Background

Beta-blockers, used for heart disease and hypertension [1], act by inhibiting beta-adrenergic receptors. Recent studies have shown that breast cancer tissue expresses beta-adrenergic receptors [2], particularly type 2 beta-adrenergic receptors [3]. Numerous in-vitro studies have demonstrated that beta-blockers can disrupt migratory activity and inhibit angiogenesis of cancer cells [4, 5]. In particular, propranolol appears to have potent anti-migratory and anti-angiogenic properties as demonstrated in cancer cell lines and animal models [4–10]. This preclinical evidence has led to calls for randomised controlled trials of propranolol as adjuvant therapy in breast cancer patients [11]; however, although early phase trials are underway [12, 13], phase 3 trials have not been conducted to date.

Only three observational studies have previously investigated the association between propranolol use and breast cancer outcomes. In 2011, an Irish study observed an 80% reduction in breast cancer-specific mortality among patients who used propranolol in the year prior to diagnosis [14]. No association was observed between propranolol use after diagnosis and breast cancer-specific mortality in an English study [15] or between propranolol use after diagnosis and breast cancer recurrence in a Danish study [16]. However, these studies had limited power because of the small numbers of breast cancer patients using propranolol, reflecting the low prevalence of propranolol use which in each study was under 5% [14–16]. Therefore, a need remains to further investigate propranolol (and other non-selective beta-blockers) and survival in breast cancer patients to inform the decision on whether to conduct large phase 3 randomised controlled trials of propranolol as adjuvant cancer therapy in breast cancer patients.

Consequently, utilising the European Cancer Pharmacoepidemiology Network [17], we conducted a pooled analysis of eight cohorts of breast cancer patients from across Europe to examine whether use of propranolol (or other non-selective beta-blockers) is associated with improved breast cancer-specific and all-cause mortality.

Methods

Data sources

Eight cohorts of breast cancer patients from across Europe (including Belgium, Denmark, England, the Netherlands, Northern Ireland, Republic of Ireland, Scotland and Sweden) were compiled through the European Cancer Pharmacoepidemiology Network [17]. Characteristics of these cohorts are presented in Table 1. The association between propranolol use and cancer mortality was examined previously within the English cohort [15] (although using a nested case–control design) and the Republic of Ireland cohort [14] (although this analysis did not investigate propranolol use after diagnosis, had shorter follow-up and had substantially fewer cases). Cancer recurrence was investigated previously in the Danish cohort [16] (although the earlier analysis was based on fewer than 20% of the breast cancer patients included in the present analysis and did not investigate mortality). Previous studies have reported detailed descriptions of the medication data available and/or linkages available in the cohorts from Denmark [18–20], England [21], the Netherlands [22, 23], Northern Ireland [24], Belgium [25], Republic of Ireland [26], Scotland [27] and Sweden [28, 29].

Table 1 Characteristics of the included cohorts
Full size table

Inclusion criteria

All cohorts identified incident invasive breast cancer patients from cancer registries. The year of diagnosis for included breast cancer patients varied across the cohorts from 1998 to 2012. Patients with other invasive cancer diagnoses (apart from non-melanoma skin cancer) prior to their breast cancer diagnosis were excluded.

Exposure

Propranolol and all non-selective beta-blocker use (including propranolol, sotalol, timolol, nadolol, carvedilol, pindolol, oxprenolol and labetolol) was ascertained from electronic dispensing records in five cohorts, GP prescribing records in two cohorts and health insurance records in one cohort (see Table 1).

Outcome

In seven of the cohorts, mortality was ascertained from national death records; social security records were used in one cohort (see Table 1). Breast cancer-specific mortality was defined as breast cancer being the underlying cause of death and was available in five cohorts. All-cause mortality was available in all cohorts.

