Background and Goal of Study: 
The usual frequentist models for meta-analysis specify a normal distribution for the true effects, but, sometimes, the effect distribution is not normal due to be skewed or non-parametric, causing an increase in false positives. From this perspective, we decided to reevaluate the results of a recent meta-analysis on the use of sugammadex versus neostigmine as reversor of neuromuscular blockade.

Materials and methods: 
The data for the construction of the models was extracted from the systematic review PROSPERO 2015:CRD42015016755, a meta-analysis of 17 trials of 1553 participants that studies the differences in muscle weakness, adverse effects, nausea and vomiting following administration of sugammadex or neostigmine. For each outcome, the result of the studies was simulated 4000 times based on their original result and its standard error, applying a conditional Dirichlet model and modifying its accuracy (M) exponentially from 0.25 (very non-parametric) to 64 (very parametric), building 9 models. For each model, we calculate the value of log RR and the probability that the actual RR was less than 0. The plausibility of each model was assessed by calculating the Bayes factor.

Results: 
Results of parametric models were similar to original meta-analysis, but were worse in non-parametric models, not reaching 95% in adverse events. However, it confirmed the strength of the result about minor signs of muscular weakness.

Conclusion: 
Bayesian simulation model can be use as a sensitivity analysis that goes beyond the funnel plot and the exclusion of trials based on their bias risk.

Sugammadex vs. neostigmine for reversal of neuromuscular blockade: a Bayesian non-parametric model of simulation of meta-analysis based on the results of a systematic review