Non-selective Beta Blockers, Hepatic Decompensation and Mortality in Cirrhosis: A National Cohort Study

Abstract

Background & Aims:

Little is known about the effectiveness of nonselective beta blockers (NSBBs) in preventing hepatic decompensation in routine clinical settings. We investigated whether NSBBs are associated hepatic decompensation or liver-related mortality in a national cohort of Veterans with Child-Turcotte-Pugh (CTP) A-cirrhosis with no prior decompensations.

Approach & Results:

In an active comparator, new user design, we created a cohort of new users of carvedilol (n=123) vs. new users of selective beta blockers (SBBs) (n=561) and followed patients for up to 3 years. An inverse probability treatment weighting (IPTW) approach balanced demographic and clinical confounders. The primary analysis simulated intention-to-treat (“pseudo-ITT”) with IPTW-adjusted Cox models; secondary analyses were pseudo-as-treated – both were adjusted for baseline and time-updating drug confounders. Subgroup analyses evaluated NSBB effects by hepatitis C (HCV) viremia status, CTP class, platelet count, alcohol-associated liver disease (ALD) etiology, and age. In pseudo-ITT analyses of carvedilol vs. SBBs, carvedilol was associated with a lower hazard of any hepatic decompensation (HR 0.59, 95% CI 0.42–0.83) and the composite outcome of hepatic decompensation/liver-related mortality (HR 0.56, 95% CI 0.41–0.76). Results were similar in pseudo-as-treated analyses (hepatic decompensation: HR 0.55, 95% CI 0.33–0.94; composite outcome: HR 0.62, 95% 0.38–1.01). In subgroup analyses, carvedilol was associated with lower hazard of primary outcomes in the absence of HCV viremia, higher CTP class and platelet count, younger age, and ALD etiology.

Conclusions:

There is an ongoing need to noninvasively identify patients who may benefit from NSBBs for the prevention of hepatic decompensation.

Graphical Abstract

INTRODUCTION

The presence of portal hypertension in cirrhosis portends clinical decompensation with ascites, hepatic encephalopathy, and variceal hemorrhage, and results in high morbidity and mortality.^{1, 2} Clinically significant portal hypertension (CSPH), defined as a hepatic venous pressure gradient (HVPG) ≥10 mmHg, is required for the development of ascites and variceal hemorrhage,^{3, 4} therefore, the prevention of CSPH and its sequelae is clinically important. Non-selective beta-blockers (NSBBs) reduce portal venous inflow and reduce portal hypertension.^{5, 6} The recent PREDESCI trial showed an 11% absolute risk reduction in the incidence of ascites over a median follow-up of 37 months among patients with compensated cirrhosis and CSPH treated with propranolol or carvedilol compared to placebo.⁷ PREDESCI trial subjects were carefully selected with baseline HVPG measurements and underwent an open-label NSBB titration period. However, data are limited regarding associations between NSBBs and decompensation in clinical settings where HVPG measurement is not readily available.

NSBBs are part of the standard of care for primary and secondary prevention of variceal hemorrhage,⁸ and are widely prescribed in cirrhosis. Furthermore, carvedilol is a widely prescribed for hypertension and other cardiovascular indications, thus making it highly practical to investigate its effects on adverse events in cirrhosis irrespective of the original prescription intent. The Baveno VII expert consensus in portal hypertension recommends carvedilol as the preferred NSBB given more effective reductions in HVPG and a better side effect profile, although large-scale studies are limited to date.⁹ To fill this gap, our study aims were to: 1) investigate the association between NSBBs, hepatic decompensation, and liver-related mortality in a national cohort of Veterans with cirrhosis,¹⁰ and 2) evaluate effects of NSBBs among clinically relevant subgroups. We performed an active comparator, new user design among patients with cirrhosis newly prescribed carvedilol who were and compared outcomes to a matched cohort who were prescribed beta-1 selective beta blockers (SBBs; atenolol, metoprolol, bisoprolol, nebivolol).¹¹

METHODS

Study Design and Data Source

We conducted a retrospective cohort study of the Veterans Health Administration (VHA), which is the single largest integrated health-care system in the U.S. and the largest healthcare provider to Veterans with cirrhosis. We used data from the Veterans Outcomes and Costs Associated with Liver Disease (VOCAL) cohort, derived from the VA Corporate Data Warehouse (CDW) – a national database with longitudinal demographic, outpatient and inpatient pharmacy, claims, clinical laboratory and imaging data on Veterans across all 50 states.¹² The VOCAL cohort includes nearly 130,000 Veterans diagnosed with cirrhosis from 1/1/2008 to 12/31/2018 and has been the source of many pharmacoepidemiologic studies. Details on cohort creation and cirrhosis identification have been previously published. ^{10, 13, 14, 15} Cirrhosis was ascertained using a validated algorithm (one inpatient or two outpatient ICD-9/10 codes for cirrhosis [571.2, 571.5, K74.6x, K70.3x]).¹⁶ The Institutional Review Board at the Corporal Michael J. Crescenz VA Medical Center in Philadelphia approved the study. All data management and analyses were performed using structured query language (SQL) and Stata/BE 17.0 (College Station, TX).

