2.4. Statistical analysis
Baseline and early mortality data were pooled using meta-analysis of
means or proportions, with individual study effect size accounted for
using inverse variance methods. Results were displayed as raw values as
well as percentages with 95% confidence intervals (CI) or 95% credible
intervals. Graft patency and long-term mortality for arterial, venous or
mixed CABG were analyzed across all arm-level studies, with direct and
indirect comparisons using a mixed treatment comparison based on a
Bayesian hierarchical model. HRs and corresponding 95% credible
intervals were calculated by Markov chain Monte Carlo methods using the
“BUGSnet” package of R software (version 3.6.3; R Foundation, Vienna,
Austria) [24]. Brooks-Gelman-Rubin plots method, trace plot and
density plot were used to access the model convergence [25].
Besides, rank probabilities were calculated to obtain the hierarchical
amount effects of multiple treatments. Given the graft type, CABG
modalities and time period over which included studies were conducted, a
random effects model was used. For the purposes of the mixed-treatment
comparison, consistency in direct and indirect effects was assumed
[26]. Heterogeneity between comparisons within the network was
analyzed by examining I values for the random-effects model.
Inconsistency was graphically examined using the BUGSnet nma.compare
function to plot the individual data points’ posterior mean deviance
contributions for the consistency model versus the inconsistency model
[27]. Early mortality and need of interventional procedures and
surgical re-interventions were reported as
pooled prevalence of adverse
outcome using the package “meta” of R software [28].