (BMI), lower educational attainment, higher anxiety and depressive symptom scores, and greater proportion of smokers among those excluded. Multivariate analyses adjusted for these characteristics.
The study protocol common to all sites was approved by each site's institutional review board, and all participants provided written, signed informed consent.

Procedures
Participants were followed with approximately annual inperson visits from 1996 to 2017. All visits included: consent; a standard protocol for measuring height with a stadiometer and weight with a balance beam scale; and interviewer-and self-administered questionnaires, including questions on demographic characteristics, medical and reproductive history, medications, psychosocial measures, and symptoms. Questionnaires were forward-and back-translated into Spanish, Cantonese, and Japanese and administered by bilingual interviewers to participants preferring these languages.

Outcome variables
At each visit, participants were asked the frequency of their hot flashes and/or night sweats in the prior 2 weeks [24][25][26]. Prior analyses [2] indicated that these two symptoms were highly correlated, providing justification for examining them jointly. Any VMS was defined as reporting at least 1 day with hot flashes or night sweats within the prior 2 weeks, and frequent VMS as having VMS on 6 or more days in the prior 2 weeks [2].

Covariates
Time-invariant factors included race/ethnicity, site, visit 1 symptom sensitivity [27], menopausal stage and use of systemic menopausal hormone therapy (HT) at the index visit, baseline history of premenstrual symptoms, education, and ever receiving BrCa chemotherapy or radiotherapy (treatment timing was unavailable). Chemotherapy and radiotherapy were ascertained from interviews, medical records, and self-reported questionnaires from cases.
All other covariates were time-varying. Because current smoking [28] was stable, missing values were interpolated using last observation carried forward. Passive smoke exposure [29] and alcohol consumption were missing in more than 2700 observations and thus excluded; adjustment for these two characteristics yielded similar results in the subsample for whom they were available (data not shown). Psychosocial characteristics included elevated anxiety score [24] for the prior two weeks (score > 4 [30]) and elevated depressive symptom score (score ≥ 16 on the 20-item Center for Epidemiology Studies-Depression (CES-D) scale for the previous week [31]), both lagged one visit. Marital status was categorized as never, previously, or currently married/ partnered. Self-assessed health was categorized as excellent, very good, good, and fair/poor. BMI was computed At all visits except 11 and 14, interviewers ascertained prescription medication use since the prior visit, transcribing preparation names from medication containers that participants brought into the visit. Medications analyzed here included HT, anti-anxiety medications, anti-depressants, BrCa SERMS (tamoxifen for all but one case), non-BrCa SERMs, and endocrine medications. BrCa treatment information (BrCa SERMS, endocrine medications [including aromatase inhibitors, GnRH agonists and (less frequently reported in our participants) estrogen receptor blockers and inhibitors of Type II 5α-reductase], chemotherapy and radiotherapy) was obtained in annual surveys and from a Pink SWAN questionnaire and medical records for women who reported incident BrCa. Medications were coded using the Iowa Drug Information Service system. Follow-up visits 11 and 14 were abbreviated; medications other than HT, employment, and self-reported health were not ascertained. Values were imputed using logistic or linear regression, based on adjacent visit values, age, menopausal status, race/ ethnicity, site, and case/control status.
Menopausal status categories-both at the index visit and time-varying-included premenopausal (menses in the prior three months without change in menstrual regularity in the prior year), early peri-menopausal (menses in the prior 3 months and changes in regularity in the prior year), late peri-menopausal (3-11 months of amenorrhea) and naturally postmenopausal (≥ 12 months of amenorrhea). Women who had a hysterectomy and/or bilateral oophorectomy were grouped with naturally postmenopausal women at the index visit due to small numbers of such surgeries and similar rates of VMS in the two groups but were separated for timevarying menopausal status. Pre-and early peri-menopausal stages were combined due to the small number of premenopausal observations post diagnosis.
For women initiating menopausal HT or oral contraception use before their final menstrual period, menopausal status could not be discerned because menstrual bleeding was masked. If a woman stopped such hormone use and resumed bleeding, menopausal status was categorized according to the criteria above. Those who had the same status at both the visits prior to and following hormone use were categorized with that menopausal stage; otherwise menopausal status was considered missing.