Eligibility Criteria

Our central study objective was to evaluate whether NSBBs are associated with hepatic decompensation and liver-related mortality among patients with well-compensated cirrhosis at risk for hepatic decompensation. At first, we created two separate subcohorts, comparing carvedilol to SBBs, and comparing propranolol/nadolol to SBBs. However, given that in our cohort propranolol and nadolol were highly linked to hepatic decompensations and are largely prescribed for secondary prophylaxis of portal hypertensive bleeding, we focused analyses on carvedilol as we were interested in evaluating primary prevention of hepatic decompensation.

Patients were excluded if they received liver transplantation or were diagnosed with hepatocellular carcinoma (HCC) prior to the index date, had less than 180 days of follow-up, or had fewer than two VHA outpatient visits in the index year as we wanted to study patients actively engaged in VA care.^{10, 15, 17} Baseline laboratory values were time-updated and obtained from the same calendar quarter as the index drug exposure date. We included patients with Child-Turcotte-Pugh (CTP) A cirrhosis. We excluded patients with metastatic cancer, baseline serum creatinine of > 2 mg/dL, total bilirubin > 3 mg/dL, and platelet count < 30,000/μL or > 150,000/μL. We also excluded patients with CTP B or C cirrhosis using our electronic-medical record (EMR)-validated, estimated CTP (eCTP) score, which is based on a combination of laboratory values (total bilirubin, albumin, INR), diagnosis codes, and pharmacy fill data (ascites, hepatic encephalopathy medications) and diagnosis codes of variceal hemorrhage.^{14, 18} We excluded patients who never received NSBBs or SBBs, and patients with NSBB or SBB exposure prior to the cirrhosis diagnosis. Patients were also excluded if prescribed vitamin K antagonists or direct acting oral anticoagulants prior to the index date. To ensure at least 3 years of follow-up, we included Veterans with NSBB index data from January 1^st, 2008 to December 31^st, 2017 with follow-up data available through December 31^st, 2020. Finally, patients were excluded if they had key missing data to adjudicate the above selection criteria.

Active comparator, new user design rationale and description

We used an active comparator, new user (ACNU) design (Figure 1) to help reduce some key challenges in observational studies: 1) confounding by indication (i.e., a patient prescribed a drug may have more severe disease and worse clinical outcomes), and 2) healthy user bias (i.e., a patient is more likely to be prescribed a drug due to being perceived as healthier, and healthier patients are more likely to adhere to a drug).^{11, 19, 20} Detailed prescription data were obtained from the VA pharmacy database and updated at 30-day intervals throughout follow-up. Eligible patients were considered NSBB-exposed if they had new prescriptions for carvedilol and had no history of any other NSBB use prior to cirrhosis diagnosis. Active comparators were new users of SBBs (atenolol, metoprolol, bisoprolol, labetalol) after the cirrhosis diagnosis. SBBs were selected as an active comparator as they are commonly used and provide a rational comparison to NSBBs regarding mechanisms of action.²¹ To increase the likelihood that the prescribed medication had ongoing use, we further restricted the cohort to patients with at least two NSBB or SBB prescriptions within a 180-day baseline period, consistent with accepted methodology.^{19, 22} The study initiation date was 180 days after the first prescription date, which had to be after the first diagnosis code of cirrhosis.²² Given that this was a secondary data analysis where the diagnosis date entered into the EMR may not be exact, we used a time lag to mitigate protopathic bias. This type of bias can give the appearance of reverse causality, if for example, carvedilol is initiated as secondary prophylaxis in response to a bleeding event, but the diagnosis code for that bleeding event in the EMR lags the actual event. To overcome this limitation, this study includes a time-lag and a ‘run-in’ baseline period of 180 days during which patients had to be free of early events to be included in the study.).²³ To isolate the effect of carvedilol versus active comparators, we excluded patients treated with both classes of medications within 180 days of each other.

