All-Cause and Cardiovascular Disease Mortality Among Breast Cancer Survivors in CLUE II, a Long-Standing Community-Based Cohort

Cody Ramin, PhD; Marcy L. Schaeffer, PhD; Zihe Zheng, MHS; Avonne E. Connor, PhD; Judith Hoffman-Bolton, AA; Bryan Lau, PhD; Kala Visvanathan, MD, MHS


J Natl Cancer Inst. 2021;113(2):137-145. 

In This Article


Study Participants and Design

The CLUE II ("Give Us a Clue to Cancer and Heart Disease") cohort was formed in 1989 when 32 894 residents of Washington County, Maryland, and the surrounding area completed an enrollment questionnaire and provided blood samples.[20] Written informed consent was obtained for all patients, and the study was approved by the institutional review board of Johns Hopkins Bloomberg School of Public Health.

For the present analysis, we identified women aged 18 years up to and including 80 years from CLUE II with no history of cancer (except for nonmelanoma skin cancer or cervical carcinoma in situ) at study enrollment who developed a first primary stage I-III breast cancer during follow-up (through December 31, 2015) (n = 628). For each woman diagnosed with breast cancer, we randomly selected 5 cancer-free women matched on age within 1 year (n = 3140). A ratio of 1:5 matching resulted in the closest matching on age at diagnosis while maintaining statistical efficiency. The index date was the date of breast cancer diagnosis for women with breast cancer and for their matched cancer-free women.

Ascertainment of Breast Cancer

Breast cancer diagnoses were ascertained through regular linkage to the Washington County hospital records and state cancer registry. In our study, all breast cancers were confirmed through pathology or medical records. Information on clinical and tumor characteristics included date of diagnosis, age at diagnosis, tumor stage, tumor size, and ER status.

Ascertainment of Death

Deaths were identified via hospital records, Maryland Vital Statistics, National Death Index, next of kin, and obituaries through December 31, 2015. Cause of death was ascertained through death certificates per Centers for Disease Control and Prevention guidelines.[21] The following International Classification of Disease codes were used to identify CVD as the primary cause of death: 390–398, 402, 404, 410–429, I00-I09, I11, I13, and I20-I51. These codes were defined a priori to primarily represent deaths due to heart disease, including ischemic heart disease, hypertensive heart disease, pulmonary heart disease, and other heart diseases (eg, cardiomyopathy and heart failure).

Ascertainment of Covariates

Covariate information was obtained from the enrollment questionnaire in 1989. The questionnaire included information on date of birth, anthropometric factors (weight, height), lifestyle behaviors (smoking, alcohol use), reproductive or hormonal factors (oral contraceptive use, hormone use, and menopause status), medication use within the past 48 hours (eg, medication for blood pressure, cholesterol, CVD, diabetes), and socio-demographic indicators (race or ethnicity, education). In addition, resting blood pressure and plasma total cholesterol were measured at the time of study enrollment.

Statistical Analysis

Characteristics of breast cancer survivors and cancer-free women were compared with frequency distributions for categorical variables and means (SDs) for continuous variables. Breast cancer clinical characteristics were also summarized for breast cancer survivors.

For all-cause mortality, we calculated Kaplan-Meier failure curves and used Cox proportional hazards regression to calculate hazard ratios (HRs) and 95% confidence intervals (CIs). Time since index date was used as the underlying time metric. Women contributed person-time from the index date to the date of death or December 31, 2015, whichever occurred first. The proportional hazards assumption was assessed graphically and with Schoenfeld residuals; there was no indication that the assumption of proportional hazards was violated. Results are presented overall, stratified by stage (I, II or III) and ER status (ER-positive, ER-negative), and restricted to older women at diagnosis (≥70 years, based on the age distribution in the cohort). To examine temporal trends, Cox proportional hazards regression models were stratified by time since diagnosis (0-5 years, >5-15 years, and >15 years). These cut points were determined a priori and based on clinically meaningful thresholds for risk of breast cancer recurrence.[22] Temporal trends were examined by stage at diagnosis among ER-positive survivors and older survivors. We used multiplicative interaction terms between breast cancer status and time since diagnosis. Overall models were stratified by index date (<2005, ≥2005) to examine whether trends in mortality differed by year of diagnosis. We also examined initial breast cancer treatment subgroups (radiation, chemotherapy, and hormone therapy) among women with an index date of 1998 or later because breast cancer treatment had less than 10% missing during this time period.

We used inverse probability weighting (IPW) to standardize the distribution of variables between survivors and cancer-free women in both Kaplan Meier failure curves and regression models for all-cause mortality.[23–26] Adjustment for covariates included age, menopausal status, education, smoking status, alcohol intake, body mass index (BMI), and history of oral hormone use. Further details on IPW and covariates are included in the Supplementary Methods (available online). To account for possible changes in covariates over time, we conducted a sensitivity analysis restricted to women diagnosed within 5 years of completing the enrollment questionnaire.

For CVD-related mortality, we used a competing risk approach to account for non-CVD mortality as a competing event.[27–29] For these analyses, there were 3 potential outcomes: CVD-related death (event of interest); non–CVD-related death (competing event), and administrative censoring (alive at the end of follow-up). Subdistribution hazard ratios using Fine and Gray regression models are presented in the text. Cause-specific hazard ratios were similar and are reported in the Supplementary Methods (available online). Details on the competing risk approach are further described in the Supplementary Methods (available online). IPW was used to standardize the distribution of variables for cumulative incidence curves and regression models. Because the assumption of proportional hazards by breast cancer status was violated for CVD-related mortality, we also report associations stratified by follow-up time.

We conducted a similar series of stratified analyses for CVD-related mortality by stage, ER status, and older age at diagnosis. However, temporal trends were stratified by 0–8 years vs greater than 8 years of follow-up. These cut points were determined empirically based on the overall cumulative incidence curves. We restricted estimates up to 15 years when examining temporal trends due to a small number of events in survivors after this cut point. Temporal trends for stage II or III and results for treatment subgroups were not included due to the small number of CVD deaths. We further conducted several post hoc analyses to examine CVD-related mortality due to ischemic heart disease.

Analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC) and Stata version 14.0 (StataCorp LP, College Station, TX). All statistical tests were 2-sided, and P values less than .05 were considered statistically significant.