Statistical analysis
All statistical analyses were performed in R version 4.0.3. For all
outcomes bar post-operative introduction of nasogastric tube, mean
difference (MD) was used as a treatment estimate and standard error of
the mean was deployed as the standard error of treatment estimate. Since
outcomes were reported in widely different ways, we reformatted them to
MD when needed using the following rules. If the means in each treatment
group were available, MD was calculated as the difference between the
two means. If median and range were present, we used the formula
proposed by Hozo et al. to calculate the means in each
group10. If range and IQR were available, we used the
formula Mean = (Quartile 1 + Median + Quartile 3) to calculate the means
in each group. For post-operative introduction of nasogastric tube, we
used the odd ratio (OR) as the unit of treatment effect. When none of
the above was available, studies were excluded from analysis in this
outcome. Since we aimed to compare pharmacological class treatment size
rather than individual treatment or dose, we could not use the multi-arm
studies option of network meta-analysis for studies comparing more than
one dose of the same treatment to placebo. Therefore, we recalculated a
common mean for all treatment arms to calculate the mean difference. The
main analysis was a Bayesian network meta-analysis using the packages
gemtc, rjags and the software JAGS. We used an identity-link for mean
difference and a log-link for the OR. We performed 10e5 iterations with
a thinning interval of 10 and a burn-in interval of 5000. Model
convergence was assessed using density and trace plots. Results are
presented as a network graph, a forest plot comparing each treatment
with placebo, ranking probability as well as a surface under the curve
cumulative ranking probabilities (SUCRA). As a sensitivity analysis, we
performed a frequentist network meta-analysis using the package netmeta.
We presented results as comparison to placebo and a matrix presenting
both direct and indirect comparisons. Finally, publication bias was
assessed for each outcome with a funnel plot and an Egger’s test.