2.6 Statistical analysis
Treatment nodes included different
routes of administration. We drew network plots with the multinmapackage in R (version 4.1.3).18 We conducted the
network meta-analysis using a random-effects model and consistency
model. This analysis was performed with the Bayesian
framework.19 We chose mean differences (MD) and 95%
credible intervals (CrI) for intraocular pressure. We used the Markov
chain Monte Carlo method, which built up four chains, and set 80,000
iterations after an initial burn-in of 20,000 and a thinning of one. We
assessed local incoherence and obtained indirect estimates using node
splitting models.20 We calculated the surface under
the cumulative ranking curve (SUCRA) to rank different administration
routes.21 SUCRA is a percentage interpreted as the
probability of a treatment that is the safest without uncertainty on the
outcome, which is equal to 1 or 0
when the treatment is certain to be the best or the worst respectively.
We performed multiple sensitivity analyses: 1) exclusion of studies
without diabetes mellitus; 2) exclusion of studies with fewer than 20
participants; and 3) exclusion of studies with combined laser therapy.
We conducted the above statistical analyses using the gemtcpackage in R (Version 4.1.3).