3.2.1. Primary outcome: Graft patency
Overall, 1191 patients were included in the present meta-analysis, with an average of 1.79 (95% CI: 1.55–2.00) grafts per patient. ITA, RA or GEA, and SV conduits were used, however exact numbers of each graft type and graft target vessels were incompletely reported. Patency assessment at follow-up was performed using either invasive angiography or non-invasive computed tomographic angiography, and significative coronary stenosis (>70%) or occlusions were considered non patent grafts. Of the 32 studies, 15 arm level papers were included to perform the Bayesian network meta-analysis, with a total of 1441 grafts with 324 significative stenosis/occlusions [30-33, 39, 41-44, 47, 49, 53-54, 57, 59]. Mean follow-up for coronary angiography across studies was 92.19 months (follow-up: min 3 – max 264 months). For the primary outcome we compared patency of ITAs, SV and other arteries (GEA and RA) at follow-up. Patency rates at follow-up for ITAs, SV and other arteries are 87.82±12.41%, 65.98%±27.84% and 77.63±22.75%, respectively. The network model, trace plot and density plot for graft patency are shown in Supplementary Figures 1 and 2.
Summary results for all grafts patency at follow up are shown in Figure 2A, with SV used as reference. These pooled results show that ITAs and other arteries (GEA and RA) are superior to SV. Rank probability analysis for graft patency follow-up demonstrates that ITA had higher probabilities of being the first most effective treatment (Figure 2B).
The efficacy of different treatments in terms of graft patency at follow-up using HR and corresponding 95% CrI is displayed in Figure 3. Pairwise comparisons for graft patency at follow-up are shown in Supplementary Figures 3A, 3B and 3C. I2 demonstrated that heterogeneity was low for SV versus ITA comparison and was moderate for SV and ITA versus other arteries (GEA and RA). After comparison of consistency and inconsistency models through individual data points’ posterior mean deviance contributions we conclude that there is a lack of evidence to suggest inconsistency within the network (Supplementary Figure 4).