Beta-Blocker Drug Use and Survival among Patients with Pancreatic Adenocarcinoma

Pancreatic carcinoma is the fourth leading cause of cancer-related death. Effective treatment strategies are currently lacking, and the 5-year survival is less than 5%, the lowest among all cancer patients (1). Adenocarcinoma, the most lethal form of pancreatic cancer, accounts for more than 85% of pancreatic tumors. Unsurprisingly, patients diagnosed with pancreatic cancer can experience higher levels of psychologic stress than other cancer patients (2, 3).

Accumulating evidence suggests that stress-induced sympathetic nervous system (SNS) activation of β-adrenergic receptor signaling may play a role in the regulation of various cancer types including pancreatic cancer (4–7). SNS activation elevates catecholamine levels released systemically from the adrenal glands (8) and locally from postganglionic sympathetic nerve fiber termini, including in the pancreas (5). This neurotransmitter secretion is influenced not only by stress but also by other factors including smoking, diabetes mellitus, chronic pancreatitis, alcohol consumption, obesity, and hypertension (9–11).

Catecholamines orchestrate various cell and tissue functions essential for the occurrence and progression of most common cancers (12, 13). In preclinical studies of pancreatic cancer, catecholamine stimulation has been associated with increased angiogenesis, tumor cell proliferation, and tumor growth, as well as with tumor cell invasion, migration, metastases, and inhibited apoptosis (5, 14–17), possibly due to modulation of proangiogenic and prometastatic factors as well as immunological dysregulation (7, 18, 19).

A growing body of evidence from preclinical studies of pancreatic cancer indicates that β-adrenergic–receptor blockers (14, 16, 17, 20–22), particularly nonselective types such as propranolol (14, 17), may inhibit downstream consequences of catecholamine stimulation, but clinical data from pancreatic cancer patients are still scarce (23, 24).

In a large, general population-based cohort of patients with pancreatic adenocarcinoma, we test the hypothesis that β-blockers may reduce cancer-specific mortality rates. Because previous research suggests that the tumor microenvironment may be most sensitive to this type of therapy in early stage disease (7), we investigate the associations not only in the entire cohort, but also in patients with localized disease.

A retrospectively defined cohort study utilized prospectively collected data available through registers, covering all Swedish residents. Linkage was performed using the unique personal identification number assigned to all Swedish residents.

The Swedish Cancer Register has recorded cancer diagnoses since 1958 with an estimated coverage of 96% of all malignancies, of which approximately 98% are verified morphologically (25). From this register, we identified all patients (aged 18 years or older) with a first primary pancreatic cancer diagnosis between January 1, 2006, and December 31, 2009, using the International Classification of Diseases (ICD) 10th revision code C25, as well as retrieving information on age and stage at diagnosis, tumor location, whether benign or malignant, histologic type, and healthcare region.

The Prescribed Drug Register, initiated in July 2005, includes all dispensed prescriptions in Sweden, and is updated monthly (26). Dispensed medications account for about 84% of the total drug utilization in the country and 98% of drugs for cardiovascular disease (26). The Prescribed Drug Register was used to identify β-blocker prescriptions redeemed by the cohort members using the Anatomic Therapeutic Chemical (ATC) classification system (Supplementary Table S1). We also identified other medications (Supplementary Table S2), some of which have either been previously associated with cancer outcomes (27) or may have an effect on adrenergic signaling. In addition, the number of distinct medication classes (medications with the same initial five characters of ATC classification) was used to derive a comorbidity score (28).

The Total Population Register and the Longitudinal Database of Education, Income and Occupation (LISA) provided information on migration and the level of attained education, respectively (29). Cancer-specific mortality was identified using data from the Cause of Death Register (30).

We excluded patients diagnosed with primary tumors of the pancreas other than adenocarcinoma and patients who had pancreatectomy before cancer diagnosis (Supplementary Fig. S1).

Patients were considered exposed to β-blockers if they had collected at least one prescription during the 90 days before their pancreatic cancer diagnosis, a definition that likely captures most current users, as in Sweden prescriptions normally cover a period of 30 to 90 days. β-Blocker exposure was further defined by receptor selectivity: nonselective (β1/β2-antagonists, ATC: C07AA) and cardioselective (β1-antagonists, ATC: C07AB).

