Statistical analysis
Paired and unpaired t tests were used for comparison of normally
distributed and Wilcoxon rank sum test used for non-normally distributed
variables. Where necessary, log transformations were employed. Data are
presented as mean (standard deviation) or median (Q1–Q3). Dichotomous
variables were compared using χ2 test or Fisher’s exact test, as
appropriate. The Kaplan-Meier survival methods with log-rank tests were
used. The patients were divided into 2 groups; (1) preoperative SR
group, and (2) preoperative AF group. To investigate the relationship
between preoperative AF and long-term mortality, univariate and
multivariable hazard regression models of Cox were used. The bootstrap
technique using one thousand samples was used as a way to account for
final multivariable model uncertainty. All study variables were first
analysed with univariate analysis and those that showed a significant
interaction (P< 0.1) were entered into the final multivariable
analysis. Furthermore, we performed propensity score matching analysis
with 1:1 matching followed by logistic regression analysis to estimate
the average treatment effect adjusted for baseline differences (namely
age, gender, hypertension, diabetes mellitus, chronic pulmonary disease,
peripheral vascular disease, previous myocardial infarction,
perioperative creatinine level and left ventricular function) between
the two groups of interest. The coefficients were converted to odds
ratios for interpretation. A P‐value of <0.05 was considered
statistically significant. Analyses were performed with Stata V.15
(StataCorp, College Station, Texas).