Eligible women were all those with a first primary breast cancer diagnosed and recorded in any of four population-based regional breast cancer registries (Auckland, Waikato, Wellington, and Christchurch)  in New Zealand between 1 Jan 2007 and 31 Dec 2016. These registers include all women diagnosed with breast cancer in their defined areas, and together cover about 70% of all breast cancer registrations in New Zealand. Using an anonymised National Health Index number, data were linked to several national data bases: the Pharmaceutical Collection (PHARMS), a national database containing dispensing information and medication identifiers from pharmacists for subsidised dispensings ; the National Minimum Dataset, relating to all day patients and inpatients discharged from both public and private hospitals; and the National Mortality Collection, with information about all certified deaths . Women were excluded if their records did not link to at least one dispensing from the pharmaceutical collection (n=14) or if their date of death was on or before their recorded date of breast cancer diagnosis (n=3). The final cohort for analyses was comprised of 14,976 women. This study was approved by the Central Health and Disability Ethics Committee (Ref: 19/CEN/4).
Exposure and outcome data
In the PHARMS database, medications dispensed after breast cancer diagnosis were determined using the therapeutic group ID, a PHARMAC identifier for each group of Anatomical, Therapeutic, and Chemical properties . All BBs were included, except those used topically for glaucoma. For each dispensing, we calculated the number of daily defined doses by multiplying the number of tablets dispensed by the dose per tablet in mg, and dividing by the daily defined dose in mg from the World Health Organisation database .
Deaths were determined from the underlying cause of death in the regional breast cancer registries and National Mortality Collection, with ICD codes C50.0 to C50.9 classified as deaths from breast cancer.
Demographic and clinical information came from the regional breast cancer registries, and covariates considered included date of diagnosis, age, ethnic group [34, 35], socioeconomic deprivation , urban/rural status , public/private status of the treatment facility, register, stage , grade , mode of detection (screen detected vs symptomatic), lymphovascular invasion, and receptor status (as defined previously , including Luminal A, Luminal B, Luminal B HER2+, HER2+ non-luminal, and triple negative). Other post-diagnostic medications included statins, aspirin and other non-steroidal anti-inflammatory medications (NSAIDs), angiotensin converting enzyme inhibitors (ACEIs), angiotensin receptor blockers (ARBs), and diuretics. Comorbidities adjusted for included any cardiac condition (angina, arrhythmia, congestive heart failure, hypertension, myocardial infarction, ‘other cardiac conditions’, and valve disease) as yes/no, diabetes, stroke, chronic obstructive pulmonary disorder, and peripheral vascular disease. We defined comorbidities as any of the above conditions appearing in a patient’s linked hospital record (inpatient admissions) in the 5-year period before their breast cancer diagnosis.
Comparisons by BB use at baseline (date of diagnosis of breast cancer) were conducted using the chi-square test. We used Cox proportional hazard models to assess hazard ratios (HRs) of breast cancer-specific mortality associated with post-diagnostic BB use vs non-use. Death registrations and Pharmaceutical Collective coverage were complete to the end of 2017, so we followed patients from their breast cancer diagnosis until death or 31 December 2017. Women with no death recorded prior to 31 December 2017 were assumed to be alive as at 31 December 2017. Medication use was conceptualised as a time-varying covariate, such that time before the first dispensing was counted as ‘nonuser’ time, and time from the first dispensing to end of follow up was counted as ‘user’ time . Models were adjusted in a systematic fashion, with the first adjustment including demographic and breast cancer clinical data, and the second adding other medication use and comorbidities.
Analyses were conducted considering BB use as a binary variable (user/nonuser), and also by splitting BB use into seven categories based on the number of daily defined doses (DDDs: categorised as 1-90 DDDs, 91-181 DDDs, 182-272 DDDs, 273-364 DDDs, 365-729 DDDs, 730-1094 DDDs, or 1095 or more DDDs, corresponding to the equivalent of 0-3 months, 3-6 months, 6-9 months, 9 months-1 year, 1-2 years, 2-3 years, and 3+ years of BB use respectively). Dose analyses were conducted using a time varying approach, such that women spent time in the lowest category before moving into the next dose category. In order to compare patterns of risk observed for BBs to those of another cardiovascular medication with similar indications , the same analyses were carried out for calcium channel blockers.
To examine the effect of BB use in early-stage patients only, an analysis was carried out restricted to patients with stage 1, stage 2, or stage 3a cancers. In this analysis, patients with an ‘unknown’ stage were excluded.
To evaluate the effect of the competing risk of death from other causes, the proportional subhazards model was also used . For this analysis, all deaths apart from breast cancer deaths were treated as competing events.
As dispensings toward the end of life may reflect changes in morbidity (including cancer recurrence/progression) or in health care related to end of life care [44, 45], we also conducted analyses lagging medication times . In these analyses, patients are initially considered nonusers and then users after a lag period has elapsed after their first medication dispensing. Using this approach, dispensings toward the end of life are removed by the lag; for example, a 6-month lag will ignore dispensings in the 6 months prior to death/last follow up and classify these women are as medication nonusers as opposed to users. To appropriately account for different periods in which end of life care may be administered, we also considered lag periods of 1 year and 2 years. In these analyses, all medications were modelled in the same fashion (for example, if BBs were lagged by 6-months, all other medications were as well).
In order to compare BB users to patients using other medications for a similar indication, a further analysis was carried out comparing BB users to BB nonusers who used another antihypertensive medication. For this comparison, other antihypertensives included ‘Potassium Sparing Combination Diuretics’, ‘Thiazide and Related Diuretics’, ‘ACE Inhibitors’, ‘ACE Inhibitors with Diuretics’, ‘Angiotensin II Antagonists’, ‘Angiotensin II Antagonists with Diuretics’, ‘Dihydropyridine Calcium Channel Blockers’, ‘Other Calcium Channel Blockers’, ‘Alpha Adrenoceptor Blockers’, and ‘Centrally-Acting Agents’. In this analysis, BB nonusers who used another antihypertensive were followed from their first post-diagnostic antihypertensive dispensing until death or 31 December 2017.
We also conducted an analysis with breast cancer recurrence (BCR) as the outcome. In this analysis, we defined a BCR as either a local/regional recurrence or distant metastasis and restricted the cohort to patients with early-stage breast cancer as above. Recurrences were determined from the breast cancer registry data through patient’s routine clinical records, and women were followed from their breast cancer diagnosis until BCR, death, last follow up date, or end of Pharmaceutical Collection coverage (31 December 2017), whichever came first. These analyses examined risk of BCR associated with BB use vs non-use, as well as risk associated with different doses of BB use vs non-use.
Results are reported as HRs and their 95% confidence intervals (CIs), with the two-sided significance level set at 0.05. Statistical analyses were conducted in STATA 13.1 (StataCorp, College Station, TX).