3.5.1. Assumption of NMA and Network Geometry

As explained in the Methods section, we decided to adopt a random effect model in the homogeneity assumption. In the transitivity assumption, the research team agreed on the transitivity of the included studies using Tables 1 and 2. We assessed the consistency assumption using a global and local approach. In the global approach, we found significant inconsistencies (*p* < 0.05). In the local approach, we found inconsistency due to a study that compared ELEC and CONV (Lei 2016 [36]). After we excluded the study (Lei 2016 [36]) according to the study protocol, the consistency assumption was satisfied at the local and global levels. Net-split graphs that include direct estimates, indirect estimates, and network estimates for consistency assessment are provided in Supplementary Materials Digital Content 5. The connectivity assumption was confirmed through network geometry (net graph), which is a visual presentation of the links in the included studies (Figure 2). After excluding the study (Lei 2016 [36]), in the network analysis of the primary outcome, there were five nodes (ELEC, BEEV, MANU, CONV, SHMA) from 16 studies and 20 pairwise comparisons from seven types of comparison pairs (edges). The number of included comparisons in each edge is shown in Figure 2.