Figure 1 –

Overview of the Active Comparator New User Design

Covariate Data

For each patient we obtained age, sex, race, body mass index (BMI), medical comorbidities, psychosocial factors, liver disease etiology and severity. Medical comorbidities were identified by diagnosis codes and/or by clinical data from the EMR.¹⁷ Diabetes was defined using a previously validated VA algorithm²⁴ and obesity was defined as having body mass index >30 kg/m² at baseline. The cirrhosis comorbidity index (CIRCOM) used for comorbidity adjustment includes ICD9-CM/ICD-10-CM diagnosis codes for acute myocardial infarction, peripheral arterial disease, epilepsy, substance abuse other than alcoholism, heart failure, cancer, and chronic kidney disease.²⁵ Alcohol use was ascertained using the Alcohol Use Disorders Identification Test-Consumption (AUDIT-C), which is annually administered in VA primary care clinics; a score of ≥4 for men and ≥3 for women was classified as alcohol misuse.²⁶ Tobacco use was obtained from the VA “Health Factors” file, which is updated annually by primary care providers. Liver disease etiology and the eCTP score were defined using previously validated methodology.^{13, 27} Total serum bilirubin, serum albumin, INR and platelet count were obtained from the laboratory database; model for end-stage liver disease-sodium (MELD-Na) was calculated. Additional covariates included hepatitis C virus (HCV) viremia based on HCV RNA quantitation and systolic blood pressure (SBP) at baseline in mmHg. Given changes in clinical care and guideline recommendations over time (e.g., the advent of all-oral HCV therapy; changes in recommendations for variceal prophylaxis), calendar year of drug initiation, and time from cirrhosis diagnosis to medication initiation were recorded. Finally, for subsequent modeling purposes, several key confounders were time-updated at 30-day windows throughout follow-up. This included statin and angiotensin converting enzyme inhibitor (ACEI) or angiotensin-receptor blocker (ARB) medication use, coronary artery disease, heart failure, diabetes, atrial fibrillation, and HCV viremia.

Study Outcomes

The primary outcomes were time to any hepatic decompensation and a composite endpoint of hepatic decompensation (ascites, variceal hemorrhage, or hepatic encephalopathy) or liver-related mortality at 3 years (Supplemental Table 1). We estimated liver-specific mortality given the high cardiovascular comorbidity burden in the veteran cohort and multiple non-liver-related competing risks of death. Because detailed cause of death data were not available for the complete cohort, to estimate liver-related mortality we constructed a multivariable logistic regression model in a separate VOCAL subcohort of 8,006 patients with cirrhosis and previously-adjudicated cause of death data (liver-related versus non-liver-related death). This model achieved an area under the receiver operating characteristic curve of 0.753 (Supplemental Figure 1); the Youden index was used to identify an optimal prediction cut point of 0.45, which was then used to generate predictions in the current dataset to classify probable liver-related versus non-liver-related mortality (i.e., if a predicted probability of liver-related death was ≥0.45 using the prediction model in a patient experiencing death, the death was classified as liver-related). Time to the individual components of hepatic decompensation, liver-related mortality alone, and all-cause mortality were secondary outcomes.

Primary Statistical Analysis

Descriptive statistics for all covariates were stratified and reported as medians and P25 and P75) for continuous data and as counts and percentages for categorical data. Statistical comparisons were made using the Wilcoxon rank-sum and Chi-squared tests for continuous and categorical data, respectively. Crude three-year event rates for all outcomes were presented for each cohort, stratified by drug exposure. To visualize the unadjusted survival distributions between groups for the primary outcome, we plotted cumulative incidence curves for each cohort. The log-rank test was used to compare distributions.

For adjusted analyses, we incorporated several advanced inferential methods to comprehensively account for potential confounding and bias. The primary approach was IPTW-adjusted Cox regression analysis using an intention-to-treat (ITT) framework¹¹, where patients classified with a given drug exposure were assumed to remain exposed throughout follow-up. In secondary analyses, drug exposure was evaluated “pseudo-as-treated” following established methodology.²² In this framework, patients were considered exposed until drug discontinuation, a switch to the comparator drug (e.g., from NSBB to SBB), or censoring due to occurrence of study outcome or end of follow-up. We gave a grace period and counted prescription coverage interruptions of ≤30 days as ongoing exposure. We used a data smoothing approach where if a patient had an isolated 30-day window where they did not have a covering prescription, but did have coverage both before and after that window, they were considered to be exposed for the bridging period. For each model, the hazard ratio (HR) and 95% confidence interval (CI) were reported. An alpha threshold of 0.05 was considered statistically significant.