Covariates

The covariates available varied between cohorts and were obtained from a number of sources including cancer registries, hospital admissions, prescriptions, GPs and health insurance databases (see Table 1). The covariates recorded included: age, year of cancer diagnosis, stage, grade, cancer treatment within the first 6 months after diagnosis (including information on cancer-directed surgery, chemotherapy, radiotherapy), medication use (including tamoxifen, aromatase inhibitors, hormone replacement therapy (prior to diagnosis), aspirin [30], statins [31]) and comorbidities prior to diagnosis. Cancer-directed surgery, chemotherapy and radiotherapy were taken from cancer registry records, apart from in Belgium where insurance claims were used and in Denmark where Patient Registry records were used. Comorbidities, largely including those in the Charlson comorbidity index [32], were taken from hospital admission records in Denmark, Scotland and Sweden, from GP records in England and from cancer registry records in the Netherlands. In the cohorts from the Netherlands, Denmark and England, adjustments for comorbidity were made for cerebrovascular disease, chronic pulmonary disease, congestive heart disease, diabetes, myocardial infarction, peptic ulcer disease, peripheral vascular disease and renal disease. In Sweden additional adjustments were made for liver disease and in Scotland additional adjustments were made for liver disease and diabetes complications. In the Republic of Ireland cohort, comorbidity information was based upon prescribing information using the RxRisk score [33]. Oestrogen use was based upon HRT use any time prior to diagnosis in the Netherlands, HRT or oral contraceptive use in the year prior to diagnosis in Denmark or HRT use in the year prior to diagnosis in Sweden, the Republic of Ireland, Belgium and Scotland. Tamoxifen and aromatase inhibitor use was obtained from prescription records, except in Denmark were a single more complete endocrine therapy variable, based upon Patient Registry records, was used instead.

Statistical analysis

We performed a two-stage analysis procedure allowing for adjustment of covariates which were not uniformly defined, coded or available across cohorts [34]. In the main analysis of medication use after diagnosis, the patients in each cohort were followed from 1 year after breast cancer diagnosis to death or end of follow-up, whichever was sooner. Patients who had died in the first year after breast cancer diagnosis (or who had less than 1 year of follow-up) were excluded because it seemed unlikely that propranolol use after diagnosis could reduce mortality within such a short period. In the main analysis, propranolol use was modelled as a time-varying covariate to avoid immortal time bias [35]; that is, patients were initially considered non-users and then became users a lag of 1 year after their first propranolol prescription. The use of a lag period is recommended in studies of medication use and cancer survival [36] because prescriptions filled shortly prior to death may reflect end-of-life treatment. In dose–response analyses, one propranolol prescription corresponded to 1 month of use, except for Denmark where one prescription corresponded to 3 months of use (based on the average duration of propranolol prescriptions in Denmark). In dose–response analyses, an individual was considered a non-user prior to 1 year after first medication usage, a user of 0–1 year for prescriptions from 1 year after first prescription to 1 year of prescriptions (considered four prescriptions in Denmark and 12 prescriptions in all other countries) and a greater user after this time. Time-dependent Cox regression models were used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for breast cancer-specific death in propranolol users compared with propranolol non-users. An unadjusted analysis was first conducted, then an adjusted analysis (including just the covariates age and year of diagnosis in the model, which were available in all cohorts) and finally a fully adjusted analysis was conducted (including all covariates available within each cohort, as presented in Table 1, in the model). The summary HRs and standard errors (SEs) from the eight cohorts were combined using random effects models to calculate pooled HRs [37] and the consistency of HRs was investigated using chi-squared tests for heterogeneity and I 2 statistics [38]. The analyses were repeated for all-cause mortality. Analyses were then repeated comparing all non-selective beta-blockers users with non-selective beta-blocker non-users and comparing all beta-blocker users with beta-blocker non-users. A sensitivity analysis was conducted restricting the cohorts to patients with stage 1–3 breast cancer, because it is plausible that the effect might be most evident in those without advanced disease.

We performed two predefined secondary analyses. First, avoiding immortal time bias without requiring complex analyses [39], we compared users of propranolol (and separately users of non-selective beta-blockers) within the first year after breast cancer diagnosis with non-users within the same period, and started follow-up 1 year after breast cancer diagnosis. Second, to investigate the potential impact of propranolol use earlier in the process of cancer development, we performed a separate analysis of medication use before breast cancer diagnosis comparing time to death in propranolol (and non-selective beta-blockers) users with non-users in the year prior to diagnosis, restricted to individuals with at least 1 year of medication records prior to diagnosis. In analysis of pre-diagnostic medication use, patients who died in the first year after diagnosis (who had follow-up of less than 1 year) were not excluded.