The dispensed daily dose was calculated for each patient [tablet strength (mg) multiplied by number of tablets used per day]. To collapse data on different types of β-blockers, we divided the dispensed daily dose with the type-specific defined daily dose (DDD), which represents the assumed average maintenance dose per day for the drug when used for its main indication in adults (31).

In a sensitivity analysis, new users were defined as those who collected β-blockers within 90 days before their diagnosis, but had no recorded collection in the previous 6 months (in patients diagnosed on or after April 1, 2006).

Cancer-specific mortality was defined using ICD-10 code C25 as the underlying cause of death. Patients were followed from the date of pancreatic cancer diagnosis until date of emigration, death, or the study end (December 31, 2010), whichever occurred first.

Age at diagnosis was modeled using restricted cubic splines with five knots. The medication-based comorbidity score was modeled as a linear measure. We categorized education as compulsory (up to 9 years), secondary (10 to 12 years), and post-secondary (more than 12 years). Patients with missing education details were excluded from the cohort (n = 32). Tumor–node–metastasis (TNM) stage was classified into stage 1 (TNM: T1-2, N0, M0), stage 2 (TNM: T3, N0, M0 or T1-3, N1, M0), stage 3 (TNM: T4, any N, M0), and stage 4 (any T, any N, M1; ref. 32). T, N, and M categories were recorded incompletely (either T, N, or M was unspecified) for 15.3%, and were missing (if TxNxMx or all three missing) for 31.1% of patients (T, N, M recording was introduced in 2004 and improved over time). These patients were classified into “recorded incompletely” and “missing” categories. Tumor location was classified as head (ICD-10: 25.0), pancreatic duct (ICD-10: C25.3), body and tail (ICD-10: C25.1, C25.2), or unspecified (ICD-10: C25.9). Use of other medications was defined using the same approach as for β-blockers.

Patient characteristics were tabulated by β-blocker use and compared using the χ² test, Fisher exact test, or t test as appropriate based on measurement type and distribution (Table 1). Tests of statistical significance were two-sided.

We estimated the observed 6-month, 1-year, and 5-year overall survival proportions using the actuarial method. Flexible parametric survival analysis (baseline hazards were modeled using splines with 5 degrees of freedom, and meansurv option of stpm2′s predict command was used; ref. 33) estimated age-adjusted pancreatic cancer survival curves by β-blocker use (Fig. 1). Cox proportional hazards models with time since diagnosis in months as the underlying time scale were fitted to estimate HRs and 95% confidence intervals (CI) for cancer-specific mortality (Tables 2 and 3) and all-cause mortality associated with β-blocker use. The analysis was performed in the entire study cohort as well as separately in patients with and without known distant metastases (M stage recorded as M1 and M0).

The assumption of proportional hazards was evaluated by applying a Grambsch–Therneau test of the scaled Schoenfeld residuals from a Cox model (34). The multivariable fractional polynomials method (35) and Grambsch, Therneau, and Fleming (GTF) smoothed plots (36) were used to assess the functional form of age and comorbidity score in the log-hazard function.

Adjusted Cox regression models included age, education, TNM stage, tumor location, healthcare region, year of diagnosis, comorbidity score, and other medications: angiotensin-converting enzyme inhibitors (ACEi), angiotensin receptor blockers (ARB), calcium channel blockers (CCB), diuretics (thiazides, loop, other), α1-receptor blockers, NSAIDs, aspirin, statins, antidiabetic agents (metformin, insulin, other), antidepressants [selective serotonin re-uptake inhibitors, and antidepressants associated with increased norepinephrine release (including tri- and tetracyclic antidepressants, serotonin-norepinephrine reuptake inhibitors, and norepinephrine-dopamine reuptake inhibitors)], anxiolytics, and antipsychotics. Since the proportional hazards assumption was violated for sex, we allowed the baseline hazard function to vary for men and women in the adjusted models. To assess the potential influence of missing information on tumor stage, we re-ran the analysis among those with complete information on stage.

The analyses were performed using Stata version 14/SE for Windows (StataCorp). The study was approved by an ethical review board in Uppsala (DNR: 2012-361).