To evaluate the association between carvedilol and the outcomes of interest, we used a survival analysis approach. Maximum follow-up was three years from drug initiation and patients were right-censored at liver transplantation, which occurred infrequently. To create pseudopopulations balanced across key demographic and clinical variables, and thereby minimize confounding between drug exposure groups, we used an inverse probability treatment weighting (IPTW) approach. We included the following variables in a logistic regression model to generate a propensity score (PS) for receipt of carvedilol versus the active comparator: age, sex, race, alcohol misuse, smoking status, BMI, etiology of liver disease, diabetes, coronary artery disease, heart failure, chronic kidney disease, chronic obstructive pulmonary disease, systolic blood pressure, HCV viremia status, MELD-Na, albumin, platelet count, time from cirrhosis diagnosis to medication start, and year of medication initiation. Inverse probability weights were then generated as 1/PS for patients who received carvedilol and as 1/(1-PS) for those who received the active comparator. ²⁸ Standardized mean differences (SMDs) were plotted for both unadjusted and IPTW-adjusted covariates. Adequate balance was defined as SMD of +/−0.1 consistent with best practice recommendations.²⁹ All models in pseudo-ITT and pseudo-as-treated analyses were adjusted for time-updated statin use, angiotensin-converting enzyme inhibitors /angiotensin-receptor blocker use, coronary artery disease, heart failure, atrial fibrillation, diabetes mellitus, and HCV viremia. For covariates where adequate balance could not be achieved through IPTW (SMD > +/−0.1), these variables were additionally included in multivariable models.

Subgroup Analyses

NSBBs have been shown to prevent hepatic decompensation among patients with CSPH, however, gold standard diagnostics such as HVPG or noninvasive surrogate tools such as liver stiffness by transient elastography are not universally available for clinical use. Given this limitation, we planned several a-priori, exploratory subgroup analyses by baseline HCV viremia, CTP score (5 vs. 6), platelet count (30–99 X10⁹/L vs. 100–150 X10⁹/L), alcohol-associated liver disease (ALD), and age ≥ 65 vs. < 65. Adjusted HRs and 95% CIs were plotted for each model in each cohort.

Sensitivity Analyses

Since Veterans have a high proportion of non-liver related comorbidities, death from non-liver causes may be an important competing risk for hepatic decompensation. Using the Fine and Gray approach, we conducted an analysis for hepatic decompensation with death as a competing risk and have generated cumulative incidence curves similar to the primary analyses.³⁰ To assess for the effect of residual unmeasured confounding, we computed the E-value, which is a value on the risk ratio scale and is defined as the minimum strength of association that an unmeasured confounder would need to have with the exposure and the outcome to negate the observed associations. The minimum E-value is 1; high E-values indicate that substantial unmeasured confounding would be needed to render the observed associations null.^{31, 32}

RESULTS

Cohort Characteristics and IPTW Adjustment

Supplemental Figure 2 shows the sample size achieved in the cohort. After applying exclusion criteria, the final analytic cohort was 684 patients with 123 new carvedilol users and 561 new SBB users (454 metoprolol [77.6%], 118 atenolol [20.2%], 9 labetalol [1.5%], 4 bisoprolol [0.7%]). Unadjusted baseline characteristics prior IPTW are shown in Table 1. The median age was 62–63. As expected, the Veteran cohort was predominantly male, and was ethnically diverse with greater than 20% of Black and Hispanic patients. The most common liver disease etiologies were HCV and ALD reflective of liver disease epidemiology in the VA with about 40% of patients HCV-viremic at the start of follow-up. There was a high baseline prevalence of medical comorbidities with greater than 60% diabetes, more than one third having CKD, and 46% with COPD. Median platelet count was about 100 × 10⁹/L, media MELD-Na score was 8 with preserved liver synthetic function (albumin, total bilirubin, and INR) and median creatinine was 0.9–1.0 mg/dL. Prior to adjustment, baseline differences were noted in the prevalence of HCV viremia, heart failure, atrial fibrillation, and CKD. However, excellent covariate balance was achieved after IPTW adjustment (Figure 2) for nearly all of the demographic and clinical features except sex, former smoking status, AUD, BMI, atrial fibrillation, and COPD. Thus, these variables were additionally adjusted for in multivariable models.