Results

Patient cohorts

The pooled analysis for breast-cancer specific and all-cause mortality comprised 55,252 newly diagnosed breast cancer patients (in whom there were 5419 breast cancer-specific deaths and 9295 all-cause deaths) and 133,251 newly diagnosed breast cancer patients (in whom there were 25,472 all-cause deaths), respectively. The maximum follow-up in each cohort after diagnosis of breast cancer ranged from 5 to 13 years (see Table 1).

Patient characteristics

Patient characteristics by propranolol (and non-selective beta-blocker) use in the first year after diagnosis are presented in Table 2. Propranolol users were slightly more likely to have an earlier year of breast cancer diagnosis. Age, stage, grade and cancer treatments were generally similar by propranolol use. There was a higher use of hormone antagonists (tamoxifen 39% versus 30% and aromatase inhibitors 26% versus 23%, respectively) in propranolol users versus non-users, but use of other medications was similar.

Table 2 Characteristics of breast cancer patients by propranolol and non-selective beta-blocker use in the year after diagnosis
Full size table

Association between propranolol use after diagnosis and breast cancer-specific and all-cause mortality

Overall 4746 breast cancer patients used propranolol at any time after diagnosis (1768 from Belgian, 1057 from Denmark, 419 from England, 151 from the Netherlands, 107 from Northern Ireland, 232 from the republic of Ireland, 629 from Scotland and 383 from Sweden). Table 3 and Fig. 1 present the findings from the main analysis. Overall, there was little difference in breast cancer-specific mortality or all-cause mortality in propranolol users compared with non-users after diagnosis (fully adjusted HR = 0.94, 95% CI, 0.77, 1.16 and HR = 1.09, 95% CI, 0.93, 1.28, respectively). The associations between propranolol and cancer-specific mortality were fairly consistent across cohorts (I 2 = 0% and heterogeneity P = 0.56), whereas the association varied more for all-cause mortality (I 2 = 65% and heterogeneity P = 0.006). On closer inspection (see Fig. 1) this heterogeneity was partly due to the Belgian estimate; once this was removed the pooled estimate was attenuated slightly (fully adjusted HR = 1.03, 95% CI, 0.88, 1.20) and the heterogeneity was reduced (I 2 = 39% and heterogeneity P = 0.02). There was little evidence of a dose–response association; compared with propranolol non-users, there was no association between use of more than 1 year of propranolol prescriptions and cancer-specific or all-cause mortality (fully adjusted HR = 0.93, 95% CI, 0.46, 1.90 and HR = 1.09, 95% CI, 0.85, 1.40, respectively). Similar null associations were observed for cancer-specific mortality when comparing users of non-selective beta-blockers with non-users of non-selective beta-blockers (see Table 3).

Table 3 Pooled analysis of the association between propranolol and non-selective beta-blocker use after breast cancer diagnosis and breast cancer-specific and all-cause mortality
Full size table
Fig. 1
figure 1

Association between propranolol and breast cancer-specific and all-cause mortality, by cohort. *Fully adjusted model contains age, year, stage and confounders presented in Table 1. CI confidence interval, HR hazard ratio

Full size image

Secondary and sensitivity analyses

Secondary and sensitivity analyses are presented in Table 4. In sensitivity analyses restricting the cohorts to stage 1–3 breast cancer patients only, the associations between propranolol and cancer-specific and all-cause mortality was similar to those for the main analysis (see Table 4). In secondary analysis there was no evidence of an inverse association between any beta-blocker use after diagnosis and cancer-specific or all-cause mortality (fully adjusted HR = 1.07, 95% CI, 0.99, 1.16 and HR = 1.12, 95% CI, 1.05, 1.20, respectively). The secondary analysis based upon medication use in the first year after diagnosis also produced similar results for propranolol and cancer-specific and all-cause mortality (fully adjusted HR = 1.07, 95% CI, 0.72, 1.60 and HR = 1.04, 95% CI, 0.89, 1.21, respectively).

