Tornado Diagram

A tornado diagram (given this name due to its resemblance to a tornado) is a graphical method for displaying a series of univariate (or one-way) sensitivity analyses that has been commonly used in cost-effectiveness analysis. Figure 1 shows an example of a tornado diagram—with incremental cost-effectiveness value ranges arranged from the largest at the top to the smallest at the bottom. In a tornado diagram, the effects of individual parameter variation on results can be compared visually, allowing analysts to intuitively communicate which parameters are more sensitive to variation— that is, parameters whose variation has the greatest effects on the results of the analysis. In the example shown in Figure 1, individual variation of Drug A costs has the greatest effects on model results, and variation of Drug A side effect utility has the least effect. More recently, the use of tornado diagrams and univariate sensitivity analyses has been downplayed due to limitations of these methods and due to the use of other techniques that overcome these limitations.

To construct a tornado diagram, the analyst varies a single parameter over its range and notes the effects of this variation on the analysis results. In Figure 1, the incremental cost-effectiveness varies from $10,000 per quality-adjusted life year (QALY) gained when Drug A costs $54 to $97,000/QALY gained when Drug A costs $139. The vertical line through the bars at $39,000/QALY shows the base case results of the analysis, the expected value of the analysis when all parameters are set at their base case point estimates. The analyst will then repeat this procedure for each of the remaining parameters to be varied. Once the result range for each parameter has been calculated, parameters are arranged in the diagram from greatest to least effect on model results.

Some decision analysis software packages allow construction of tornado diagrams within the program itself. Tornado diagrams may also be constructed using the graphing functions of Microsoft Excel or through the use of a variety of Excel add-in programs.

Tornado diagrams have a number of limitations. They can be useful when two strategies are being compared, since tornado diagrams typically depict changes in the incremental cost-effectiveness ratio between two strategies. However, when three or more strategies are being considered, tornado diagrams become more difficult to use due to the differential effects of individual-parameter variation on multiple competing strategies and, at times, the effects of strategy dominance. Due to these shortcomings, alternative structures for tornado diagrams have been proposed. One such structure uses the net benefit framework to convert incremental cost-effectiveness ratios to either net monetary or net health benefits, thus bypassing some of the shortcomings of the conventional tornado diagram. However, this formulation gives a less intuitive picture of sensitive parameters and requires, [Page 1139]due to the net benefit calculation, the choice of a single cost-effectiveness acceptability threshold.

Figure 1 A tornado diagram, depicting changes in incremental cost-effectiveness ratios (the x-axis) when individual parameter values are varied

Note: The numbers at either end of the horizontal bar are the high and low parameter values considered in a univariate sensitivity analysis. The vertical line depicts the base case incremental cost-effectiveness ratio of the model.

...