Risk of Breast Cancer With Long-term Use of Calcium Channel Blockers or Angiotensin-Converting Enzyme Inhibitors Among Older Women

Marsha A. Raebel; Chan Zeng; T. Craig Cheetham; David H. Smith; Heather Spencer Feigelson; Nikki M. Carroll; Kristin Goddard; Heather M. Tavel; Denise M. Boudreau; Susan Shetterly; Stanley Xu

Disclosures

Am J Epidemiol. 2017;185(4):264-273. 

In This Article

Methods

Setting and Population

This retrospective cohort study was conducted at Kaiser Permanente (KP), a US health-care provider. KP includes 7 sites that together comprise the largest private, not-for-profit, integrated health-care delivery system in the United States. The 3 sites participating in this study, KP Colorado, KP Northwest (Oregon and Washington), and KP Southern California, had a combined membership exceeding 4.5 million in 2013 (KP, unpublished data). Members receive outpatient health care at KP-owned facilities; medical facilities additionally provide laboratory, radiology, and pharmacy services. All KP facilities have fully integrated ambulatory electronic health records in all patient care areas, with access from any KP location.

The overall population included women aged ≥55 years with KP membership between January 1, 1997 (KP Southern California), January 1, 1998 (KP Northwest), or January 1, 1999 (KP Colorado) and April 30, 2012 (all sites). We studied women aged 55 years or older because both breast cancer and hypertension occur more often in older women. Further, 55 years is a commonly used age cutpoint for defining menopause.[15] From the overall population, we identified the source population as persons with at least 2 coded hypertension diagnoses (International Classification of Diseases, Ninth Revision, Clinical Modification, codes 401.*–405.*) on separate dates during outpatient visits between study start and end dates. From this source population, women with a pharmacy benefit and dispensing of a CCB or ACEi were then identified.

Women with any CCB or ACEi dispensing were stratified into new users and prevalent users (Figure 1). New users were defined as having no dispensing of either a CCB or an ACEi during a 1-year look-back period prior to the first dispensing in the study date range. Women with any CCB or ACEi dispensing within <1 year after they met study inclusion criteria were defined as prevalent users and were not included in the study.

Figure 1.

Selection of participants for a study of the relationship between long-term use of calcium channel blockers (CCBs) or angiotensin-converting enzyme inhibitors (ACEis) and the risk of breast cancer at 3 Kaiser Permanente sites, 1997–2012. Women were enrolled in the cohort beginning on January 1, 1997, at Kaiser Permanente Southern California, January 1, 1998, at Kaiser Permanente Northwest, and January 1, 1999, at Kaiser Permanente Colorado and were followed through April 30, 2012 (all 3 sites). Diagnoses were based on having International Classification of Diseases, Ninth Revision, Clinical Modification, codes 401.*–405.* recorded on separate dates during outpatient visits between the site-specific beginning date and the study end date. New users were defined as women with no pharmacy dispensing of an ACEi or CCB during a 1-year look-back period. Prevalent users were defined as women with any dispensing of an ACEi or CCB during the look-back period. Censoring events included death, disenrollment, prophylactic mastectomy, switching from a CCB to an ACEi, or switching from an ACEi to a CCB.

For new users, the date of the first CCB or ACEi dispensing was the index date. We followed these women from the index date to the date of breast cancer diagnosis or a censoring event, including a switch from a CCB to an ACEi, a switch from an ACEi to a CCB, death, disenrollment from KP, prophylactic mastectomy, or April 30, 2013, whichever occurred first.

We excluded women with any breast cancer diagnosis before the index date. Breast cancers diagnosed within 12 months after the index date were considered preexisting cancers. We incorporated a 1-year lag period after the first CCB or ACEi dispensing; women who developed breast cancer or were otherwise censored during this first year did not contribute to analyses.

The study protocol was approved by the KP Colorado Institutional Review Board; KP Southern California and KP Northwest ceded oversight to KP Colorado. The requirement for informed consent was waived.

Data Sources

The data source for inclusion criteria, exposures, and most covariates was the KP Virtual Data Warehouse. The Virtual Data Warehouse contains administrative, electronic health record, and other data that have been extracted, loaded, and maintained in identically formatted tables using standardized variable names and values.[16,17] The data source for breast cancer outcomes was the Cancer Research Network's Virtual Tumor Registry.[18] The Virtual Tumor Registry contains data that adhere to the standards of the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) program and the North American Association of Central Cancer Registries.[19] To populate the Virtual Tumor Registry, tumor registrars review and abstract data from medical records.[20–23] Virtual Tumor Registry and Virtual Data Warehouse data were linked through a common, unique study identifier.

Exposures and Outcome

The first exposure of interest was at least 1 year of use (i.e., ≥365 days' supply dispensed) of any CCB-containing product (i.e., amlodipine, felodipine, isradipine, nicardipine, nifedipine, nimodipine, nisoldipine, diltiazem, or verapamil). The second exposure of interest was at least 1 year of use of any ACEi-containing product (i.e., benazepril, captopril, enalapril, fosinopril, lisinopril, moexipril, perindopril, quinapril, ramipril, or trandolapril). Duration of use was calculated from total number of dispensings, dispensing dates, and each days' supply dispensed. Women accumulated person-time in the CCB or ACEi group until they discontinued the medication class, switched from an ACEi to a CCB or from a CCB to an ACEi, or had another censoring event.

