Cardiovascular Outcomes of α-Blockers vs 5-α Reductase Inhibitors for Benign Prostatic Hyperplasia

This cohort study compares the cardiovascular safety profile of α-blockers vs 5-α reductase inhibitors for patients with benign prostatic hyperplasia.


Introduction
Cardiovascular diseases and benign prostatic hyperplasia (BPH) are common conditions with shared risk factors among older men. 1 The most prescribed class of medications for BPH consists of α-1 blockers (ABs), particularly selective antagonists of the α-1A adrenergic receptor (α1-A-AR) subtype. 2 Interestingly, the α1-A-AR subtype is expressed in prostate and cardiovascular tissues.Preclinical studies found that the α1-A-AR subtype was associated with cardioprotective outcomes. 3,4rthermore, the landmark Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT) 5,6 was terminated early because doxazosin (a nonselective AB) was associated with an increased risk of adverse cardiac events, most notably HF, angina, and coronary revascularization.
However, previous investigations of the cardiac safety of ABs have produced conflicting results. 3,7r 2021 study 8 found that use of ABs compared with no AB use was associated with an increased risk of mortality among patients undergoing percutaneous coronary intervention for myocardial infarction (MI).This finding highlights the need to investigate AB safety on a broader scale.In this study, we hypothesized that ABs compared with 5-α reductase inhibitors (5-ARIs), the second most prescribed medication class for BPH, would be associated with increased cardiovascular risks among Medicare beneficiaries with BPH. 2

Methods
This cohort study was approved by the University of North Carolina at Chapel Hill (UNC-CH) Institutional Review Board, which granted a waiver of informed consent because the risk to patients was no more than minimal.This report follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Data Source
This study used insurance claims from a 20% random sample of Medicare beneficiaries in the US (2007-2019).This database contains longitudinal claims on inpatient, outpatient, and prescription drug services and demographic data.We obtained data under a data use agreement with the Centers for Medicare & Medicaid Services as maintained by the UNC-CH Cecil G. Sheps Center for Health Services Research.

Study Cohort
This was an active comparator study comparing new users of ABs vs 5-ARIs for risk of adverse outcomes. 9We constructed a cohort of individuals identified as males who were aged 66 to 90 years at new use; had continuous enrollment in fee-for-service Medicare plans A, B, and D for 12 or more months (with a 45-day gap allowed) prior to new use; and had 1 or more diagnosis codes for BPH at 12 months or less prior to new use (eTable 1 in Supplement 1; eTable 24 in Supplement 2).We excluded patients with chemotherapy within 6 months, hospice care within 12 months, or prostatectomy or prostate cancer at any time prior to new use (eTables 25-27 in Supplement 2).

Exposure Assessment
We identified medications via prescription claims from Medicare part D claims.We considered 2 comparator drug classes (generic or branded): ABs (alfuzosin, doxazosin, tamsulosin, terazosin, silodosin, and prazosin) and 5-ARIs (finasteride and dutasteride) (eTable 2 in Supplement 1 and eTable 28 in Supplement 2).Medication initiation was defined as a prescription fill after January 1, 2008, without fills 12 months or longer prior.To minimize exposure misclassification due to nonadherence, we required individuals to refill the study drug 30 days or less after finishing their first prescription.We excluded individuals who filled a prescription for the other drug class or experienced an outcome between the first and second fill.Follow-up began at the second prescription fill to ensure equal follow-up across treatment groups (Figure 1).

Outcomes
1][12][13][14][15][16] Outcomes were hospitalization for heart failure (HF), composite major adverse cardiovascular events (MACE; hospitalization for stroke, MI, or death), composite MACE or HF hospitalization, and death.Hospitalization outcomes were identified from the in-patient record only (eTable 3 in Supplement 1).Death was identified using the Medicare Master Beneficiary Summary File: National Death Index Segment. 17

Covariates
9][20] Variables were identified based on claims 12 months or less prior to new use.
[23] Medications were identified using National Drug Codes.5][26] Categories included American Indian or Alaska Native, Asian or Pacific Islander, Black, Hispanic, non-Hispanic White, other (race not within other categories), and unknown.Combined race and ethnicity was included in our analyses as a proxy for processes of marginalization impacting health, not as a biological construct. 18,27,28

