Statistical analysis
Statistical analyses were performed using the Statistica (TIBCO v.13.3) software. Continuous variables are expressed as mean ± standard deviation (SD) or median range as appropriate.  Confidence intervals (95% CI) for individual groups and the whole group were measured. For a subset of studies with analyzable and comparable data, the results were synthesized quantitatively by performing random-effects model meta-analyses to compute absolute net changes in continuous variables (for example, time to PNI or temperature to PNI) and pooled OR for binary variables (for example, PNI with CMAP and no phrenic nerve injury). All pooled estimates were displayed with a 95% CI. For PNI resolution the Wilcoxon Peto’s log-rank test was used because higher sensitivity at early survival times than late ones. The existence of heterogeneity among effect sizes of individual studies was assessed using an I2 of higher than 50% with p value of < 0.1, test, and the T 2 index with a value of ≥ 95%, indicating medium-to-high heterogeneity. A funnel plot and Begg and Mazumdar’s test were used to investigate publication bias for the primary clinical outcome.