Table 4 Secondary and sensitivity analyses for pooled analysis of the association between propranolol and non-selective beta-blocker use and breast cancer-specific and all-cause mortality
Full size table

Table 4 also presents results for the analysis of propranolol use before diagnosis. Propranolol use in the year before diagnosis was not associated with reduced cancer-specific or all-cause mortality (fully adjusted HR = 1.03, 95% CI, 0.86, 1.22 and HR = 1.02, 95% CI, 0.94, 1.10, respectively). In all secondary analyses of non-selective beta-blocker use, similar associations were observed to those for propranolol use (see Table 4).

Discussion

This large pooled analysis of breast cancer patients did not present convincing evidence of reduced cancer-specific or all-cause mortality in breast cancer patients who used propranolol or non-selective beta-blockers either before or after breast cancer diagnosis.

Our pooled analysis supports the findings of two earlier epidemiological studies of the association between propranolol use after diagnosis and cancer outcomes [15, 16]. The first, an earlier analysis of Danish data [16], showed no association between propranolol use after diagnosis and recurrence (adjusted HR = 1.3, 95% CI, 0.92, 1.9); however, that study did not investigate mortality or the influence of propranolol use before diagnosis. The second study, an earlier analysis of English data [15], based upon a case–control design, showed no association between propranolol and cancer-specific mortality (adjusted HR = 0.98, 95% CI, 0.57, 1.71).

Our pooled analysis also showed no reduction in cancer-specific mortality associated with propranolol use before diagnosis and therefore does not support the results of an earlier Irish study, the only previous study to investigate this association, which observed an 80% reduction in breast cancer-specific mortality (adjusted HR = 0.19, 95% CI, 0.06 0.60) in 46 breast cancer patients using propranolol in the year prior to diagnosis [14].

The main strength of our analysis is statistical power; this is the largest study yet to investigate the association between use of propranolol and cancer outcomes in breast cancer patients. Despite this, there remains the possibility of type 2 error and we cannot rule out a weak protective effect of propranolol on cancer-specific mortality. Other strengths include the long duration of follow-up, which was up to 13 years following breast cancer diagnosis in some cohorts. The use of routinely recorded drug information allowed precise evaluations of temporal relationships between propranolol use and mortality and eliminated the potential for recall bias incurred in questionnaire-based studies. Misclassification due to over-the-counter use was likely to be minimal because propranolol can be obtained only by prescription in the included countries.

A weakness of the study is the potential for bias due to the misclassification of breast cancer-specific cause of death on death certificates. However, simulations from a recent methodological study indicate that misclassification of breast cancer-specific cause of death is likely to have relatively small impact on comparisons between groups, assuming misclassification of cancer-specific death is not differential [40]. It should be noted that cohorts from three of the contributing countries [14–16] had been analysed previously with respect to propranolol; however, over 80% of the breast cancer patients included in the pooled analysis had not been analysed previously, and these earlier analyses covered different time periods [14, 16], were based on different study designs [14, 15], used a different outcome [16] or investigated only exposure before diagnosis [14]. There were some differences in the ascertainment of medication use (five studies used dispensing records, two used GP prescribing records and one used health insurance records) and in the ascertainment of mortality (seven studies used national mortality records and one used social security records). These differences may have contributed to the heterogeneity of the association between propranolol and all-cause mortality. This was partly due to the estimate in the Belgian cohort, and after removal of this study the heterogeneity was markedly reduced, but findings for all-cause mortality were similar. In contrast, there was little evidence of heterogeneity in the association between propranolol and cancer-specific mortality.

Oestrogen receptor status was not available in all of the cohorts; however, reanalysis of the propranolol association in the Swedish and Scottish cohorts additionally adjusting for oestrogen receptor status (after including tamoxifen and aromatase inhibitors in the model) made little difference to the estimates (data not shown), suggesting that oestrogen receptor status had limited potential to confound our results. BMI was also not available. The lack of adjustment for BMI could have attenuated propranolol associations because breast cancer patients with higher BMI have worse survival [41]. Similarly, we cannot rule out the effect of residual confounding on the observed associations from other unrecorded variables (such as trastuzumab use, diet, alcohol intake and physical activity) or for variables which were recorded differently between cohorts (such as use of hormone replacement therapy).

Conclusions

In this large pooled analysis, propranolol and non-selective beta-blocker use, either before or after diagnosis, was not associated with improved breast cancer-specific or all-cause mortality.