Table 1 –

Cohort characteristics prior to inverse probability treatment weighting

	Carvedilol vs. SBB
Factor	Carvedilol (n=123)	SBB (n=561)	p-value
Age, median (P25, P75)	63 (58, 68)	62 (57, 66)	0.24
Male Sex	122 (99.2%)	550 (98.0%)	0.38
Race			0.65
White	72 (58.5%)	352 (62.7%)
Black	30 (24.4%)	111 (19.8%)
Hispanic	11 (8.9%)	40 (7.1%)
Asian	2 (1.6%)	8 (1.4%)
Other	8 (6.5%)	50 (8.9%)
Smoking Status			0.25
Never smoker	34 (28.1%)	184 (33.0%)
Former smoker	48 (39.7%)	178 (32.0%)
Current smoker	39 (32.2%)	195 (35.0%)
Alcohol Use Disorder	21 (17.1%)	136 (24.2%)	0.087
BMI, median (P25, P75)	30.0 (25.2, 34.1)	29.0 (25.5, 33.0)	0.38
Etiology of Liver Disease			0.76
HCV	25 (20.3%)	129 (23.0%)
HBV	2 (1.6%)	5 (0.9%)
EtOH	36 (29.3%)	147 (26.2%)
HCV+EtOH	35 (28.5%)	182 (32.4%)
NAFLD	22 (17.9%)	81 (14.4%)
Other	3 (2.4%)	17 (3.0%)
HCV Viremia	40 (32.5%)	246 (43.9%)	0.021
Obesity	61 (49.6%)	239 (42.6%)	0.16
Diabetes Mellitus	78 (63.4%)	343 (61.1%)	0.64
Coronary Artery Disease	45 (36.6%)	172 (30.7%)	0.20
Heart Failure	52 (42.3%)	95 (16.9%)	<0.001
Atrial Fibrillation	13 (10.6%)	116 (20.7%)	0.009
CKD	51 (41.5%)	171 (30.5%)	0.018
COPD	56 (45.5%)	256 (45.6%)	0.98
CIRCOM Category			0.11
1+0	19 (15.4%)	118 (21.0%)
1+1	33 (26.8%)	176 (31.4%)
3+0/3+1	71 (57.7%)	267 (47.6%)
Systolic BP mmHg, median (P25, P75)	132 (118, 148)	133 (121, 147)	0.77
Sodium mmol/L, median (P25, P75)	138 (136, 140)	138 (136, 140)	0.70
Creatinine mg/dL, median (P25, P75)	1.0 (0.9, 1.3)	0.9 (0.8, 1.1)	<0.001
Albumin g/dL, median (P25, P75)	3.6 (3.3, 4.0)	3.7 (3.3, 4.0)	0.56
T. Bilirubin mg/dL, median (P25, P75)	0.9 (0.7, 1.4)	0.9 (0.6, 1.3)	0.54
Platelet Count X 109/L, median (P25, P75)	102 (84, 123)	106 (84, 131)	0.22
INR, median (P25, P75)	1.1 (1.1, 1.3)	1.1 (1.0, 1.3)	0.55
MELD-Na, median (P25, P75)	8 (6, 12)	8 (6, 11)	0.23
Time from Cirrhosis Diagnosis to Med Start (days), median (P25, P75)	510 (210, 900)	360 (210, 750)	0.057
Cirrhosis diagnosis year			<0.001
2008	0 (0.0%)	13 (2.3%)
2009	2 (1.6%)	27 (4.8%)
2010	5 (4.1%)	51 (9.1%)
2011	5 (4.1%)	55 (9.8%)
2012	9 (7.3%)	60 (10.7%)
2013	15 (12.2%)	63 (11.2%)
2014	9 (7.3%)	75 (13.4%)
2015	19 (15.4%)	75 (13.4%)
2016	27 (22.0%)	78 (13.9%)
2017	32 (26.0%)	64 (11.4%)

Abbreviations: BMI-body mass index, CIRCOM-cirrhosis comorbidity index, CKD-chronic kidney disease, COPD-chronic obstructive pulmonary disease, IPTW-inverse probability of treatment weighting, P25–25^th percentile, P75–75^th percentile, HCV-hepatitis C virus, EtOH-alcohol, MELD-model for end-stage liver disease, SBB-selective beta blockers

Figure 2 –

Covariate Balance Prior to and After Inverse Probability Treatment Weighting

Associations between NSBB Exposure and Hepatic Decompensation/Liver-related Mortality Crude Event Rates

Crude 3-year event rates for primary and secondary outcomes are shown in Table 2 and are juxtaposed with comparable event rates from NSBB clinical trials; PREDESCI trial, which compared propranolol or carvedilol to placebo among patients with well-compensated cirrhosis and HVPG ≥10 mmHg and a trial that compared timolol to placebo among patients with HVPG ≥6 mmHg.^{7, 33} In the VOCAL cohort, any decompensation occurred in 8.8% with carvedilol and in 17% of the SBB-treated patients, respectively. The composite outcome of any decompensation or liver-related mortality was 11% with carvedilol and 22% with SBBs. The more common decompensations were ascites and hepatic encephalopathy; variceal hemorrhage was rare at about 1%. All-cause mortality was similar in each group; 22% with carvedilol and 24% with SBBs whereas liver-related mortality was 2.0% with carvedilol and 5.1% with SBBs. Unadjusted cumulative incidence curves for the primary outcomes and all-cause mortality for each cohort in Figure 3, Panels A-C. Compared to SBBs, carvedilol exposure was associated with a lower crude cumulative incidence of the composite outcome of any hepatic decompensation or all cause-mortality (p=0.03), Panel C.

