• Current treatment guidelines recommend a “start low, go slow” strategy when initiating β-blocker (BB) treatment for patients with heart failure with reduced ejection fraction (HFrEF).

• A fourth of patients were initiated on a starting dose of BB that was not adherent to the guideline recommendation among Chinese hospitalized patients with HFrEF during hospitalization.

This study retrospectively recruited Chinese patients with HFrEF using electronic medical records (EMRs) from the West China Hospital of Sichuan University (Zhou et al., 2021a) who 1) were discharged between January 1, 2011 and September 30, 2018; 2) were ≥18 years old; 3) had an LVEF <40% (echocardiography reading); 4) had a length of stay >2 days; 5) initiated BBs during hospitalization (did not receive BBs within the two calendar days after admission but a new BB on the third calendar day or later); and 6) had essential records of laboratory tests [serum creatinine, N-terminal pro-B type natriuretic peptide (NT-proBNP), low-density lipoprotein (LDL-C), hemoglobin and cardiac troponin T (cTNT)] and vital signs [heart rate, and systolic and diastolic blood pressures (BPs)]. We excluded patients underwent any non-interventional surgery.

This study was approved by the ethical committee of West China Hospital, Sichuan University (No. 2019-472). The patient consent was waived for data collection of this study was based on the EMR system retrospectively.

We collected the information of age, sex, smoking, alcohol consumption, surgery, prescription, admission department, discharging diagnoses with International Classification of Diseases, 10th Revision (ICD-10 codes), and patient status on discharge in the EMR system for each patient. We also extracted vital signs (systolic and diastolic BPs, heart rate, and respiration rate), height, and weight from the nursing system, laboratory test results from the laboratory information system (LIS), and LVEF from the echocardiography reading. For patients admitted more than once, we collected the data derived from their first hospitalization. Body mass index (BMI) was calculated as weight (kg) divided by height (m) squared. We used records of prescriptions within the two calendar days after admission, the first record of each vital sign, the first laboratory test results during hospitalization as baseline characteristics for each patient. The estimated glomerular rate filtration (eGFR) was calculated according to the chronic kidney disease epidemiology collaboration formula (Levey et al., 2009; De Cosmo et al., 2016) using age, sex, and serum creatinine values. Hypertension, diabetes mellitus, coronary heart disease (CHD), arrhythmia, cardiomyopathy, rheumatic heart disease, and acute myocardial infarction were identified using the discharging ICD-10 codes I10 to I15, E10 to E14, I20 to I25, I44 to I49, I42, I01 to I09, and I21 to I23, respectively. We calculated the Charlson comorbidity index (CCI) using the ICD-10 codes of discharging diagnoses (Charlson et al., 1987; Fraccaro et al., 2016).

A standardized initiating dose for each patient was calculated as his/her original initiating dose recorded in the EMR divided by the BB target dose. Recent CPGs recommend that BBs should be initiated at a dose of no more than 1/8 of the target dose (Chinese Society of Cardiology of Chinese Medical Association, 2014; Ponikowski et al., 2016). Patients were identified to be non-adherent to the CPG if they had a standardized initiating dose >1/8, and otherwise in the adherence group.

Time to event outcomes included profound bradycardia (defined as the first episode of heart rate <50 beats per minute during BB administration), hypotension (defined as the first episode of systolic BP <90 mmHg during BB administration), cardiogenic shock requiring intravenous inotropes (defined as the first episode of receiving intravenous milrinone, dobutamine or noradrenaline during BB administration), severe bronchospasm requiring intravenous steroid (defined as the first episode of receiving intravenous methylprednisolone during BB administration), and dose reduction or withdrawal [defined as the first episode of either receiving a standard dose of BB that was less than that of the previous day (reduction) or stopping BB during hospitalization (withdrawal) for any reason]. All patients were followed up throughout the hospitalization. The survival time of a given time-to-event outcome started on the date of the initiating BB dose and ended at the occurrence of this outcome. Patients were considered as censored if they did not have a given outcome before the earlier date of the last BB dose or 7 days after initiation of BBs.

Descriptive statistics were used for baseline patient characteristics. The normality of variables was tested using the Kolmogorov-Smirnov test. Normal continuous variables were described as mean ± standard deviation (SD) and compared between groups using the two-sided student’s t-test. Non-normal continuous variables were shown as median (25% quantile–75% quantile) and compared using the Mann–Whitney U test. Categorical variables were reported as frequencies (percentages) and compared using the Chi-square test.

We adopted inverse probability weighting (IPW) to minimize the bias due to potential confounders between two groups. (Cole and Hernán, 2008) We derived propensity score (PS, the estimated probability of initiating BBs with a dose >1/8 of the target dose) for each patient using a multivariable logistic regression model, adjusting for age, sex, baseline heart rate, baseline systolic BP, baseline NT-proBNP, baseline LVEF, baseline eGFR, and CCI. We used estimated PSs to calculate stabilized inverse probability weights which were used in the following analyses to weight the individual contribution to models or cumulative probability (Cole and Hernán, 2008). We presented the effective sample size and the between-group difference of PS before and after applying IPW. (Supplementary Figure S1; Supplementary Table S1 ) Absolute standardized differences before and after applying IPW were calculated. (Supplementary Figure S2 ) The baseline characteristics were considered comparable if the absolute standardized difference was less than 0.10. (Supplementary Figure S2).

