Data are presented as mean ± one SD for continuous variables or as percentages for dichotomous variables. Continuous numerical variables were tested for normality using the Shapiro–Wilk test and then compared between groups with unpaired Student’s t-test if normally distributed or Mann–Whitney U-test if not normally distributed. In the case of dichotomous variables, Pearson’s χ2 or Fisher’s exact test were used as appropriate. Predictors for composite outcome were tested using a multivariable logistic regression model: a stepwise selective approach was conducted by backward and forward selection methods using Akaike information criterion as discrimination criterion between models. Event-free survival curves were compared between the groups by Kaplan–Meier method and subsequently compared with the log-rank test. To further adjust for patient selection and preoperative characteristics, we developed 2 propensity score-matched analyses including in the analysis all the baseline variables available. In Matched Analysis A, we included all the identified redo coronary surgery patients, regardless of the type of primary operation. In Matched Analysis B, we included only patients who had undergone isolated coronary surgery as their primary operation. In both cases, patients undergoing redo-OPCAB were matched (1:1) to the group undergoing redo-CABG by all the preoperative variables. Intraoperative variables were not included in the model as they occurred during the surgery. The nearest neighbour method was used with a caliper of 0.2 and the balance after matching was evaluated with standardized mean differences. After propensity score matching, variables were compared using paired Student’s t-test or paired Wilcoxon test for continuous variables and McNemar (for dichotomous variables) and χ2 test for ordinal categorical variables. A conditional logistic regression model was developed to evaluate the predictors of the composite outcome including the same variables used for the non-matched unconditional logistic regression plus the matching index. All tests were two-sided with the α level set at 0.05 for statistical significance. Clinical data were recorded and subsequently tabulated with Microsoft Excel (® Microsoft Corp, Redmond, WA, USA). The statistical analysis was computed using R version 3.0.2 [R Core Team (2014), R Foundation for Statistical Computing, Vienna, Austria]. The propensity score matching was computed with the package MatchIt [15].