One of the advantages of network meta-analysis is that it can provide information about the ranking of all evaluated interventions for the studied outcome [2] (link), [42] (link). Probabilities are often estimated for a treatment being ranked at a specific place (first, second, etc.) according to the outcome.
Ranking of treatments based solely on the probability for each treatment of being the best should be avoided. This is because the probability of being the best does not account for the uncertainty in the relative treatment effects and can spuriously give higher ranks to treatments for which little evidence is available. So-called rankograms and cumulative ranking probability plots have been suggested as a reliable and comprehensive graphical way to present ranking probabilities and their uncertainty [2] (link). A rankogram for a specific treatment j is a plot of the probabilities of assuming each of the possible T ranks (where T is the total number of treatments in the network). The cumulative rankograms present the probabilities that a treatment would be among the n best treatments, where n ranges from one to T. The surface under the cumulative ranking curve (SUCRA), a simple transformation of the mean rank, is used to provide a hierarchy of the treatments and accounts both for the location and the variance of all relative treatment effects [2] (link). The larger the SUCRA value, the better the rank of the treatment.
Themvmeta command can provide ranking probabilities using the option pbest(min|max, all zero) . Options min or max specify whether larger or smaller treatment effects define a better treatment, while all and zero specify the estimation of probabilities for all possible ranks including the reference treatment. The estimated probabilities can be stored as additional variables in the dataset by adding the suboption gen() in pbest() and predictive ranking probabilities (the probability that each treatment will be placed in each rank in a future study [39] (link), [40] ) can be estimated with the suboption predict .
Our STATA commandsucra produces rankograms and computes SUCRA values using the ranking probabilities (e.g. as estimated with the mvmeta ) as input. If prob1 prob2 etc, is a list of variables including all ranking probabilities (one variable per treatment for each possible rank) as derived from mvmeta then typing
. sucra prob*,mvmetaresults plots the cumulative rankograms for all treatments.
InFigure 7 we present cumulative rankograms for the network of rheumatoid arthritis trials. The SUCRA values provide the hierarchy for the six active treatments; 1.8%, 59.9%, 66.2%, 21.8%, 75.9%, 41%, 83.4% for placebo, abatacept, adalimumab, anakinra, etanercept, infliximab, rituximab respectively. The cumulative rankograms can also be used to compare different models. In Figure 7 we present also the results from a network meta-regression accounting for small-study effects (using the variance of the log-odds ratios as covariate). The graph shows that small-study effects materially alter the relative effectiveness and ranking of treatments and adjustment will put etanercept and anakira in more favourable order compared with rituximab and abatacept respectively. The option compare() in the command sucra can be used to compare two ranking curves.
Ranking of treatments based solely on the probability for each treatment of being the best should be avoided. This is because the probability of being the best does not account for the uncertainty in the relative treatment effects and can spuriously give higher ranks to treatments for which little evidence is available. So-called rankograms and cumulative ranking probability plots have been suggested as a reliable and comprehensive graphical way to present ranking probabilities and their uncertainty [2] (link). A rankogram for a specific treatment j is a plot of the probabilities of assuming each of the possible T ranks (where T is the total number of treatments in the network). The cumulative rankograms present the probabilities that a treatment would be among the n best treatments, where n ranges from one to T. The surface under the cumulative ranking curve (SUCRA), a simple transformation of the mean rank, is used to provide a hierarchy of the treatments and accounts both for the location and the variance of all relative treatment effects [2] (link). The larger the SUCRA value, the better the rank of the treatment.
The
Our STATA command
In
Full text: Click here