Data analyses
Participant characteristics were compared for cases and controls using chi-square or Fisher's exact test for categorical variables and Student's t test for continuous variables.
In addition to case/control status, the primary predictor was years before/after diagnosis/index date with diagnosis/ index date = time 0. Diagnosis date was used as a surrogate for treatment initiation, which we assumed occurred shortly after diagnosis. Analyses of prevalent VMS omitted data outside the range − 10 to + 10 years to avoid sparse data.
Patterns of prevalent and incident VMS were examined in relation to the diagnosis/index date to distinguish patterns of persistent VMS in women who already had VMS from newly developed VMS, respectively. For both prevalence and incidence, any and frequent VMS were modeled separately.
Prevalence of VMS was modeled using random effects logistic regression [32]. Separately for cases and controls, we used nonparametric, locally weighted scatterplot smoothing (LOESS) regression by years before/after diagnosis/index [33]. Possible knots-points of inflection with a change in slope-suggested by LOESS were tested in piecewise linear modeling to identify the best-fitting knot placement. To allow cases and controls to have different slopes in segments delineated by knots, we included interactions of case/control status and segment-specific time variables in logistic regression models. Results from these models are presented as: (a) point prevalences at segment start and (b) segmentspecific per-year adjusted odds ratios (aOR); the former indicates the point prevalence level at the start of segments, and the latter reflects change in odds with an additional 1 year elapsed.
To address hypothesis #1, in cases we compared point prevalence (PP) of VMS at diagnosis and post-diagnosis, and per-year aORs in different segments. For hypothesis #2, we compared PP and per-year aORs for cases versus controls in each segment. Covariate adjustment was performed in three stages: covariates other than BrCa treatments; adding medications (hypothesis #3); and adding chemotherapy and radiotherapy (hypothesis #3). For hypothesis #4, we tested interactions between case/control status and VMS risk factors previously identified in SWAN [2], stratifying on pre and post diagnosis/index. We also examined changes in HT use, as cases may have been more likely to stop HT postdiagnosis, potentially leading to "rebound" VMS.
Incidence of VMS was analyzed in participants without baseline VMS using Kaplan-Meier plots-and covariateadjusted discrete survival analysis accounting for "late entry"-participants were excluded from risk sets earlier than their study entry [34].

