Javelin diagrams: A graphical tool for probabilistic sensitivity analysis

To demonstrate post hoc robustness of decision problems to parameter estimates, analysts may conduct a probabilistic sensitivity analysis, assigning distributions to uncertain parameters and computing the probability of decision change. In contrast to classical threshold proximity methods of sensitivity analysis, no appealing graphical methods are available to present the results of a probabilistic sensitivity analysis. Here we introduce an analog of tornado diagrams for probabilistic sensitivity analysis, which we call javelin diagrams. Javelin diagrams are graphical augmentations of tornado diagrams displaying both the probability of decision change and the information value associated with individual parameters or parameter sets. We construct javelin diagrams for simple problems, discuss their properties, and illustrate their realistic application via a probabilistic sensitivity analysis of a seven-parameter decision analysis from the medical literature.