Table 2 –

Crude Three-Year Event Rates (%) in the VOCAL Cohort, and in previously published PREDESCI⁷, and Timolol trials²⁹

	VOCAL NSBB vs. SBB Cohort		PREDESCI Carvedilol/propranolol vs. placebo (median follow up 37 months)			Timolol vs. placebo (median follow-up 55 months)
Outcome	Carvedilol	SBB	Outcome	NSBB	Placebo	NSBB	Placebo
Any decompensation	9%	17%	---	---	---	---	---
Any decompensation or liver-related mortality	11%	22%	Any decompensation or mortality	16%	27%
Variceal hemorrhage	0.8%	1.2%	GI bleeding	4%	3%	2.8%	2.9%
Ascites	6.1%	9.8%	Ascites	9%	20%	3.7%	5.7%
Hepatic encephalopathy	3.4%	11%	Hepatic Encephalopathy	4%	5%	2.8%	1.9%
All-cause mortality	22%	24%	All-cause mortality	8%	11%	9.3%	14%
Liver-related mortality	2.0%	5.1%	---	---	---	---	---

--- indicates outcome was not separately reported in the trial

Abbreviations: NSBB-nonselective beta blocker, SBB-selective beta blocker, GI-gastrointestinal

PREDESCI trial examined propranolol or carvedilol to placebo for the primary outcome of hepatic decompensation and death.

Timolol trial compared timolol to placebo for the primary outcome of the development of gastroesophageal varices or variceal hemorrhage.

Figure 3 –

Unadjusted Cumulative Incidence Curves for (A) All-Cause Mortality, (B) Any Hepatic Decompensation and (C Any Hepatic Decompensation or Liver-Related Mortality for Comparisons between carvedilol and Selective Beta Blockers (SBBs)

IPTW adjusted analyses

Using the primary pseudo-ITT approach (first column of Table 3), carvedilol vs. SBBs was associated with a lower hazard of any hepatic decompensation (HR 0.59, 95% CI 0.42–0.83) and the composite of any hepatic decompensation/liver-related mortality (HR 0.56, 95% CI 0.41–0.76). Carvedilol was also associated with a lower hazard of the secondary outcomes of ascites (HR 0.64, 95% CI 0.40–1.00), HE (HR 0.44, 95% CI 0.28–0.71), liver-related mortality (HR 0.27, 95% CI 0.13–0.59), and all-cause mortality (HR 0.74, 95% 0.58–0.96) with no significant association observed for variceal hemorrhage (HR 0.36, 95% CI 0.06–2.04). In pseudo-as-treated analyses (second column of Table 3), carvedilol was associated with lower hazard of any hepatic decompensation (HR 0.55, 95% CI 0.33–0.94). The associations with any hepatic decompensation/liver-related mortality and individual hepatic decompensations were similar in magnitude to the pseudo-ITT analyses, but did not reach statistical significance.

Table 3 –

Inverse Probability Treatment Weighting-Adjusted Cox Regression Accounting for Time-Updated Confounders in Pseudo Intention to Treat (ITT) and Pseudo As-Treated Analyses^*

Carvedilol versus SBB †
	Pseudo-ITT		Pseudo-As-Treated
Primary Outcomes	HR (95% CI)	p-value	HR (95% CI)	p-value
Any Hepatic Decompensation	0.59 (0.42–0.83)	0.003	0.55 (0.33–0.94)	0.03
Any Hepatic Decompensation or Liver-Related Mortality	0.56 (0.41–0.76)	<0.001	0.62 (0.38–1.01)	0.057
Secondary Outcomes	HR (95% CI)	p-value	HR (95% CI)	p-value
Variceal Hemorrhage	0.36 (0.06–2.04)	0.25	-	-
Ascites	0.64 (0.40–1.00)	0.05	0.52 (0.24–1.10)	0.09
Hepatic Encephalopathy	0.44 (0.28–0.71)	0.001	0.76 (0.40–1.44)	0.40
All-Cause Mortality	0.74 (0.58–0.96)	0.02	0.89 (0.55–1.45)	0.65
Liver-Related Mortality	0.27 (0.13–0.59)	0.001	1.28 (0.25–6.48)	0.76

^*Each model adjusted for time-updating statin use, ACE inhibitor/angiotensin-receptor blocker use, coronary artery disease, heart failure, atrial fibrillation, diabetes mellitus, and HCV viremia

^†Models additionally adjusted for sex, smoking status, alcohol use disorder, body mass index, and chronic obstructive pulmonary disease

Subgroup analyses

Results from exploratory subgroup analyses stratified by baseline HCV viremia, CTP score (5 vs. 6), platelet count (30–99 × 10⁹/L vs. 100–150 × 10⁹/L), ALD, and age are shown in Figure 4 with effect estimates shown in Supplemental Tables 3 and 4. Carvedilol was associated with reduced hazard of any hepatic decompensation and hepatic decompensation/liver-related mortality among subgroups with absence of HCV viremia, CTP-6 (vs. CTP-5) cirrhosis, platelet count 100–150 ×10⁹/L (vs. 30–99 ×10⁹/L), ALD etiology, and age <65.