The IPW-adjusted cumulative incidence was constructed for all outcomes. A Cox proportional hazards regression model was used to assess the association of each time-to-event outcome with guideline non-adherence, weighted by the inverse probability weights. The proportional hazards assumption was evaluated by the proportional risk assumption test. The hazard ratios (HRs) and 95% confidence intervals (CIs) of non-adherence were reported. For data that did not satisfy the assumption of proportional hazards (global P < 0.05), we utilized an accelerated failure time model with Weibull distribution to explore the HR and 95% CI of the non-adherence group, adjusting for age, sex, baseline heart rate, baseline systolic BP, baseline NT-proBNP, baseline LVEF, baseline eGFR, and CCI (Lee and Go, 1997).

Patients were stratified by sex (female/male), age (>/≤60 years of age), CHD (with/without), and hypertension (with/without) for subgroup analyses.

We also conducted two sensitivity analyses. The first was to check whether the results were robust to the addition of additional covariates, whether use of CCB, ARB, and venous furosemide within the two calendar days after admission. The performance of IPW and absolute standardized differences between two groups before and after applying IPW were presented in Supplementary Figures S3, S4. The second sensitivity analyses adjusted existed covariates, department of admission (cardiology vs others) and whether the use of oral thiazide at baseline in the generation of the inverse probability weights and multivariable accelerated failure time models. Subgroup analyses included sex, age, status of hypertension, status of CHD, and individual BB (metoprolol succinate vs bisoprolol). The performance of IPW and absolute standardized differences between two groups before and after applying IPW were presented in Supplementary Figures S5, S6.

All analyses were conducted using R Studio (R Pack Version 4.1.0, R Core Team, 2021, R Foundation for Statistical Computing, Vienna, Austria) (R Core Team, 2021), and figures were produced using the package ggplot2 (Wickham, 2016), survival (Therneau T, 2020), and forestplot (Max Gordon and Thomas Lumley, 2020). A two-sided P value < 0.05 was considered statistically significant.

Among 8,864 patients with LVEF <40%, this study included 1,104 HFrEF inpatients who initiated BBs during hospitalization (Figure 1). The baseline characteristics of the patients are shown in Table 1. Among the included patients, 356 (32.2%) patients were female; 633 (57.3%) patients were aged >60 years; 525 (47.6%) patients had CHD; and 446 (40.4%) patients had hypertension at baseline. The median length of hospitalization was 12 days [interquartile range (IQR), 8–17 days].

The clinicians prescribed BB with a higher dose than the guideline recommendation in 304 (27.5%) patients (median: 1/4 of the target dose). Most patients were initiated BBs within the first few calendar days of admission (Supplementary Figure S7), with a higher portion in non-adherent group (Supplementary Figure S8). Patients in the non-adherence group received more prescriptions of bisoprolol (48.4 vs 11.5%), but less metoprolol succinate (49.7 vs 79.8%) than the adherence group.

Patients in the non-adherence group were with younger age (58.6 ± 15.6 years vs 62.2 ± 14.9 years, P < 0.001), less baseline CHD (38.5 vs 51.0%, P < 0.001), less baseline acute myocardial infarction (9.9 vs 20.4%, P < 0.001); lower NT-proBNP level (median, 3,506.5 pg/ml vs 4,198.0 pg/ml; P = 0.01); more prescriptions of ARB or CCB (37.2 vs 29.6%, P = 0.02; 25.7 vs 17.6%, P = 0.003), and less prescriptions of oral furosemide, venous furosemide, or aldosterone receptor antagonist (23.7 VS 32.9%, P = 0.003; 64.1 vs 71.1%, P = 0.03; 78.9 vs 86.5%, P = 0.002) than the adherence group.

After applying IPW, baseline characteristics were well balanced between groups. (Supplementary Figure S2) As shown in Figure 2, guideline non-adherence was not significantly associated with a higher risk of profound bradycardia (HR, 1.49; 95% CI, 0.87 to 2.56; P = 0.15), hypotension (HR, 1.19; 95% CI, 0.89 to 1.59; P = 0.24), cardiogenic shock requiring intravenous inotropes (HR, 1.11; 95% CI, 0.74 to 1.65; P = 0.63), and severe bronchospasm requiring intravenous steroid (HR, 1.59; 95% CI, 0.90 to 2.80; P = 0.11), but was significantly related to a higher risk of BB dose reduction or withdrawal (HR, 1.78; 95% CI, 1.42 to 2.22; P < 0.001).

However, we did not find any subgroup effects by age, sex, with/without CHD, or with/without hypertension on the association of non-adherence to adverse events (all P interaction>0.05) (Figure 3).

The first sensitivity analyses showed the results were robust to the addition of additional covariates, whether use of CCB, ARB, and venous furosemide within the two calendar days after admission. (Supplementary Figures S9, S10). The second sensitivity analyses confirmed the robustness of the results and did not find any subgroup effect by individual BB. (Supplementary Figure S11).