Outcomes included incident invasive lobular or ductal carcinoma (International Classification of Diseases for Oncology codes C500–C506 and C508–C509) diagnosed after the index date and by April 30, 2013.

Covariates

Covariates initially included age,[13] body mass index (weight (kg)/height (m)2),[24–26] education (2000 US Census data), year of cohort entry, alcohol abuse (International Classification of Diseases, Ninth Revision, code 291.xx or 303.xx; yes/no), tobacco use (never, former, or current smoking), history of hysterectomy, history of cardiovascular disease other than hypertension, diabetes, use of other selected medications (statins, insulin, metformin, estrogen replacement with/without progestin, and nonsteroidal antiinflammatory agents), number of ambulatory visits, receipt of mammograms (a proxy for screening bias), KP site, and race/ethnicity. We considered use of β-blockers, angiotensin receptor blockers, and diuretics because these antihypertensive agents are often used concomitantly with CCBs or ACEis, and concerns have been raised about whether they are associated with increased breast cancer risk.[1,4,5,9,27,28]

Statistical Methods

We characterized the cohort sociodemographically and checked data quality and consistency and correlations. Two covariates, history of other cardiovascular disease and tobacco use, were removed because they correlated with other variables (r > 0.40). Covariates selected a priori for inclusion were age, race/ethnicity, mammography, KP site, year of cohort entry, hysterectomy, alcohol abuse, body mass index, and use of angiotensin receptor blockers, β-blockers, diuretics, and/or statins. All other covariates were tested for significance (P < 0.10) in univariate outcome models, and the following were included in the final models: education, diabetes, and estrogen replacement therapy.

We used discrete-time survival analyses to examine associations between cumulative durations of CCB and ACEi use and breast cancer. In this method, each woman's duration of use was broken into discrete yearly periods—that is, 1–<2 years, 2–<3 years, etc., through 11–<12 years of cumulative CCB or ACEi exposure. In this expanded person-period data set, each woman had 1 record for each discrete yearly period in which she was observed and an event indicator for whether breast cancer occurred during that period. Crude breast cancer hazards for individual periods were obtained using the life-table approach. We then applied the complementary log-log binary model, regressing incident breast cancer on the yearly periods to estimate duration associations after adjusting for covariates. We selected discrete-time survival analysis with a complementary log-log link function because it is appropriate for events that can happen at any time but are only observed in discrete intervals, such as breast cancer, and the parameter estimates provide the relative hazard interpretation (hazard ratios and 95% confidence intervals).[29,30] Each adjusted model included time-varying covariates (age, body mass index, hysterectomy, diabetes, alcohol abuse, estrogen replacement, mammography, and use of angiotensin receptor blockers, β-blockers, diuretics, and/or statins) and fixed covariates (KP site, race/ethnicity, education, and year of cohort entry). Time-varying covariates were updated at the beginning of each period. Subsequent to assessing the hazard ratio for breast cancer for each yearly period for CCB and ACEi separately, we fitted discrete-time models treating period as a continuous variable to obtain estimated slope coefficients for examination of the linear trend in duration association. We additionally examined whether the linear slopes for duration associations were similar for CCB and ACEi and tested for significance by means of an interaction between yearly duration and drug exposure.

In the primary analyses, duration of CCB or ACEi use was computed using the cumulative numbers of days' supply of CCB or ACEi dispensed, counting the days' supply of overlapping dispensings as stockpiled medication. For example, if a woman had 20 dispensings containing 60 days' supply each, her cumulative days' supply was 1,200, regardless of whether she obtained those dispensings at 60-day intervals or earlier. To determine the robustness of our computations, as a secondary analysis we recomputed durations of CCB and ACEi use not including overlapping dispensings as stockpiled medications. Using the same example woman as above, if she had 20 dispensings containing 60 days' supply each and consistently obtained refills 50 days after the previous dispensing, her cumulative days' supply was 1,000 in the secondary analysis.

We did not include a variable for length of study follow-up in the primary or secondary adjusted analyses, because duration of CCB or ACEi use and length of study follow-up were measured on the same time scale and were highly correlated (r = 0.76). However, because women with very long follow-up times had a greater opportunity to have breast cancer detected than women with shorter follow-up, to adjust for any differential length of study follow-up between CCB and ACEi users we identified a subcohort of women with frequency-matched lengths of follow-up. After preparing the matched subcohort, we completed analyses using the same techniques as in the primary analysis.

Because we required women to have ≥2 hypertension diagnoses but both diagnoses did not have to occur prior to CCB or ACEi initiation, there was potential for selection bias. To assess this, we conducted an analysis restricted to women with ≥2 hypertension diagnoses prior to initiation of CCB or ACEi use. All statistical analyses were conducted using SAS, version 9.2 (SAS Institute, Inc., Cary, North Carolina).

processing....