Results
Controls had a slightly but not statistically significantly higher point prevalence (PP) of VMS at baseline (39.3%) than cases (34.7%) ( Table 1). Demographic, psychosocial and health characteristics did not differ between cases and controls, except cases tended to drink more alcohol (Table 1). Factors associated with VMS at baseline largely did not differ from those previously reported in SWAN [2] (data not shown). The LOESS plot of the PP of any VMS among cases (hypothesis #1) suggested two knots with an increase starting at diagnosis (year 0) and a decrease starting at 2.75 years post-diagnosis (Fig. 2a). Among controls, VMS decreased gradually over time (hypothesis #2). Patterns were similar for the PP for frequent VMS (Fig. 2b). Consequently, all models used two knots, estimating separate per-year aORs for cases and controls in each segment: Segment 1 = up to 10 years pre-diagnosis/index; Segment 2 = diagnosis/index to 2.75 years post-diagnosis/index; and Segment 3 = 2.75-10 years post-diagnosis/index. For hypothesis #1, analyses only in cases, adjusted PP of any VMS declined slightly pre-diagnosis (per-year aOR = 0.91; Table 2, first model), but increased sharply in Segment 2 with PP at 2.75 years post diagnosis was almost twice the PP at diagnosis (77.9% versus 42.6%, p < 0.001). The per-year aOR for cases in Segment 2 is 1.76, indicating that for each additional passing year, the odds of any VMS increased by a factor of 1.76. Subsequently, prevalence rates declined (Segment 3 per-year aOR = 0.72). Pairwise differences between segment-specific per-year aORs were statistically significant (p < 0.004).
In hypothesis #2 analyses, adjusted PP of any VMS was somewhat higher in cases than controls pre-diagnosis/index (p = 0.053) (  Adjusted PPs for frequent VMS were lower (6.7-9.5% among controls, 7.6-27.1% among cases), but case/control differences in PP and per-year aORs were consistent with those for any VMS (data not shown). New HT use at the first post-diagnosis/index visit was low in both cases and controls (3.9% versus 4.3%, respectively, p = 0.85). However, continued use among cases was significantly lower than among controls at the first post-diagnosis visit (28.6% versus 61.8%, p = 0.003). Adjustment for HT patterns had a negligible effect on per-year aORs, however (data not shown), nor was the interaction of HT pattern with case/control status statistically significant (p = 0.67).
Use of BrCa SERMS occurred almost entirely among cases post-diagnosis (29.0% of Segment 2 observations, 14.4% of Segment 3 observations), with only two Segment 1 observations (0.01%) in controls. Endocrine medication use was reported in 15.6% and 18.2% of cases' Segment 2 and Segment 3 observations, respectively, versus 0.10% and 0.03% among controls. In hypothesis #3 analyses, concurrent BrCa SERM use was positively associated with prevalence of any VMS (aOR = 2.95, 95% CI 1.34, 6.52), but concurrent endocrine medication use was unrelated (aOR = 0.96, 95% CI 0.46, 1.97). The latter result may reflect endocrine medication use only during postmenopause, rather than in late peri-menopause when VMS PP tends to peak [2]. Table 2 Multivariable random effects binomial logistic regression for prevalence of any VMS from 10 years pre-to 10 years post-diagnosis, Pink SWAN a Per-year odds ratio reflects the change in odds associated with a 1-year increase in that segment b Adjusted for: menopause status and hormone use at first post-diagnosis visit, site, race/ethnicity and time-varying menopause transition stage and hormone use, age, body mass index, education, self-reported health, marital status, employment status, symptom sensitivity at visit 01, 1-year lagged anxiety score > 4, 1-year lagged anxiety medication use, and use of anti-depressants, history of premenstrual symptoms and smoking status c For prevalences at start of segments, p value indicates statistical significance for the case-control difference in VMS prevalence. For per-year ORs, the p value indicates statistical significance for the case-control difference in the time slope, i.e., for the interaction between case-control status and segment time Similar to previous SWAN findings [22], Kaplan-Meier plots suggested slightly lower incidence of any VMS (Fig. 3a) and frequent VMS (Figs. 3b) in cases than controls with almost all incidence for both groups occurring prediagnosis/index. This resulted in small numbers of incident VMS post-diagnosis/index (19-21 incidents in cases), making multivariate analyses of these trends unstable. Overall, aHR (95% CI) for cases versus controls were 1.16 (0.79, 1.70) for any VMS and 1.10 (0.74, 1.63) for frequent VMS. Among participants with no VMS pre-diagnosis, corresponding aHRs post-diagnosis/index were 1.37 (0.61, 3.07) and 1.55 (0.84, 2.87), respectively.
In hypothesis #4 analyses, no statistically significant case/control interactions with risk factors were observed for prevalence of frequent VMS, either pre-or post-diagnosis/ index. For prevalence of any VMS, two interactions were statistically significant: race/ethnicity pre-diagnosis (interaction p = 0.043) and menopause status post-diagnosis (interaction p = 0.026): aORs for Black versus white were 9.04 for cases versus 2.09 for controls, and for Hispanic versus white were 24.4 for cases versus 0.52 for controls; aORs for late versus early peri-menopause were 13.0 for cases but 3.77 for controls; aORs for post-versus early peri-menopause were 12.5 for cases but 2.80 for controls. Given the large number of comparisons and small cell counts (7 Hispanic observations, 33 late peri-menopause observations, and 36 pre-/early peri-menopause in cases), these significant interactions may have been due to chance.

