A New Approach to One-Way Sensitivity Analysis and the Tornado Plot

OBJECTIVES: Traditional one-way sensitivity analysis (OWSA) involves setting each parameter at its upper and lower bound values while all others stay at their mean. It is a systematic series of scenario analyses with individual parameters set at their extreme values and is interpreted as a measure of the influence of individual parameters on uncertainty. We explored a simple inversion of the analysis, where one parameter is set at its mean while in a probabilistic model. This approach gives an estimate of the magnitude of uncertainty eliminated by fixing the parameter, from which we infer the uncertainty attributable to the parameter.

METHODS: This study used a Markov cost-utility model of herpes zoster vaccination in the general population at age 50 years, run for 20,000 simulations. The outcome was expressed as net monetary benefit (NMB) at a willingness-to-pay threshold of €20,000 per QALY. We conducted OWSA using both the traditional approach (set one parameter at upper and lower bounds with all others at their mean) and the alternative (set one parameter at its mean and vary all others).

RESULTS: The mean NMB was €23.5 million (95% CI: €18.8m to €28.3m). Using the traditional approach to OWSA, one parameter was prominent: the cost of the vaccine. Further parameters that were considered important included the probability of developing post-herpetic neuralgia and the cost of administering the vaccine. Using the alternative approach, fixing the cost of the vaccine at its mean reduced the width of the confidence bounds of the NMB by 66% (from €9.5m to €3.2m). No other single parameter reduced the width of the confidence bounds by more than 2%.

CONCLUSIONS: Analysis of eliminating uncertainty offers a different perspective on the influence of each parameter on overall uncertainty, and provides a useful additional tool alongside the traditional tornado plot.