Nonselective Beta-blockers Are Associated With a Lower Risk of Hepatocellular Carcinoma Among Cirrhotic Patients in the United States

Karn Wijarnpreecha; Fang Li; Yang Xiang; Xun Xu; Cong Zhu; Vahed Maroufy; Qing Wang; Wei Tao; Yifang Dang; Huy Anh Pham; Yujia Zhou; Jianfu Li; Xinyuan Zhang; Hua Xu; C. Burcin Taner; Liu Yang; Cui Tao

Disclosures

Aliment Pharmacol Ther. 2021;54(4):481-492. 

In This Article

Methods

Data Source

This is a retrospective cohort study based on the de-identified Cerner EHRs data. Data comprise demographics, encounters, diagnoses, procedures, lab results, medications, and other clinical observations. Cerner Corporation has established Health Insurance Portability and Accountability Act-compliant (HIPAA) operating policies to establish de-identification for Health Facts[15] (Supplementary material). UTHealth has agreements with Cerner to use data for research purposes, and the institutional review board (IRB) at UTHealth approved the study protocol.

The data were processed using a Python-based EHR data processing pipeline on an Intel Xeon CPU server.

Inclusion and Exclusion Criteria

Patients, who were diagnosed with cirrhosis (using International Classification of Disease [ICD] 9 and 10-CM) and had consecutive records in Cerner between January 2, 2000, and December 31, 2017, were identified as potential inclusion in this study (n = 217 804). The index date was defined as the first date of diagnosis of cirrhosis. Patients were excluded if they met any of the following criteria: (a) missing values, including without medication information (n = 47 681), or without age information (n = 1027) after age imputation, or without sex information (n = 8); (b) younger than 18 years at the index visit (n = 4064); (c) the follow-up time less than 30 days, that is, for HCC patients, development of HCC within 30 days from the index date (n = 7661); for non-HCC patients, the last encounter from the index date less than 30 days (n = 50 720); (d) taking multiple types of NSBB (n = 2126), for example, having more than one type of NSBB in records; (e) the total continued use of NSBBs less than 30 days, including carvedilol (n = 6148), nadolol (n = 3593) and propranolol (n = 7185). These five exclusion conditions were not mutually exclusive and could overlap; 107 428 patients were finally included as the cohort. In the stratified analyses, patients were divided into four groups according to their monotherapy types: carvedilol (n = 5581), nadolol (n = 2353), propranolol (n = 4273), and no beta-blocker (95 228). Figure 1 (part A) demonstrates the flowchart of patient selection in this study. This cohort study was conducted in accordance with the STROBE (The Strengthening the Reporting of Observational Studies) statement.[16]

Figure 1.

Flow chart of cohort selection and propensity score matching

Study Outcomes and Variable Definitions

The primary outcome of the study was to determine the incidence of HCC among cirrhotic patients who were treated with carvedilol and another two NSBBs (nadolol and propranolol) compared with those not treated with any beta-blockers. The diagnosis of HCC was made according to the ICD codes. Patients were followed up from the index visit to the date of the diagnosis of HCC (for HCC patients) or to the last clinic visits (for non-HCC patients).

The primary exposure of interest was the prescription of carvedilol. Since previous studies reported that traditional NSBBs (nadolol and propranolol) were associated with a lower risk of HCC,[10,17,18] we also included them in this study to validate their protective effect using the EHR data. NSBB use (including carvedilol, nadolol, or propranolol), was defined as receiving continuous therapy for more than 30 days and with monotherapy (ie, only one type of these three NSBBs documented in a patient's record).

The patient variables (characteristics) retrieved from Cerner included (a) basic demographics, including the age at the index visit, sex, and race/ethnicity; (b) complications of cirrhosis, including ascites, hepatic encephalopathy, esophageal varices, hepatorenal syndrome, portal hypertension and spontaneous bacterial peritonitis (SBP); (c) possible risk factors and comorbidities, including cerebrovascular disease, diabetes, heart disease, hepatitis B, hepatitis C, nonalcoholic fatty liver disease (NAFLD); (d) comedications, including aspirin and statins use; and (e) the follow-up time from the index visit to the last visit and the total follow-up time from the first visit to the last visit. The disease variables were defined according to the ICD codes and the medications were defined by their generic names in the prescription records (see Table S1 and Table S2 for the specific ICD codes and generic names).

Statistical Analyses

For descriptive statistics, the baseline characteristics were reported by percentages for categorical variables and means ± standard deviation (SD) for continuous variables. Comparisons between groups (P values) were estimated by the chi-square test for categorical variables and the Kruskal Wallis test for continuous variables. The cumulative HCC incidence rate was estimated using a Kaplan-Meier method that can be used with time-varying covariates. The log-rank test was applied to compare differences in HCC incidence between groups, and the Z-test was applied to compare differences in the cumulative HCC incidence at a specific time point (100-month in this study) between groups. Cox proportional hazards regression (Hazard ratio [HR, 95% CI]) was used to estimate the association between NSBBs use and HCC development. We performed both univariate and multivariate Cox analyses. For the multivariate Cox analysis, demographics (age, sex, and race), possible risk and protective factors (diabetes, viral hepatitis B, viral hepatitis C, NAFLD, cerebrovascular disease, heart disease, aspirin use, and statins use) were considered as potential confounders for the association between NSBB use and HCC development. To assess the HCC protective effect of carvedilol in the subpopulation of cirrhotic patients, we also performed both univariate and multivariate Cox analyses in subgroups stratified by etiologies of cirrhosis and cirrhosis with complications. Subgroup analyses include alcoholic cirrhosis and nonalcoholic cirrhosis, cirrhosis with complications (decompensated liver cirrhosis), and with varices or portal hypertension.

In addition, we conducted an assessment of effect modification for the risk of HCC among NSBBs groups stratified by age, sex, race/ethnicity, and risk factors (diabetes, viral hepatitis B, viral hepatitis C, and NAFLD). Odds ratios (ORs) and P values of the risk of HCC were calculated using Fisher's exact test.

To minimise the probability of selection bias, statistical analyses were performed after propensity score matching (PSM).[19] We used the nearest neighbor matching for reference (no beta-blocker)—exposure (NSBBs) with a ratio of 2:1. Since we had three exposure groups (carvedilol, nadolol, and propranolol) and one reference group (no beta-blocker), to get fully matched datasets that had the same severity of cirrhosis and feature distribution, we matched the reference group to each exposure group respectively, using all possible factors that were regarded as relevant to evaluate the severity. The matching factors included age, sex, complications (ascites, hepatic encephalopathy, hepatorenal syndrome, portal hypertension, SBP, and varices), risk factors (diabetes, NAFLD, viral hepatitis B, and viral hepatitis C), as well as comorbidities and comedications (essential hypertension, cerebrovascular diseases, heart disease, vitamin D deficiency, aspirin use, and statin use). After PSM, the final matched cohort comprises carvedilol: no beta-blocker (5581:11 162), nadolol: no beta-blocker (2353:4706), and propranolol: no beta-blocker (4273:8546) (Figure 1, part B). Our statistical analyses, including Kaplan-Meier estimate, Cox proportional hazards regression, as well as the assessment of effect modification, were all based on the datasets that applied this PSM method.

Statistical analyses were performed in RStudio with R version 4.0.3. Two-tailed P ≤ 0.05 was considered significant.

processing....