The study cohort comprised 2,394 patients with a primary pancreatic adenocarcinoma, of whom 517 were known to be free of distant metastases at diagnosis. The cancer diagnosis was based on histopathology in 75.4% of the patients, on cytology in 24.4% and on X-ray, scintigraphy, ultrasound, MRI, CT, or equivalent in 0.2%.

Patients were followed for a maximum of 5 years. Over 20,625 person-months of follow-up, 2,187 (91%) patients died (2,054 of pancreatic cancer, 33 of cardiovascular disease, and 100 of other causes). The median survival was 5.1 months for the overall study cohort (2.5 months for 799 patients with distant metastases, and 25.4 months for 106 patients with stage 1 disease).

The estimated 6-month, 1-year, and 5-year overall survival proportions were 46%, 25%, and 3%, respectively. The corresponding estimates among patients without distant metastases were 77%, 53%, and 9%.

Overall, 522 (22%) patients were exposed to β-blockers, and 103 (20%) among patients without known metastases at diagnosis. Most patients used noncombination β-blockers (97%), whereas very few used combination tablets [α1- and β-blocking agents (n = 10), or selective β-blocking agents with other antihypertensive agents (n = 7)]. The most commonly prescribed β-blockers belonged to the β1-cardioselective (89%) class, consisting mostly of metoprolol (n = 281), followed by atenolol (n = 132) and bisoprolol (n = 55). Only 21 patients used the nonselective β-blocker propranolol. Number of DDDs for a patient per day was calculated for 99% of β-blocker users and ranged from 0.06 to 2.67 (median, 0.50).

Table 1 shows the distribution of potential confounding and prognostic factors as well as other prescriptions at baseline among users and nonusers of β-blockers. Age at diagnosis ranged from 19 to 93 years. In general, β-blocker users tended to be older and had a higher comorbidity score. They were also more likely to have prescriptions for other antihypertensive medications, loop and other diuretics, aspirin, statins, digoxin, and antidiabetic agents.

The crude cancer-specific mortality rates (per 100 person-months) among β-blocker users and nonusers were 9.78 (95% CI, 8.90–10.75) and 10.01 (95% CI, 9.53–10.51), respectively, producing a near-null association estimate in the unadjusted analysis (HR, 0.99; 95% CI, 0.89–1.10; P = 0.792). The age-adjusted pancreatic cancer survival rate was somewhat higher among β-blocker users compared with nonusers with a corresponding HR of 0.90 (95% CI, 0.81–1.00; P = 0.058; Fig. 1, Table 2). After multivariable adjustment, β-blocker use compared with nonuse was associated with a statistically significant 21% reduced rate of cancer-specific mortality (Table 2). Stepwise adjustment showed that age, tumor stage, comorbidity score, and use of loop diuretics accounted for the most of the progression from near-null unadjusted estimates to inverse statistically significant adjusted associations between β-blocker use and cancer-specific mortality with an HR of 0.77 (95% CI, 0.69–0.87; P < 0.001).

Among patients without distant metastases at diagnosis, β-blocker users had a more pronounced mortality rate reduction, especially users of higher daily doses (Fig. 1, Table 2). Among patients with distant metastases, the magnitude of the association was weaker and statistically nonsignificant (adjusted HR, 0.88; 95% CI, 0.71–1.09; P = 0.232). These associations did not differ by sex [patients without distant metastases: adjusted HR, 0.61; 95% CI, 0.37–1.00, P = 0.055 among men (n = 255; β-blocker users n = 53); and 0.63 (95% CI, 0.39–1.02), P = 0.062 among women (n = 262; β-blocker users n = 50); P_interaction = 0.389].

The magnitude of the associations was somewhat greater for nonselective β-blockers in the overall cohort; however, no statistically significant difference was detected by receptor selectivity (Table 2).

Analysis excluding patients with unknown stage produced similar adjusted HRs for cancer-specific mortality as those in the main analysis [all with known stage (n = 1,284; β-blocker users n = 271): adjusted HR, 0.79 (95% CI, 0.66–0.94, P = 0.008); without distant metastases (n = 485; β-blocker users n = 94): HR, 0.65; 95% CI, 0.46–0.92, P = 0.014].