Figure 4 –

Adjusted Relative Hazard for Any Hepatic Decompensation and Hepatic Decompensation/Liver-Related Mortality in Intention to Treat Models with Interaction Subgroup Analyses

Abbreviations: ALD- alcohol-associated liver disease, CTP-Child-Turcotte-Pugh, HCV-hepatitis C virus, LRM-liver-related mortality, PLT-platelet count

Sensitivity analyses –competing risks

In sensitivity analyses for hepatic decompensation with all-cause mortality as a competing risk, results were similar to the primary analyses. Carvedilol was associated with a lower hazard of hepatic decompensation in pseudo-ITT: sHR 0.59, 95% CI 0.42–0.82, p=0.002 and pseudo-as-treated analyses: sHR 0.55, 95% CI 0.33–0.91, p=0.02. Cumulative incidence function curves are in Supplemental Figure 3. Additional sensitivity analyses for unmeasured confounding are in Supplemental Table 2. E-values ranged from 1.88–6.87. This suggests that unmeasured confounders would need to have moderate to strong relative effects on the exposures and the outcomes to negate our observed associations.

DISCUSSION

The efficacy of NSBBs in the primary and secondary prevention variceal hemorrhage is well-established⁶, but data in prevention of decompensation are lacking outside of a handful of clinical trials.^{7, 33} In the landmark PREDESCI study, patients with well-compensated cirrhosis, had a lower incidence of the primary composite endpoint of decompensation and death.⁷ However, in most practice settings, HVPG assessment is not standard practice and NSBB performance in prevention of adverse liver events is unknown. We investigate the role of carvedilol in preventing hepatic decompensation or mortality in a well-characterized, population-based cirrhosis cohort with detailed prescription and clinical outcome data with a larger sample size than previous NSBB trials.^{7, 33}

Main Finding

After three years of follow-up, in intention-to-treat analyses, we found that among patients with CTP-A cirrhosis with platelet counts between 30–150 × 10⁹/L and no previous history of hepatic decompensation, carvedilol was associated with a lower hazard of hepatic decompensation and the composite outcome of hepatic decompensation/liver-related mortality. To contextualize our findings, we need to compare patient characteristics and clinical outcomes between the VOCAL cohort and clinical trials evaluating the effects of NSBBs on adverse liver outcomes. Adverse events occurred with comparable frequencies to clinical trials. Similar to the PREDESCI trial where the 3-year incidence of ascites was 9% in the NSBB group, ascites incidence in our cohort ranged was 6% among carvedilol-exposed patients. The composite outcome of any hepatic decompensation/liver-related mortality was 22% in our active comparator arm compared to 27% in the PREDESCI trial. Furthermore, our effect estimate for carvedilol on hepatic decompensation and mortality was remarkably similar to the PREDESCI trial (HR 0.59 VOCAL, HR 0.51 PREDESCI) including in sensitivity analyses for just hepatic decompensation with death as competing risk (sHR 0.55) lending further face validity to our findings. Our findings also support those of a recent meta-analysis of 4 randomized, controlled trials showing that carvedilol reduced hepatic decompensation largely due to a reduced risk of ascites (sHR 0.51).³⁴ Though liver-related mortality was lower in the NSBB group in the pseudo-ITT analyses, this was not the case in the pseudo-as-treated analyses. The latter finding could be explained by discontinuation of beta-blockers prior to death and the low number of available cases to evaluate given this discontinuation.

However, there are several ways our cohort differs from the clinical trial populations. We observed a higher all-cause mortality in the VOCAL cohort, which is not surprising given the high prevalence of concomitant hypertension, obesity, and diabetes among Veterans. We also found higher incidence of HE in the SBB active comparator group in the VOCAL cohort compared to PREDESCI trial. We cannot fully explain these differences, but hypothesize that these may have been due to patient characteristics, for example, patients in the VOCAL cohort were older and had a higher prevalence of alcohol-associated liver disease. Although notably different from clinical trial populations, the VOCAL cohort may be reflective of the global increase in metabolic comorbidities observed among patients with advanced liver disease.³⁵ We acknowledge that carvedilol was largely prescribed for non-liver indications in this well-compensated cohort, however, that only supports it as valid use case to analyze its effects on the primary prevention of hepatic decompensation outside of a clinical trial setting. Liver stiffness measurements would have complemented this analysis, however, are not collected by the 128 VA facilities in a standardized fashion and are also not routinely available outside of tertiary care centers.

