4.4. Quantitative Synthesis and Statistical Analysis

We anticipated that a variety of pedicle clamping techniques would be reported throughout the literature. Therefore, we classified the distinct comparisons made in the individual studies by constructing a network plot, as described by Chaimani et al. [64]. In such a plot, the size of each dot represents the number of study arms, and the thickness of each connecting line represents the number of comparisons made between the given groups (e.g., whole pedicle clamping vs. no clamping, intermittent clamping vs. continuous clamping, prolonged vs. short continuous clamping). Whenever data were sufficient, outcomes were pooled and compared by random effects meta-analysis [65]. The primary comparison of interest to this meta-analysis looked at (a) patients undergoing portal triad clamping (either intermittent or continuous) versus (b) those undergoing no clamping at all. In this main analysis, studies reporting more than one comparison of clamping type or duration were considered as separate studies [66]. The aggregated effect size was expressed as a standardized mean difference in survival.

Between-study heterogeneity was calculated using the I² statistic, and was explored using several approaches. First, sensitivity analyses were performed to address the putative sources of heterogeneity. As another approach, we conducted meta-regression analyses to test for an association between bibliometric and clinical characteristics of the individual studies and their respective effect estimates. Factors assessed by meta-regression included study quality, the impact factor of the journal where studies were published, the extent of the liver resection, the number of tumor nodules, the size of the largest tumor, the median age of the cohort.

In case of missing summary statistics (such as mean values and standard deviations), we estimated them from medians and percentiles as proposed elsewhere [66,67]. Moreover, when necessary, relevant data were obtained by digitalizing results from the original figures, via high magnification and point estimation with the software GetData Graph Digitizer [68]. Statistical analyses were done using Stata software (v.15, College Station, TX, USA).