Introduction to OssaNMA

ssnma()

Assuming a new two-arm trial comparing treatment 1 and treatment 2 is to be planned. The two treatments exist in the existing network, which serves as a foundation to analyze the new trial with the existing network using network meta-analysis (NMA).

Given that the risk of treatment 1 is 0.2, the risk of treatment 2 is 0.3, and the standard error of the estimated effect size between two treatments from the existing NMA is 0.3, ssnma() can be applied to solve the minimum required total sample size for the new trial to achieve a power of 0.8 and allocate it to each treatment group under different allocation method (even or uneven) and analysis method (with or without the existing network):

# Analyze the new trial with the existing network
ssnma(p1 = 0.2, p2 = 0.3, enma_sigma = 0.3, power.level = 0.8, 
        sig.level = 0.05, method = "with", allocation = "uneven")
#> $sample_size
#> [1] 187 163
#> 
#> $power
#> [1] 0.801

# Analyze the new trial with the existing network and 
# keep the sample size of each group to be the same
ssnma(p1 = 0.2, p2 = 0.3, enma_sigma = 0.3, power.level = 0.8, 
        sig.level = 0.05, method = "with", allocation = "even")
#> $sample_size
#> [1] 176 176
#> 
#> $power
#> [1] 0.801

# Analyze the new trial without the existing network
ssnma(p1 = 0.2, p2 = 0.3, enma_sigma = 0.3, power.level = 0.8, 
        sig.level = 0.05, method = "without", allocation = "uneven")
#> $sample_size
#> [1] 317 277
#> 
#> $power
#> [1] 0.801

# Analyze the new trial without the existing network and 
# keep the sample size of each group to be the same
ssnma(p1 = 0.2, p2 = 0.3, enma_sigma = 0.3, power.level = 0.8, 
        sig.level = 0.05, method = "without", allocation = "even")
#> $sample_size
#> [1] 298 298
#> 
#> $power
#> [1] 0.801

ssanma()

Assume that we have the same new trial planned as the previous section, the goal in this section is to calculate the optimal sample size allocation to each treatment group with a fixed total sample size of 200 to maximize the power, ssanma() is used:

# Analyze the new trial with the existing network
ssanma(p1 = 0.2, p2 = 0.3, enma_sigma = 0.3, N = 200, sig.level = 0.05,
        method = "with")
#> $sample_alloc
#> [1] 107  93
#> 
#> $power
#> [1] 0.679

As we can see, the optimal way is to allocate 107 subjects to group 1 and 93 subjects to group 2. The corresponding power is 0.679.

We may wonder, if we were to analyze it traditionally without the existing network, what would be the optimal sample allocation and power? By changing the method to without, we have:

# Analyze the new trial without the existing network
ssanma(p1 = 0.2, p2 = 0.3, enma_sigma = 0.3, N = 200, sig.level = 0.05,
        method = "without")
#> $sample_alloc
#> [1] 107  93
#> 
#> $power
#> [1] 0.37

The optimal sample allocation when we analyze the new trial traditionally is the same to previously. However, the power decreased greatly compared to analyzing it with the existing network.

We may also wonder, what’s the power if we allocate it evenly to each group? The parameter allocation can be used:

# Even allocation and analyze the new trial with the existing network
ssanma(p1 = 0.2, p2 = 0.3, enma_sigma = 0.3, N = 200, 
        sig.level = 0.05, method = "with", allocation = "even")
#> $sample_alloc
#> [1] 100 100
#> 
#> $power
#> [1] 0.678

# Even allocation and analyze the new trial without the existing network
ssanma(p1 = 0.2, p2 = 0.3, enma_sigma = 0.3, N = 200, 
        sig.level = 0.05, method = "without", allocation = "even")
#> $sample_alloc
#> [1] 100 100
#> 
#> $power
#> [1] 0.369

Application: Designing a new trial based on an existing NMA

In this section, we will show how to use the netmeta package to calculate the standard error of the estimated effect size between two treatments in the existing network, which would be used as the input parameter of the functions in OssaNMA package and to assist designing a new trial based on an existing NMA. An example dataset is loaded for illustration:

# load the example dataset in package OssaNMA
data(BRDdat)
head(BRDdat)
#>   studlab            treat1       treat2       TE      seTE
#> 1       1 No active control  Florfenicol 1.817766 0.4198201
#> 2       2 No active control Enrofloxacin 3.471966 1.5027924
#> 3       3 No active control Enrofloxacin 3.201584 0.4781349
#> 4       4 No active control Enrofloxacin 1.434531 0.3878773
#> 5       5 No active control Enrofloxacin 3.197472 0.4938823
#> 6       6 No active control Enrofloxacin 1.945910 0.3854496

This example dataset represents a previously published network of interventions for the treatment of Bovine Respiratory Disease (BRD) in feedlot cattle (O’Connor, Yuan, Cullen, Coetzee, Da Silva, and Wang, 2016). The dataset is comprised of 98 trials, 13 treatments and 204 arms. Each row represents the summary statistics for a pairwise comparison between two treatment in a trial. See the meaning of each column below:

• studlab: study id
• treat1: name of treatment 1
• treat2: name of treatment 2
• TE: estimated treatment effect size (log odds ratio) between treat1 and treat2
• seTE: standard error of TE

