An overview on rankinma

Brief overview

Package rankinma supports users to easily obtain and visualize various metrics of treatment ranking from network meta-analysis no matter using frequentist or Bayesian approach. This package not only accepts manual-prepared data set of treatment ranking metrics from users, but also can help users to gather various treatment ranking metrics in network meta-analysis. Users can use functions in rankinma by calling the library with following syntax:

library(rankinma)

Feature

rankinma allows users to visualize various treatment ranking metrics in network meta-analysis based either common-effect model or random-effects model. The current version includes three common metrics of treatment ranking.

Probabilities: probabilities of every available treatment on each possible rank.
SUCRA: the value of surface under the cumulative ranking curve using Bayesian approach.
P-score: the value of SUCRA using frequentist approach.

Briefly, rankinma can be used for visualization of both detailed metrics of probabilities and global metrics (i.e. SUCRA and P-score). Besides, rankinma provides users multiple types of plots to illustrate aforementioned treatment ranking metrics, and current version consists of five types of plots with six sub-types.

Beading plot: a novel graphics for displaying global metrics of treatment ranking (i.e. SUCRA, P-score, and P-best) based on numeric line plot.
Bar chart: a classic graphics for most metrics of treatment ranking (i.e. probabilities, SUCRA, and P-score), and rankinma supports two sub-type of bar chart in terms of side-by-side bar chart and cumulative bar chart.
Line chart: a classic graphics for most metrics of treatment ranking (i.e. probabilities, SUCRA, and P-score), and rankinma supports two sub-type of line chart in terms of simple line chart (a line on a chart) and composite line chart (multiple lines on a chart).
Heat plot: a new graphics for showing global metrics of treatment ranking (i.e. SUCRA and P-score) for each outcome, and rankinma supports to gather all heat plots of outcomes with interests on a plot.
Spie chart: a new graphics proposed in 2020 for displaying multiple global metrics of treatment ranking (i.e. SUCRA and P-score) from outcomes with interests by each treatment, and rankinma supports to place all spie charts on a plot.

Flow and functions

Users can visualize treatment ranking after network meta-analysis in five steps, but have to check condition before using rankinma.

Situation 1: Users have data for network meta-analysis of a single outcome but do not get treatment ranking metrics yet.

Situation 2: Users have data for network meta-analysis of various outcomes but do not get treatment ranking metrics yet.

For situation 1:

Step 1. Load data and do network meta-analysis.

Step 2. Get treatment ranking metrics from the network meta-analysis using function GetMetrics().

Step 3. Setup data in rankinma format using function SetMetrics().

Step 4. Visualization using function PlotBeads(), PlotHeat(), PlotBar(), or PlotLine().

For situation 2:

Step 1. Load data and do network meta-analysis.

Step 2. Get treatment ranking metrics from the network meta-analysis using function GetMetrics().

— Repeat step 1 and 2 for each outcome, and keep output of them for the further steps. —

Step 3. Combine treatment ranking metrics using function rbind() in R base.

Step 4. Setup data in rankinma format using function SetMetrics().

Step 5. Visualization using function PlotBeads(), PlotHeat(), PlotBar(), or PlotLine().

Usage and examples

The following steps and syntax demonstrate how user can illustrate a summary of treatment ranking metrics on various outcomes from network meta-analysis.

Example 1 for illustrating line chart when users have data for network meta-analysis of a single outcome but do not get treatment ranking metrics yet.

STEP 1. Load data

library(rankinma)
library(netmeta)
data(Senn2013)
nmaOutput <- netmeta(TE, 
                     seTE, 
                     treat1, 
                     treat2, 
                     studlab, 
                     data = Senn2013, 
                     sm = "SMD")

STEP 2. Get Probabilities

dataMetrics <- GetMetrics(nmaOutput, 
                          outcome = "HbA1c.random", 
                          prefer = "small", 
                          metrics = "Probabilities", 
                          model = "random", 
                          simt = 1000)

STEP 3. Set data for rankinma

dataRankinma <- SetMetrics(dataMetrics, 
                           tx = tx, 
                           outcome = outcome, 
                           metrics.name = "Probabilities")

STEP 4. Visualize the probabilities of treatments among possible ranks

If the users wish to visualize the information using line chart, the following syntax could be a reference.

PlotLine(data = dataRankinma, 
         compo = TRUE)

Figure 1. Composite line chart for probabilities of treatments on each rank.

If the users wish to visualize the information using stacked bar chart, the following syntax could be a reference.

PlotBar(data = dataRankinma, 
        accum = TRUE)

Figure 2. Accumulative bar chart for probabilities of treatments on each rank.

Example 2 for illustrating beading plot when users have data for network meta-analysis of multiple outcomes but do not get treatment ranking metrics yet.

