Comment on: 'Vitruvian plot: a visualisation tool for multiple outcomes in network meta-analysis'.

Abstract

Ostinelli et al developed an interesting visualisation tool, the Vitruvian plot, to present multiple outcomes in network metaanalysis. We write to make some suggestions and potential improvements. First, to present the strength of statistical evidence, the authors colour the sectors according to the p values of a Ztest. According to the Cochran handbook, this could lead to overreliance and misinterpretation of p values, and assertive judgements about imprecision. One solution is to use a partially contextualised approach, according to the guidance of the GRADE working group. This approach also encourages researchers to use absolute values. After choosing reference intervention, researchers need to set thresholds for effects, which classify interventions into those with a trivial, small, moderate or large effect. The specific magnitude of the potential benefit or harm is more conducive to helping readers understand the evidence accurately. Second, in the Vitruvian plot, the increase in absolute event rate is in a 1:1 ratio with the increase in sector radius. Taking into account the general reader’s understanding of statistics, this could lead to misunderstandings. For example, if the radius is doubled, the area will be quadrupled. The sector area cannot accurately represent the corresponding absolute estimates. We suggest that the absolute event rate should be comparable to the ratio of shaded sectors to total sector area, which will avoid misunderstandings by graphically sensitive readers. Third, the original Vitruvian plot cannot show both the specific magnitude of the potential benefit or harm and the certainty of evidence at the same time. Especially when we use effect size to colour the sector, the certainty of evidence could present information of imprecision. We propose improvements to the Vitruvian plot to increase the presentation of the certainty of evidence. This facilitates the user to read the evidence coherently. Based on the original design, we improved the original Vitruvian plot as shown in figure 1. The data presented in the figure are fictitious. For demonstration purposes, we set the same threshold for each outcome. The green and red in the sector represent the two directions of the effect. The ratio of the shaded sector area to the total sector area is consistent with the absolute event rate. The area outside the circle is used to present the certainty of evidence. We find the Vitruvian plot to be very useful and sincerely hope that our suggestions could refine this visualisation tool for multiple outcomes in network metaanalysis.