Let’s conduct a network meta-analysis(NMA) using this dataset:

library(netmeta)
#> Loading required package: meta
#> Loading 'meta' package (version 6.0-0).
#> Type 'help(meta)' for a brief overview.
#> Readers of 'Meta-Analysis with R (Use R!)' should install
#> older version of 'meta' package: https://tinyurl.com/dt4y5drs
#> Loading 'netmeta' package (version 2.5-0).
#> Type 'help("netmeta-package")' for a brief overview.
#> Readers of 'Meta-Analysis with R (Use R!)' should install
#> older version of 'netmeta' package: https://tinyurl.com/kyz6wjbb
nma_res <- netmeta(TE,seTE,treat1,treat2,studlab,
                   data=BRDdat,
                   sm="OR",comb.fixed = T,comb.random = F)

Assuming a new two-arm trial comparing Ceftiofur pin and Tildipirosin is to be planned. To apply the functions in OssaNMA to help to plan the new trial, we need to have the standard error of the estimated effect size between the two treatments, Ceftiofur pin and Tildipirosin, from the existing network. We can get the value by:

enma_sigma <- nma_res$seTE.fixed['Ceftiofur pin','Tildipirosin']
enma_sigma
#> [1] 0.2231349

Also, we need to know the risk of two treatments in the new trial. Some options are:

• Take ‘No active control’(NAC) as a baseline treatment, we can calculate the estimated log odds ratio between NAC and other treatments using NMA. As for the risk of NAC, we can get it pooling the arm-level data from the existing network if any.
• Use other source of evidence to assign the risk of two treatments.

Take the first option for example:

# The risk of NMA is calculate by pooling the arm-level data from the existing network. 
# The arm-level data is not provided in the package so the value is given directly here.
p_nac <- 0.68
# extract the log odds ratio between NAC and two treatments from nma_res
lor_nac_enro <- nma_res$TE.fixed['No active control','Ceftiofur pin']
lor_nac_flor <- nma_res$TE.fixed['No active control','Tildipirosin']
# calculate risk of Ceftiofur pin, name it as p1
p1 <- p_nac/(p_nac + exp(lor_nac_enro)*(1-p_nac))
# calculate risk of Tildipirosin, name it as p2
p2 <- p_nac/(p_nac + exp(lor_nac_flor)*(1-p_nac))

Take a look at p1 and p2:

p1
#> [1] 0.1868955

p2
#> [1] 0.2366667

With p1, p2, and enma_sigma obtained from the existing NMA, we can solve the minimum required total sample size for the new trial to achieve a pre-specified power using ssnma() or calculate the optimal sample size allocation to each treatment group with a fixed total sample size to maximize the power using ssanma(), as how we applied the two functions in the first two sections when we have specified values of the input parameters.

See the application using the p1, p2, and enma_sigma obtained from the existing NMA below:

# Analyze the new trial with the existing network
ssnma(p1 = p1, p2 = p2, enma_sigma = enma_sigma, power.level = 0.8, 
        sig.level = 0.05, method = "with", allocation = "uneven")
#> $sample_size
#> [1] 853 782
#> 
#> $power
#> [1] 0.8

# Analyze the new trial with the existing network and 
# keep the sample size of each group to be the same
ssnma(p1 = p1, p2 = p2, enma_sigma = enma_sigma, power.level = 0.8, 
        sig.level = 0.05, method = "with", allocation = "even")
#> $sample_size
#> [1] 819 819
#> 
#> $power
#> [1] 0.8

# Analyze the new trial without the existing network
ssnma(p1 = p1, p2 = p2, enma_sigma = enma_sigma, power.level = 0.8, 
        sig.level = 0.05, method = "without", allocation = "uneven")
#> $sample_size
#> [1] 1106 1015
#> 
#> $power
#> [1] 0.8

# Analyze the new trial without the existing network and 
# keep the sample size of each group to be the same
ssnma(p1 = p1, p2 = p2, enma_sigma = enma_sigma, power.level = 0.8, 
        sig.level = 0.05, method = "without", allocation = "even")
#> $sample_size
#> [1] 1062 1062
#> 
#> $power
#> [1] 0.8

# Analyze the new trial with the existing network
ssanma(p1 = p1, p2 = p2, enma_sigma = enma_sigma, N = 800, sig.level = 0.05,
        method = "with")
#> $sample_alloc
#> [1] 417 383
#> 
#> $power
#> [1] 0.588

# Analyze the new trial without the existing network
ssanma(p1 = p1, p2 = p2, enma_sigma = enma_sigma, N = 800, sig.level = 0.05,
        method = "without")
#> $sample_alloc
#> [1] 417 383
#> 
#> $power
#> [1] 0.406

# Even allocation and analyze the new trial with the existing network
ssanma(p1 = p1, p2 = p2, enma_sigma = enma_sigma, N = 800, 
        sig.level = 0.05, method = "with", allocation = "even")
#> $sample_alloc
#> [1] 400 400
#> 
#> $power
#> [1] 0.587

# Even allocation and analyze the new trial without the existing network
ssanma(p1 = p1, p2 = p2, enma_sigma = enma_sigma, N = 800, 
        sig.level = 0.05, method = "without", allocation = "even")
#> $sample_alloc
#> [1] 400 400
#> 
#> $power
#> [1] 0.405