Statistical Analysis
0][31] Treatment was determined at drug initiation, and patients were retained in that group over follow-up, regardless of treatment changes.Explicitly, we did not censor patients at treatment discontinuation or switching.
We thus aimed to estimate the effect of initiating ABs vs 5-ARIs on adverse outcomes; however, because these are nonrandomized, secondary data and we cannot confirm that causal criteria (eg, no confounding) were met, we report associations throughout.For confounding, we calculated propensity scores (PSs) using logistic regression, modeling the probability of treatment with ABs vs 5-ARIs dependent on measured confounders (eTable 5 in Supplement 1).3][34] We assessed covariate balance before and after IPTW using absolute standardized mean differences, with those Յ0.1 indicating adequate balance.
Patients were censored only at Medicare Parts A or B disenrollment.To account for this, we calculated stabilized inverse probability of censoring weights (IPCWs), 35 using pooled logistic regression to estimate an individual's probability of not being censored at each quintile of follow-up, dependent on treatment and factors associated with censoring (using the same variables as the PS model).Stabilized IPCWs were estimated as the probability of not being censored, dependent on treatment, divided by an individual's probability of not being censored, dependent on treatment and factors associated with censoring.
We estimated cumulative incidence, risk ratios (RRs), and risk differences (RDs) from the second prescription fill through 1 year of follow-up using Kaplan-Meier and Aalen-Johansen estimators for mortality and nonmortality outcomes, respectively.Point estimates and CIs were calculated using nonparametric bootstrapping, drawing 500 random samples with replacement.RR and RD estimates and their SDs were calculated as the mean and standard error of point estimates across 500 samples.We calculated 2-sided 95% CIs, which when compared with the null value, indicate 2-sided statistical significance at P < .05.7][38][39][40] To understand the magnitude of our exposure-outcome association, we calculated the number needed to harm (NNH) as the inverse of the RD estimate for outcomes with 95% CIs that did not overlap with 0. We plotted cumulative incidence curves in the nonbootstrapped sample.

A Priori Sensitivity Analyses
First, we used asymmetric propensity score trimming to understand the robustness of study results against uncontrolled confounding. 41,42Second, to understand residual confounding by hypertension, we repeated primary analyses considering only the ABs tamsulosin and silodosin, which are not indicated to treat hypertension. 43Third, to investigate our BPH definition, we required patients have  44 Sixth, we limited our study period to start follow-up after October 1, 2015, so that all outcomes would be captured via ICD-10 codes.

Post Hoc Sensitivity Analyses
9][50] Third, we repeated primary analyses among patients with 2 or more outpatient or 1 or more inpatient claims with a BPH diagnosis.Fourth, to investigate exposure-outcome associations among patients with a poor cardiovascular profile, we repeated the primary analysis among patients with 1 or more inpatient hospitalizations for MI, stroke, or HF within 12 months prior to cohort entry.Finally, to investigate confounding by socioeconomic status, we repeated primary analyses additionally controlling for eligibility for Medicaid dual enrollment 51 and the Medicare Part D low-income subsidy. 52

Results
Among days prior to study drug initiation was comparable to the primary cohort (eTable 13 in Supplement 1), as were treatments (eFigure 10 in Supplement 1).Associations were attenuated when additionally controlling for indicators of BPH severity (eTable 14 in Supplement 1).Associations were consistent with primary results after excluding patients with prior anticoagulant use (eTable 15 in Supplement 1).
Similarly, results were comparable, although less precise, when restricting to new-use episodes on or after October 1, 2015 (eTable 16 in Supplement 1).
In a post hoc sensitivity analysis evaluating risk of hospitalization for injury or poisoning, we found an RR of 1.06 (95% CI, 0.99 to 1.13) and an RD per 1000 individuals of 2.87 (95% CI, −0.42 to a Distribution is given in the primary cohort after trimming nonoverlapping regions of the propensity score distributions across 2 treatment groups.Results are presented before and after applying IPTWs.
b Patients who experienced heart failure or a major adverse cardiovascular event between their first and second prescription fill are included.The sum of the standardized weight among the AB treatment group was 163 840, and the sum among the 5-ARI group was 26 009.
c CIs were calculated using the proc surveymeans procedure in SAS statistical software using bootstrapping (with 250 resamples) to account for interdependence caused by the use of IPTWs.
d These variables were not included in the primary propensity score model.
e Identified as 2 or more outpatient or 1 or more inpatient claims with a diagnosis code for heart failure in any position.
6.16) for AB vs 5-ARI groups (eTable 17 in Supplement 1).Estimates from the quantitative bias analysis for misclassified smoking were attenuated from primary results for composite MACE and HF hospitalization (RR range, 0.94 to 1.03 for AB vs 5-ARI groups), although comparable when adjusted for misclassified obesity (RR range, 1.07 to 1.11 for AB vs 5-ARI groups) (eTable 18 in Supplement 1).Outcomes were A, hospitalization for heart failure (HF), B, major adverse cardiovascular events (MACE), C, composite MACE or hospitalization for HF, and D, all-cause mortality.5-ARI indicates 5-α reductase inhibitor; AB, α-blocker.