Interpretation of the findings

Our data must be interpreted in the context of the study design. Outside of randomized trials, outcomes for patients started on NSBB must be compared to control therapies. In clinical settings, ‘placebo-controlled’ comparisons are at risk for healthy-user bias. To overcome this limitation, we employed a methodologically robust new-user, active comparator design and adjusted for a multitude of demographic and clinical confounders including concomitant statins, angiotensin-receptor inhibitors and angiotensin receptor blockers. We compared carvedilol to selective beta-blockers and achieved comparable characteristics even prior to IPTW supporting our choice of active comparator. Our findings highlight the application of carvedilol in preventing adverse liver events and the need for prospective studies to further examine NSBB effects in population-based cohorts.³⁶

The presence of CSPH and NSSB response are key factors guiding assignment in clinical trials. Patients were selected for PREDESCI following documented HVPG ≥10 mmHg and with HVPG response to intravenous propranolol guiding the choice of NSBB (propranolol for HVPG responders and carvedilol for nonresponders). In fact, a randomized trial comparing timolol to placebo in the presence of any portal hypertension (HVPG ≥6 mmHg) found that there was no difference in the primary composite endpoint of gastroesophageal varices or variceal hemorrhage or hepatic decompensation supporting that CSPH is necessary to achieve NSBB benefit.^{7, 33} Yet, HVPG measures is rarely available in most clinical practices and even liver stiffness measurement (LSM) use is quite variable. Given these challenges, we explored several different subgroups that may derive NSBB benefits and found more beneficial associations with carvedilol in the absence of HCV viremia, younger age, more advanced CTP score (6 rather than 5), and a higher platelet threshold. Associations were similar across subgroups stratified by platelet count, however, more pronounced benefits were noted for subgroups without HCV viremia, higher CTP scores, and ALD etiology. We did not find benefits of carvedilol among patients with cirrhosis age ≥65. This may be due to multiple competing causes of mortality among older adults highlighting the need for caution for preemptive NSBBs in this age group. The mechanisms behind the differences are unclear, however, the patient populations are different in terms of extrahepatic comorbidity burden. However, given the small sample sizes of the subgroups, we view these results as hypothesis-generating to inform the design of future prospective studies.

Limitations

There are certain limitations we must acknowledge. This is an observational cohort study that is largely male and older potentially limiting generalizability. There is always the potential for unmeasured residual confounding, time-related biases or ascertainment biases in any retrospective study. However, the active-comparator new user design, IPTW, and 180-day time lag techniques help reduce these biases.²³ Furthermore, we provide a sensitivity analysis of the strength of unmeasured confounders that would be needed to alter the observed associations. We were not able to evaluate the effects of propranolol or nadolol on hepatic decompensation as the indications for these medications were nearly always in secondary prevention of portal hypertensive bleeding. Untreated or inactive comparator comparisons (i.e., to emulate placebo) are challenging to perform in observational studies as the placebo patients receiving these comparators are often much healthier or more ill than those receiving the primary exposure, introducing bias. We did not have measures of HVPG, LSM, or a reliable way to assess presence of varices on endoscopy reports and did not assess therapy adherence or hemodynamic response. Finally, a theoretical critique of ACNU ITT approaches is that exposure misclassification may bias results towards a null hypothesis and therefore potentially miss meaningful associations.¹¹ However, in our study we rejected the primary null hypothesis, and our results may in fact be conservative. Moreover, results were generally consistent between “pseudo-ITT” and “pseudo-as-treated” approaches, further supporting the validity of our findings. Despite the limitations, we evaluated outcomes in a national, large, and diverse cohort. We have adjusted for a multitude of demographic, clinical factors, time-updated confounders, and concurrent medications such as statins, angiotensin-converting enzyme inhibitors, and angiotensin-receptor blockers.

Conclusion

Our findings have important clinical implications. First, we highlight the benefits and challenges of applying clinical trial results with carefully selected criteria to clinic-based populations. Second, we show the promising nature of carvedilol in preventing adverse liver events, and the need to further study how to best select patients with compensated advanced liver disease for NSBB therapy. Third, we bring attention to the need to study novel approaches to noninvasively identify CSPH. As vibration-controlled transient elastography use becomes more widespread, future efforts should focus on whether this modality or other biomarkers can aid in identifying patients at risk for decompensation.

Abbreviations:

AUDIT-C

Alcohol Use Disorder Identification Test

BMI

body mass index

CI

confidence interval

CPT

current procedural terminology

CTP

Child-Turcotte-Pugh score

HCV

hepatitis C virus

IPTW

inverse probability of treatment weighting

M

mean

MELD

model for end stage liver disease

NAFLD

non-alcoholic fatty liver disease

NASH

non-alcoholic steatohepatitis

NSBB

non-selective beta-blocker

P25

25^th percentile

P75

75^th percentile

PS

propensity score

SBB

selective beta-blocker

SD

standard deviation

sHR

subdistribution hazard

VA

veterans affairs