Discussion
Our results provide a new contribution by comparing preand particularly post-diagnosis trajectories of VMS among women with incident BrCa to those of controls, revealing that much of the increase post-diagnosis was not attributable to treatment, and revealed that VMS risk factors did not differ in cases and controls.
Point prevalence of any VMS was similar for both groups pre-diagnosis/index, consistent with prior SWAN findings [20]. This differed from findings from a nearly 18-year follow-up of 25,499 postmenopausal non-HT users in the Women's Health Initiative, which revealed that women with VMS lasting 10 + years had higher BrCa incidence than those who never reported VMS [35]. In our study, the PP of VMS increased in cases from diagnosis to + 2.75 years but remained stable among controls, suggesting treatment effects in cases, supported by the significance of the BrCa SERMs variable, consistent with prior findings [21]. However, adjustment for BrCa SERM treatment only somewhat attenuated the still significantly elevated per-year aOR among cases, suggesting that treatment only partially accounted for the VMS increase in contrast to findings from prior studies Fig. 3 a Kaplan-Meier Plot of cumulative probability (with 95% Confidence Intervals) of any incident VMS over time prior to and after diagnosis/index date for cases and controls. b Kaplan-Meier plot of cumulative probability (with 95% Confidence Intervals) of frequent incident VMS over time prior to and after diagnosis/index date for cases and controls of BrCa cases taking aromatase inhibitors [6,7] or tamoxifen [8]. Consistent with prior results [3], chemotherapy had no additional impact on VMS prevalence, which declined in cases and controls after 2.75 years after diagnosis/index with a steeper decline among cases.
Most incident VMS in both groups occurred pre-diagnosis/index, resulting in small numbers of incident VMS post diagnosis/index; thus, incidence trends were unstable, but indicated no statistically significant case/control differences in post-diagnosis/index incident VMS. Because analyses of prevalent VMS included women who had VMS in prior segments, but analyses of incident VMS did not, the higher per-year aOR for prevalence of VMS among cases (after adjustment for BrCa SERMs), but not incidence of VMS in Segment 2 could reflect a predisposition for VMS among cases or common underlying factors (eg, inflammation) for BrCa [36] and VMS [37].
Our study's strengths included: (a) the longitudinal design, permitting assessment of temporal relations of VMS, BrCa and VMS risk factors; (b) inclusion of five racial/ethnic groups, providing good generalizability; and (c) standardized data, enabling statistical adjustment for many potential covariates, reducing residual confounding. The study had limitations, however. First, most notably, the number of incident BrCa cases was relatively small, potentially resulting in some modest but meaningful associations not detected as statistically significant, such as when examining incident VMS and case/control status interaction with risk factors. Second, recall errors and misclassification could have resulted from self-reported data on VMS, BrCa treatments, and medications; however, we verified reported medications against containers for prescribed medications that women brought to each visit. Third, we lacked information regarding timing of chemotherapy or radiotherapy so that we could not examine concurrent treatment but only any vs. no treatment, which may explain the lack of significant associations. Finally, most cases occurred postmenopausally so that the menopause-inducing effects, including VMS, of some BrCa therapies reported in prior studies [9,10] were not observable.
In conclusion, risk factors for VMS did not differ between cases and controls, but trajectories of prevalence of VMS differed significantly. Prevalence of VMS remained stable among controls, but increased in cases in the immediate post-diagnosis segment, partially reflecting treatment but also possibly reflecting common underlying factors for BrCa and VMS or a predisposition for VMS among cases. However, incidence of VMS did not increase post-diagnosis, arguing against predisposition, but not against common factors for VMS and BrCa. These findings suggest that prevalence of VMS post-diagnosis was only partially explained by treatment and possibly by prior prevalence of VMS in cases because analyses of VMS prevalence included women who had VMS previously, again suggesting shared underlying factors for VMS and BrCa. That post-diagnosis VMS were not entirely attributable to treatments which is important clinically because VMS could lead to lower adherence to BrCa treatment. Additionally, we did not find higher HT use in cases than controls pre-diagnosis so that higher prevalence of VMS soon after BrCa diagnosis does not seem attributable to prior HT use. use agreement between the applicant's institution and the University of Pittsburgh.
Code availability Code for the analyses for this paper will be shared on reasonable request to the SWAN Coordinating Center and will require a formal use agreement between the applicant's institution and the University of Pittsburgh.

Declarations
Conflict of interest None, not applicable.

Ethical approval
The institutional review boards at each sites approved the study protocol.
Consent to participate All sites obtained written informed consent from all of their study participants.

Consent for publication
All authors made substantial contributions to the conception or design of the work; the acquisition, analysis, or interpretation of data; or the creation of new software used in the work; drafted the work or revised it critically for important intellectual content; approved the version to be published; and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.