Discussion
In this large cohort study of Medicare beneficiaries with BPH, we found that initiation of ABs compared with 5-ARIs was associated with an increased risk of death, MACE, and composite MACE and HF hospitalization.We did not identify an increased risk for HF hospitalization alone.IPCWadjusted risks of these outcomes were low.To our knowledge, this is the largest study assessing cardiovascular risk in patients taking ABs and the first to focus on mortality and MACE among patients with BPH.
Residual confounding could contribute to these results.However, even a small difference in cardiovascular outcomes associated with ABs vs 5-ARIs would have substantial public health implications given how widely ABs are prescribed for BPH.Per our NNH estimate, for every 136 individuals (95% CI, 84-348 individuals) with BPH treated with ABs over 5-ARIs, we would expect 1 additional MACE or HF hospitalization within 1 year after initiation.To provide perspective, 163 829 individuals in our 20% Medicare sample (86.3%) were prescribed ABs over 5-ARIs.When this number is divided by the NNH of 136 individuals, this is potentially associated with 1205 additional events over 1 year.The importance of these findings may also be amplified because most patients with BPH require long-term medical therapy; thus, the risk of AB exposure could be cumulative.
Furthermore, more than 5 million people filled a prescription for tamsulosin alone in the US in 2020, underscoring the potential magnitude of outcomes associated with these medications. 53r findings regarding HF potentially contrast with those of ALLHAT, wherein a 2-fold risk of incident HF led to discontinuation of the nonselective AB group.Initial hypotheses regarding adverse effects of doxazosin pointed to less effective blood pressure control in this population compared with individuals receiving chlorthalidone.However, further analyses did not support this. 54Apparent differences between the 2 studies also could arise from ALLHAT's examination of hospital and home-treated HF, whereas we considered only HF hospitalization.Furthermore, our cohort's mean age was approximately 7 years older than that of participants in ALLHAT and included only male patients with an established diagnosis of BPH, whereas ALLHAT included male and female participants.
Our finding that ABs were not associated with HF hospitalization risk may also appear to contrast with an RR of 1.10 among users of ABs compared with those using 5-ARIs reported by Lusty et al. 55 However, the Ontario study focused on new diagnosis of HF, whereas we studied HF hospitalization.Furthermore, they indexed patients into the study at the time of BPH diagnosis, introducing potential selection bias that our approach avoids. 9,56A meta-analysis of randomized trials among patients with indications for ABs by Sousa et al 57 found that patients receiving ABs compared with those in active or placebo control groups had increased risk of acute HF (odds ratio = 1.78 [95% CI, 1.46-2.16]).They found no association with mortality (odds ratio = 1.10 [95% CI, 0.84-1.42]),although they did not analyze outcomes with respect to a relevant comparator group as we have done with patients taking 5-ARIs.
In contemporary practice, the most prescribed ABs for BPH are α-1A subtype-selective antagonists that have minimal impact on blood pressure. 58We found that increased risks of adverse cardiovascular outcomes were maintained when the exposure was limited to subtype-selective ABs in sensitivity analyses.This finding was unsurprising given that 141 398 ABs (86.3%) prescribed in our study were subtype selective.Collectively, these data align with the concept that activation of endogenous cardiac α1-A-ARs is cardioprotective, as demonstrated in preclinical studies. 3,59,60