In the sensitivity analyses including patients diagnosed on or after April 1, 2006, using the new-user definition of exposure, β-blocker use [48 patients in the entire cohort (n = 2,246) and 11 in the metastases-free subgroup (n = 487)] compared with nonuse was still associated with a reduced cancer-specific mortality rate, with HRs of 0.79 (95% CI, 0.57–1.10, P = 0.163) in the entire cohort and 0.64 (95% CI, 0.25–1.61, P = 0.343) among patients without distant metastases.

Similar associations were observed for β-blocker use in relation to all-cause mortality (data not shown).

ACEi were the second most commonly prescribed antihypertensive medications in the study population after β-blockers, followed by CCB, ARB, thiazide diuretics, and α1-receptor blockers in descending order of frequency. Noncombination tablets accounted for most of the ACEi (89%) and ARB (69%) prescriptions.

We observed near-null associations between use of any ACEi or thiazide diuretics and cancer-specific mortality (Table 3). Use of any ARB was associated with an increased rate of cancer-specific mortality in the overall cohort.

Of the remaining medications in the multivariable model (Table 3), only anxiolytics were associated with a statistically significant reduced rate of cancer-specific mortality. This group of medications includes preparations such as benzodiazepines (37) and hydroxyzine (38) that are used for multiple indications including anxiety, insomnia, muscle relaxation, neuroses, and epilepsy (benzodiazepines comprised 72% of the used anxiolytics). Use of antidepressants that boost norepinephrine levels was associated with an increased cancer-specific mortality rate (marginally statistically significant) in the entire cohort. An increased cancer-specific mortality rate was associated with loop diuretics (both in all patients and those without distant metastases), NSAIDs, and antipsychotics (for all patients).

In this large, general population-based cohort of pancreatic cancer patients, use of β-blockers was associated with longer survival, once age was taken into account. The magnitude of the inverse association increased with further adjustment for potential confounding and prognostic factors, and was most pronounced among patients with localized disease, as observed for other cancer types (28, 39).

Widely used cardiovascular medications have previously been studied as cancer therapeutics with variable results (27, 40). Here, the statistically significant inverse association with mortality was specific to β-blockers and not other treatments for cardiovascular disease including aspirin and statins.

To our knowledge, no epidemiologic study has evaluated specifically the association of β-blocker use with pancreatic cancer survival. In a retrospective study of multiple cancers based on the UK primary care Doctors' Independent Network database (23), β-blocker use (compared with use of other antihypertensive drugs) was associated with an increased all-cause mortality risk among pancreatic cancer patients (n = 140). A subsequent study replicating these analyses in the primary care Clinical Practice Research Datalink database (n = 376) found no association (24). Neither of these studies took cancer stage into account.

Some studies, but not all (41, 42), have reported improved survival among β-blocker users for cancers including breast (28), ovarian (43), non–small cell lung (44), and malignant melanoma (39). The inconsistency of results between and within cancer types could be due to not only heterogeneity in tumor biology, but also methodological variations, including differences in exposure definitions and population selection. There may also be influences of variation in type and duration (10, 45) of β-blocker use in earlier studies.

Reduced mortality primarily among patients with nonmetastatic disease indicates the possibility of greater influence early in the disease course (7), and that inhibition of β-receptors is unlikely to influence progression once the tumor is established.

Our analysis, although limited by a low prevalence of nonselective β-blocker users, did not indicate any clear difference between selective and nonselective β-blockers. β2-Adrenoreceptors predominate in pancreatic carcinoma cell lines (21), and some preclinical studies suggest that nonselective β-blockers may be more effective cancer progression inhibitors than β1-blockers (22). However, β1 and β2 receptors are very similar, and β-blockers used most frequently in clinical practice (including metoprolol and atenolol) have rather poor β1/β2 selectivity (46, 47).

Other evidence suggests that catecholamine signaling may be involved in inhibition of cancer progression. We found that use of antidepressants that boost norepinephrine levels was associated with an increased mortality rate, whereas use of anxiolytics was associated with a reduced rate. The latter class of medications includes benzodiazepines that reduce norepinephrine release through potentiation of gamma-aminobutyric acid (GABA) at GABA-A receptors in the central nervous system (37). While β-blockers predominantly affect peripheral targets, benzodiazepines exert their influence through receptors in the central nervous system, although peripheral influences (48, 49) may also be relevant including in the pancreas. Hydroxyzine exerts its anxiolytic effect mainly as an inverse agonist of the H1 histamine receptor (38), which plays a role in regulating sympathetic nerve activity (50).

