### INTRODUCTION

### STATISTICAL APPROACH OF NETWORK META-ANALYSIS

### Bayesian method

#### Prior and posterior distribution in Bayesian inference

#### Markov chain Monte Carlo simulation

^{2}=1/4*3.142*1

^{2}, which is 0.7855.

#### Bayesian hierarchical model

_{i}is the actual observation effect of the

*i*th study, and v

_{i}is the variance of the

*i*th study. Here, θ is the true value of the treatment effect and a common effect size to be inferred by the fixed effect model.

^{2}) are input to the prior distribution. In turn, μ follows a hyperprior distribution, which is a normal distribution with μ

_{0}as mean and η

_{0}

^{2}as variance, and τ

^{2}follows a hyperprior distribution with p as mean and q as variance. These parameters μ

_{0}, η

_{0}

^{2}, p, and q of the prior distributions μ and τ

^{2}are hyperparameters.

_{i}to θ

_{k}. Thus, it is no longer an independent model but a hierarchical model [5].

#### Summary of the Bayesian method

### Frequentist method

*i*th study in the

*d*th study design.

*i*th study of the

*d*th study design, considering heterogeneity between studies and inconsistency between study designs. It is called mean difference, log risk-ratio, or log odds ratio (OR). Next, δ

^{AJ}represents the treatment contrast as a primary index of interest, and the effect size of J treatment in contrast to the A reference treatment. Then,

^{2}with a random effect model in pairwise meta-analysis. Then,

### BAYESIAN NMA USING R "*gemtc*" PACKAGE

*gemtc*" for NMA using the Bayesian method. When coding the data first, you must set the variable names in accordance with the relevant function. The process is as follows: network setup -> select a network model (fixed or random) -> select the MCMC convergence optimal model -> statistical reasoning in the final model (Figure 3).

*gemtc*” for Bayesian NMA and “netmeta” for frequentist NMA. Before starting the analysis, you must install the packages with the following commands. For a more detailed explanation, you can refer to the detailed code, data, and references for each package [7].

*gemtc*”)

### Data coding and loading

*gemtc*” package to perform Bayesian NMA.

*gemtc*” package has many sub-functions. Among them, the “mtc.network” function can be run only if the data function name is a specific name. In the binary data, it must be “study,” “responders,” “sampleSize,” or “treatment.” In this example, the variable name is different; thus, you must change the variable name with the “colnames” command as follows:

### Network setup

### Network model

### Markov chain Monte Carlo (MCMC) simulation and convergence diagnosis

#### Running MCMC simulation

#### MCMC simulation and convergence status

*pD*is an estimated value of the parameter. Thus, the DIC considers both the fitness and complexity of the model, and the smaller the DIC is, the better the model.

**Selecting the final model for MCMC simulation**

#### Consistency test

#### Forest plot

#### Treatment ranking

### FREQUENTIST NMA USING R “netmeta” PACKAGE

#### Data coding and loading Load the “netmeta” package to perform frequentist NMA.

#### Network model

**Network plot**

**Network model summary estimates**

#### Consistency test

**Global approach**

**Local approach**

#### Forest plot

#### Treatment ranking

### COMPARISON OF NMA RESULTS: BAYESIAN VS. FREQUENTIST METHOD AND R VS. STATA SOFTWARE

### CONCLUSION

*gemtc*” package and the frequentist network meta-analysis using the “netmeta” package. We found that these two methods produced the same results. Refer to the references for detailed descriptions for continuous data, besides the binary data presented in the examples in this study [2].