Strengths and Limitations
This study has important strengths.Our source population was Medicare beneficiaries enrolled in Parts A and B fee-for-service plans.Given that approximately 54% of Medicare beneficiaries were enrolled in these plans in 2019, 61 our study represents a substantial percentage of US adults aged 65 years or older.Moreover, we adhered to prespecified analyses and used rigorous methodology, such as IPTWs and IPCWs to account for systematic bias.We conducted a new-user, active comparator cohort study that minimized confounding by indication and avoided immortal time bias. 9,620][31] Finally, our choice of active comparator is clinically relevant: ABs and 5-ARIs are first-line treatments for BPH, but adverse cardiac outcomes are hypothesized only with ABs. 3 This choice ensured that our study addressed a pertinent clinical question and so may provide valuable insights into the safety of medications commonly prescribed for BPH management.
It is important to acknowledge several limitations in our study.First, like other studies using insurance claims data, this study is susceptible to variable misclassification.In sensitivity analyses, we found that our results were robust against poorly classified obesity but that if smoking operated as a confounder beyond what we have captured, residual confounding may explain our results.
4][65] Given that we controlled for COPD, there may be less residual confounding than this sensitivity analysis suggested.Second, despite substantial effort to control for confounding, some confounding may persist. 7Our results were consistent across multiple sensitivity analyses (eg, asymmetric PS trimming), although when controlling for indicators of BPH severity, for example, we found attenuated point estimates.Given the sizes of our association estimates, a post hoc negative control outcome analysis was uninformative (eAppendix in Supplement 1).Third, we evaluated only 1 year of follow-up; we were concerned about our ability to capture reasons for treatment changes (eg, BPH symptom severity), 35 and treatment changes were common.Furthermore, we did not consider medication dosing given that clinicians often instruct patients to take a different number of the same pill dosage according to their BPH symptoms, which cannot be captured in claims.Fourth, this study may not be generalizable to all patients with BPH.We excluded beneficiaries enrolled in Medicare Advantage plans, required 2 or more medication fills, and included claims only through 2019.These limitations should not impact our study's internal validity; however, any differences in the distribution of variables that modify the effect of ABs vs 5-ARIs on cardiovascular outcomes (ie, effect measure modifiers) will impact the transportability of our results to other populations. 66,67

Conclusions
This cohort study is the largest study of the association of ABs with cardiovascular events to date, to our knowledge.We found that new prescription of ABs compared with 5-ARIs was associated with a higher risk of all-cause mortality and MACE among patients with BPH.Although we present the most extensive analysis, to our knowledge, of the cardiovascular safety of ABs heretofore, further investigation with more detailed clinical data is warranted to guide ongoing clinical practice.
To ensure comparable outcome identification across the transition from International Classification of Diseases, Ninth Revision (ICD-9) to International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes (October 2015), we visually assessed monthly prevalence estimates for each outcome among all Medicare enrollees from 2013 to 2017 (eFigures 1-3 in Supplement 1).

Figure 1 .
Figure 1.Study Design Diagram With Inclusion and Exclusion CriteriaFirst prescription fill

Furthermore, we evaluated
treatment discontinuation and switching patterns to understand treatment persistence.Analyses were conducted using SAS statistical software version 9.4 (SAS Institute).Data were analyzed from January 2007 through December 2019; follow-up started January 2008.

