Study design
The China Heart Failure Surgery Registry (China-HFSR) was led by Fuwai Hospital and other representative cardiac centres in different regions around China. In total, 94 centres with annual surgery volumes > 100 were included as participants in the study. We included patients ≥ 17 years old who underwent CABG from January 2012 to June 2017 with documented LVEF < 50%. Patients were excluded if they underwent concomitant valve or other surgeries or non-elective surgeries. We also excluded patients who have preoperative intra-aortic balloon pump insertion, cardiogenic shock and third degree heart block. (Fig. 1) These patients were then stratified according to preoperative beta-blocker administration. All CABG procedures represented standard surgical approaches to surgical myocardial revascularization with and without the use of cardiopulmonary bypass support. This study was approved by the institutional review board at Fuwai Hospital (approval number 887, April 25th, 2017) and carried out in accordance with relevant guidelines and regulations. The informed consent was provided by participants.
Data collection
All data were collected at the local sites from the medical records. The requirements for data collection and the definitions of variables were clearly identified. Standardized electronic case report forms were completed at the local sites and then submitted online to the data processing centre. All data were into the database separately by two trained technicians. Two separate reviewers from the data processing centre randomly selected and assessed 5–10% of each of the participating centres’ medical records during annual on-site audits. We compared the data in the database and the original medical records. A committee composed of physicians and surgeons determined the correct final value when there was a disagreement. In all patients included in China-HFSR database, 90 (1.5%) patients were without listed height, 74 (1.2%) without listed weight and 1 without data regarding smoking history. Considering the fact that they only accounted for a very small proportion of our patients, we imputed missing continuous variables (height and weight) with different mean values for the sexes. The missing categorical variable (smoking history) was imputed with negative values. So that patients who might have experienced the end point (in-hospital death) would not be excluded from analysis simply for 1 or 2 missing variables among the many examined.
Clinical data
The preoperative variables included age, gender, body mass index, smoking history, New York Heart Association (NYNH) classification, Canadian Cardiovascular Society (CCS) classification, diabetes mellitus (DM), hypertension, hyperlipidemia, renal failure, chronic obstructive pulmonary disease (COPD), cerebrovascular accident, carotid disease and other peripheral arterial disease, preoperative atrial fibrillation, previous myocardial infarction (MI), percutaneous transluminal coronary angioplasty (PTCA) history, No. of diseased vessels, Left main CAD, LVEF, preoperative creatinine and prior cardiovascular surgeries.
The major postoperative complications included re-intubation, MI, mediastinal infection, stroke, renal failure, multiple organ dysfunction syndrome, postoperative atrial fibrillation and reoperation for bleeding. MI was counted as a complication if it newly occurred postoperatively and was defined as any one of the following: MI documented in the medical record with an elevation of cardiac troponin values with at least one value above the 10 times 99th percentile upper reference limit or electrocardiograph documented ST-segment elevation in evolution, Q waves 0.03 seconds in width and/or one-third or greater of the total QRS complex in 2 or more contiguous leads, or new left bundle branch block.[11] Mediastinal infection was defined according to the expert consensus.[12] Stroke was defined as a central neurological deficit persisting > 24 hours (i.e., extremity weakness or loss of motion, loss of consciousness, loss of speech, visual field cuts). Renal failure was defined as an increase in serum creatinine level to > 4 mg/dL, 3× the most recent preoperative creatinine level, or a new postoperative need for dialysis. Reoperation for bleeding was defined as chest tube drainage ≥ 200 mL/h for at least 3 hours requiring surgical intervention.
Statistical analysis
Continuous variables are expressed as either mean ± standard deviation or medians and quartiles depending upon overall variable distribution. Categorical variables are presented as frequencies and percentages. We performed a t-test for normally distributed continuous variables; otherwise, the Mann-Whitney U test or Kruskal-Wallis H test was used. Chi square tests or Fisher’s exact tests were used for categorical variables. Cochran-Armitage trend test were used to examine the trend of beta-blocker use during the study period.
We used the following 2 techniques to adjust for selection bias when comparing outcomes of the beta-blocker vs non-beta-blocker groups: multiple logistic regression modelling and propensity matching. For the regression-based analyses, the association between preoperative beta-blocker use and each clinical end point were adjusted for baseline patient risk by inclusion of the following validated and widely accepted measures of patient-level covariates: age, body mass index, sex, smoking history, diabetes mellitus, hypertension, hyperlipidemia, chronic obstructive pulmonary disease, cerebrovascular accident, previous MI, PTCA history, LVEF, preoperative creatinine, CCS classification, NYHA classification, No. of diseased vessels, Left main CAD, preoperative atrial fibrillation and prior cardiovascular surgery history. Each logistic model also included year of surgery and a set of fixed-effect hospital-specific intercept variables.[13] Model results are reported as odds ratios (OR) with a 95% confidence interval.
The second method of adjusting for selection bias involved matching patients with similar estimated probability of receiving beta-blockers (propensity score). The propensity score was calculated by a multivariable logistic regression model which was developed using the same covariates listed above for the regression-based analyses. Then we matched patients in a 1:1 fashion without replacement.[14] ORs with 95% CIs comparing the frequency of each end point for patients receiving vs not receiving beta-blockers were estimated using univariable logistic regression.
Additional analysis were performed to examine whether the association between beta-blockers and mortality differed across prespecified subgroups based on age, sex, ejection fraction, diabetes mellitus, hypertension and chronic lung disease. Subgroup-specific ORs were estimated and displayed with 95% CIs.
All reported P values are 2 sided, and values of P < 0.05 were considered to indicate statistical significance. All statistical analysis was performed using SPSS version 22.0 (IBM Corp., Armonk, NY).