Network meta-analysis using R: a review of currently available automated packages

Abstract

Network meta-analysis (NMA)--a statistical technique that allows comparison of multiple treatments in the same meta-analysis simultaneously--has become increasingly popular in the medical literature in recent years. The statistical methodology underpinning this technique and software tools for implementing the methods are evolving. Both commercial and freely available statistical software packages have been developed to facilitate the statistical computations using NMA with varying degrees of functionality and ease of use. This paper aims to introduce the reader to three R packages, namely, gemtc, pcnetmeta, and netmeta, which are freely available software tools implemented in R. Each automates the process of performing NMA so that users can perform the analysis with minimal computational effort. We present, compare and contrast the availability and functionality of different important features of NMA in these three packages so that clinical investigators and researchers can determine which R packages to implement depending on their analysis needs. Four summary tables detailing (i) data input and network plotting, (ii) modeling options, (iii) assumption checking and diagnostic testing, and (iv) inference and reporting tools, are provided, along with an analysis of a previously published dataset to illustrate the outputs available from each package. We demonstrate that each of the three packages provides a useful set of tools, and combined provide users with nearly all functionality that might be desired when conducting a NMA.

Figure 1. Network plots created by R packages a) gemtc, b) pcnetmeta, and c) netmeta.

Figure 2. Inconsistency-detecting heat map function netheat from the netmeta package applied to the diabetes data set.

Figure 3. A forest plot of the estimates of odds ratios between each treatment and the reference placebo created using the gemtc R package and diabetes data.

Figure 4. A sample of the detailed comparison-wise forest plots available from the gemtc R package outlining odds ratio estimates from contributing studies, direct evidence and indirect evidence using treatments 5 (diuretic) and 6 (placebo) from the diabetes data.

Figure 5. A forest plot of the estimates of odds ratios between each treatment and the reference placebo created using the netmeta R package and diabetes data.

Figure 6. A confidence interval plot from the pcnetmeta R package displaying estimates of the event rates for all treatments in the diabetes dataset.

Figure 7. A density plot from the pcnetmeta R package displaying posterior densities for estimates of the event rates for all treatments in the diabetes dataset.

Figure 8. A rank plot created using the rankogram function from the gemtc R package applied to the diabetes dataset illustrating empirical probabilities that each treatment is ranked 1^st through 6^th (left to right).