Figure 2 .No
Figure 2. Cumulative Incidence for Primary Outcomes

Table 2 .
Association of Treatment With Study Outcomes aResults were consistent among a population with 2 or more outpatient or 1 or more inpatient claims with a BPH diagnosis code (eTables 19 and 20 in Supplement 1).Estimates among patients with prior hospitalization for HF, MI, or stroke were too imprecise for meaningful conclusions (eTables 21 and 22 in Supplement 1).Finally, controlling for socioeconomic status yielded generally consistent results (eTable 23 in Supplement 1).
Abbreviations: 5-ARI, 5α-reductase inhibitor; AB, α-blocker; HF, heart failure; IPTW, inverse probability of treatment weight; MACE, major adverse cardiovascular event; RD, risk difference; RR, risk ratio.aResultsare among the primary study population after trimming nonoverlapping regions of the propensity score distributions comparing initiators of ABs vs 5-ARIs.Results are presented with and without IPTWs; all results incorporate inverse probability of censoring weights.JAMA Network Open | CardiologyCV Outcomes of α-Blockers vs 5-α Reductase Inhibitors for Benign Prostatic Hyperplasia JAMA Network Open.2023;6(11):e2343299. doi:10.1001/jamanetworkopen.2023.43299(Reprinted) November 14, 2023 8/16 Downloaded from jamanetwork.comby guest on 11/17/2023 This is an open access article distributed under the terms of the CC-BY License.© 2023 Zhang J et al.JAMA Network Open.Seamans MJ, Hong H, Ackerman B, Schmid I, Stuart EA.Generalizability of subgroup effects.Epidemiology.2021;32(3):389-392.doi:10.1097/EDE.000000000000132967.Webster-Clark M, Breskin A. Directed acyclic graphs, effect measure modification, and generalizability.Diagnosis and Procedure Codes for Inclusion and Exclusion Criteria eTable 2. National Drug Codes Used to Identify Study Exposures eTable 3. Diagnoses Codes Used to Identify Study Outcomes eTable 4. Diagnosis Codes, Procedure Codes, and Drug Names Used to Identify Confounders and Their Operationalization eFigure 1. Month-Level Prevalence of Primary Outcome Definition for Inpatient Hospitalization for Heart Failure Over ICD-9 to ICD-10 Transition eFigure 2. Month-Level Prevalence of Primary Outcome Definition for Inpatient Hospitalization for Stroke Over ICD-9 to ICD-10 Transition eFigure 3. Month-Level Prevalence of Primary Outcome Definition for Inpatient Hospitalization for Myocardial Infarction Over ICD-9 to ICD-10 Transition eFigure 4. Directed Acyclic Graph Used to Identify Important Potential Confounders in This Study eTable 5. Logistic Regression Model Specification Used to Generate Propensity Scores and Inverse Probability of Censoring Weights in Primary Analyses eTable 6. Diagnosis Code List to Identify Hospitalization Due to Injury or Poisoning as Negative Control Outcome eMethods.Quantitative Bias Analysis for Smoking and Obesity eFigure 5. Diagram Demonstrating Flow of Study Patients and New-Use Episodes Through Inclusion and Exclusion Criteria eFigure 6. Annual Estimates of Proportion of New-Use Episodes Attributed to Each Study Drug in Primary Population eFigure 7. Propensity Score Distributions Prior to Trimming Patients eFigure 8. Propensity Score Distributions After Trimming Nonoverlapping Propensity Scores and Refitting Logistic Regression Model in Included Patient Population Without Bootstrapping eTable 7. Descriptive Statistics of Stabilized Inverse Probability of Treatment Weights in Primary Patient Population Without Bootstrapping eFigure 9. Histogram of No. Days From Second Fill Date Until Discontinuation, Fill for Other Drug Class, or Censoring eTable 8. Counts and Percentages of Medications Attributed to New-Use Episodes by Year in Primary Study Population eTable 9. Descriptive Statistics of Days From Second Prescription Fill Until Discontinuation, Filling Prescription for Other Study Drug, or Censoring by Initial Treatment and Amount of Follow-Up eTable 10.Primary Study Results After 1-y Follow-Up Among Population After Asymmetric Propensity Score Trimming eTable 11.Descriptive Table of Study Population Limited to New-Use Episodes by New Users of α-Blockers Attributed to Tamsulosin or Silodosin eTable 12.Primary Study Results After 1-y Follow-Up Limited to New-Use Episodes Among α-Blockers Attributed to Tamsulosin and Silodosin, Selective α-1A Adrenergic Receptor Antagonists eTable 13.Descriptive Statistics of Unweighted Patient Population With Benign Prostatic Hyperplasia Diagnosis in 180 d Prior to New-Use Episode eFigure 10.Distribution of Medication Fill at an Individual's Index Date Into the Cohort Across Calendar Years eTable 14.Primary Study Results After 1-y Follow-Up Additionally Controlling for Indicators of Severity of Benign Prostatic Hyperplasia eTable 15.Study Results After Removing Patients With History of Anticoagulant Fill Within 1-y Prior to New-Use Episode eTable 16.Study Results Restricting to New-Use Periods on or After October 1, 2015 eTable 17.Study Estimates for Hospitalization for Injury or Poisoning as Negative Control Outcome eTable 18. Results from Quantitative Bias Analysis eTable 19.Descriptive Statistics of Study Population Restricted to Individuals With Ն2 Outpatient or Ն1 Inpatient Diagnosis Codes for Benign Prostatic Hyperplasia eTable 20.Study Results Among Patients With Ն2 Outpatient or Ն1 Inpatient Diagnosis Codes for Benign Prostatic Hyperplasia eTable 21.Descriptive Statistics of Study Population Restricted to Individuals With Ն1 Inpatient Hospitalization for Myocardial Infarction, Stroke, or Heart Failure Within 1 y Prior to Cohort Entry eTable 22. Study Results Among Patients With History of Hospitalization for Heart Failure, Myocardial Infarction, or Stroke Within 1 y Prior to New-Use of Study Medications eTable 23.Study Results Among Primary Study Population Additionally Adjusting for Dual Eligibility for Medicaid and Receipt of Medicare Part D Low-Income Subsidy eAppendix.Discussion on Limitations of Hospitalization for Injury or Poisoning as Negative Control Outcome Analysis eReferences.