Heterogeneity assessment: bias exploration
A risk of bias analysis was performed for all included studies as per the Cochrane guidelines . Overall, there was a high level of bias due to the fact that a majority of studies were not randomized or blinded. Moreover, we assigned scores for each of the domains D1-D7. No study fulfilled all of these criteria (figure 3). Funnel plots were used to assess publication bias for all primary and secondary outcomes. There was no funnel plot asymmetry for the primary outcome late survival (supplementary figure 3) and tumor relapse at follow-up (supplementary figure 4).