The main strengths of our study include use of high-quality prospectively collected data from registers with almost complete population coverage and follow-up in the setting of a tax-supported universal health care system. The general population-based design minimizes the risk of selection bias, and the universal health care provision substantially reduces the likelihood of findings being confounded by socioeconomic characteristics, as does adjustment for attained level of education.

Nonrandomized allocation of treatment in observational studies may produce spurious associations when treatment indications are related to future health risks. However, unlike in this study, confounding by indication is more likely to arise in studies that assess outcomes the treatment is intended to affect (51, 52). Concerns about confounding by indication were further alleviated by the varying associations observed for other antihypertensive medications. We also found that β-blocker users had higher comorbidity scores and were more likely to have received loop diuretics, widely used in the therapy of heart and renal failure. The observed inverse association with cancer-specific mortality is therefore less likely to be explained by selective avoidance of prescribing β-blockers to frail patients.

The register data allowed us to control for important potential confounding and prognostic factors. Nevertheless, the possibility of residual confounding cannot be excluded in an observational study. In order to explain the inverse association between β-blockers and mortality, an uncontrolled confounder would, however, need to be strongly associated with avoidance of β-blockers and increased mortality rate, and be sufficiently prevalent (53). For example, although some studies have reported positive associations between smoking, obesity, and mortality in patients with pancreatic cancer (54, 55), it is unlikely these characteristics are notably inversely associated with β-blocker use (56, 57). Therefore, these uncontrolled factors are unlikely to explain the observed inverse association between β-blocker use and pancreatic cancer mortality.

Our definition of exposure to β-blockers classifies patients as exposed if they redeemed their medication during the 90-day period prior to diagnosis. We acknowledge there may be changes to prescribed medication use during follow-up. Therefore, our classification may not be reflective of the β-blockers use over the entire follow-up. However, the adopted approach resembles the intention-to-treat analysis in randomized studies and eliminates the possibility of immortal time bias.

Since the drug exposure data were only available after the start of the register in July 2005, we could not evaluate cumulative exposure before diagnosis or long-term patterns of β-blocker use prior to diagnosis. In our sensitivity analysis where new of β-blocker use was assessed, the magnitude of the association was consistent with the main results, although it did not reach statistical significance.

The Prescribed Drug Register does not provide information on compliance. However, we defined use of medications based on when they were dispensed to patients at pharmacies, which along with the requirement for patients to pay a component of the cost when dispensed should increase the probability of actual use.

In conclusion, our results support the hypothesis that inhibition of β-adrenergic receptor signaling pathways may impede progression of pancreatic adenocarcinoma and suggest that β-blockers may complement existing treatment modalities for these patients. These findings may inform future randomized controlled trials to clarify if β-blockers, which have been used for decades worldwide as a cardiac medication and are regarded as safe, can be included in pancreatic cancer treatment regimens.

S. Montgomery reports receiving commercial research grant from grants to his institution from Roche, Novartis, and AstraZeneca for research into multiple sclerosis. K.E. Smedby reports receiving a commercial research grant from Janssen Cilag and is a consultant/advisory board member for Celgene. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R. Udumyan, F. Fang, U. Valdimarsdottir, A. Ekbom, K. Fall

Development of methodology: R. Udumyan, U. Valdimarsdottir

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.E. Smedby, K. Fall

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Udumyan, S. Montgomery, F. Fang, H. Almroth, U. Valdimarsdottir, A. Ekbom, K. Fall

Writing, review, and/or revision of the manuscript: R. Udumyan, S. Montgomery, F. Fang, H. Almroth, U. Valdimarsdottir, A. Ekbom, K.E. Smedby, K. Fall

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Udumyan, S. Montgomery

Study supervision: S. Montgomery, K. Fall

This study was funded by a grant from the Swedish Cancer Society (CAN 2013/650 to Dr. K. Fall).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.