STEP 1. Load data

library(rankinma)
library(netmeta)
data(Senn2013)
nmaOutput <- netmeta(TE, 
                     seTE, 
                     treat1, 
                     treat2, 
                     studlab, 
                     data = Senn2013, 
                     sm = "SMD")

STEP 2. Get global metrics Due to several rank metrics in network meta-analysis, users have to decide which global metric would be used in the analysis; accordingly users choose argument for parameter metrics in the function GetMetrics(). The following syntax is for users who would like to obtain the Surface Under the Cumulative Ranking Curve (SUCRA).

Get SUCRA

nmaRandom <- GetMetrics(nmaOutput, 
                        outcome = "HbA1c.random", 
                        prefer = "small", 
                        metrics = "SUCRA", 
                        model = "random", 
                        simt = 1000)
nmaCommon <- GetMetrics(nmaOutput, 
                        outcome = "HbA1c.common", 
                        prefer = "small", 
                        metrics = "SUCRA", 
                        model = "common", 
                        simt = 1000)

If users wish to obtain the P-score, they can utilize the following syntax.

Get P-score

nmaRandom <- GetMetrics(nmaOutput, 
                        outcome = "HbA1c.random", 
                        prefer = "small", 
                        metrics = "P-score", 
                        model = "random", 
                        simt = 1000)
nmaCommon <- GetMetrics(nmaOutput, 
                        outcome = "HbA1c.common", 
                        prefer = "small", 
                        metrics = "P-score", 
                        model = "common", 
                        simt = 1000)

If users wish to obtain the P-best, they can assign the argument P-best to the parameter metrics in the function GetMetrics(). The following syntax is for reference.

Get P-best

nmaRandom <- GetMetrics(nmaOutput, 
                        outcome = "HbA1c.random", 
                        prefer = "small", 
                        metrics = "P-best", 
                        model = "random", 
                        simt = 1000)
nmaCommon <- GetMetrics(nmaOutput, 
                        outcome = "HbA1c.common", 
                        prefer = "small", 
                        metrics = "P-best", 
                        model = "common", 
                        simt = 1000)

STEP 3. Combine metrics from multiple outcomes

dataMetrics <- rbind(nmaRandom, nmaCommon)

STEP 4. Set data for rankinma As mentioned above, users choose argument for parameter metrics in the function GetMetrics() according to which global metric would be used in analysis due to several rank metrics in network meta-analysis. The following syntax is for users who would like to use SUCRA.

Set data based on SUCRA

dataRankinma <- (dataMetrics, 
                 tx = tx, 
                 outcome = outcome, 
                 metrics = SUCRA, 
                 metrics.name = "SUCRA")

Set data based on P-score

dataRankinma <- (dataMetrics, 
                 tx = tx, 
                 outcome = outcome, 
                 metrics = P.score, 
                 metrics.name = "P-score")

Set data based on P-best

dataRankinma <- (dataMetrics, 
                 tx = tx, 
                 outcome = outcome, 
                 metrics = P.best, 
                 metrics.name = "P-best")

STEP 5. Visualize global ranking metrics

Illustrate beading plot based on SUCRA (without relative effect)

PlotBeads(data = dataRankinma)

Figure 3A. Beading plot for SUCRA on two outcomes

Illustrate beading plot sorted by P-score (with relative effects and equidistant by rank)

PlotBeads(data = dataRankinma, 
          scaleX = "Rank",
          txtValue = "Effects")

Figure 3A. Beading plot for SUCRA on two outcomes

Illustrate colorblind-friendly beading plot based on P-best

PlotBeads(data = dataRankinma,
          lgcBlind = TRUE)

Figure 3C. Colorblind friendly beading plot for P-score on two outcomes

Illustrate spie plot based on P-score

PlotSpie(data = dataRankinma)

Figure 3B. Spie plot for P-score on two outcomes

Reference

Salanti, G., Ades, A. E., & Ioannidis, J. P. (2011). Graphical methods and numerical summaries for presenting results from multiple-treatment meta-analysis: an overview and tutorial. Journal of clinical epidemiology, 64(2), 163-171.

Chaimani, A., Higgins, J. P., Mavridis, D., Spyridonos, P., & Salanti, G. (2013). Graphical tools for network meta-analysis in STATA. PloS one, 8(10), e76654.

Van Valkenhoef, G., Tervonen, T., Zwinkels, T., De Brock, B., & Hillege, H. (2013). ADDIS: a decision support system for evidence-based medicine. Decision Support Systems, 55(2), 459-475.

Rücker, G., & Schwarzer, G. (2015). Ranking treatments in frequentist network meta-analysis works without resampling methods. BMC medical research methodology, 15(1), 1-9.

Daly, C. H., Mbuagbaw, L., Thabane, L., Straus, S. E., & Hamid, J. S. (2020). Spie charts for quantifying treatment effectiveness and safety in multiple outcome network meta-analysis: a proof-of-concept study. BMC Medical Research Methodology, 20, 1-13.

Balduzzi, S., Rücker, G., Nikolakopoulou, A., Papakonstantinou, T., Salanti, G., Efthimiou, O., & Schwarzer, G. (2023). netmeta: An R package for network meta-analysis using frequentist methods. Journal of Statistical Software, 106, 1-40.