RM41 - A novel approach to ranking parameter uncertainty in one-way sensitivity analysis: What tornado diagrams are missing

Abstract

Economic evaluations of health technologies have long relied on one-way sensitivity analysis (SA) to examine the impact of parameter uncertainty on modeling outcomes. Traditionally, this impact has been measured and ranked based on absolute changes in the incremental cost-effectiveness ratio (ICER) across plausible parameter values and presented in a tornado diagram. This format does not adequately identify or prioritize parameters where the range of uncertainty causes the ICER to change quadrants in the cost-effectiveness (CE) plane. However, these quadrant changes, which represent fundamental changes to the CE conclusion, are arguably more meaningful than changes in the ICER within a quadrant. This research illustrates a novel approach to presenting one-way SA results that focuses on identifying parameters with the greatest potential to change the overall CE conclusion rather than narrowly focusing on changes to the ICER. We developed a comprehensive algorithm for ranking the parameters varied in a one-way SA. Broadly, we first prioritize parameters with the potential to qualitatively change the CE conclusion and then rank parameters based on quantitative changes to modeling outcomes. Changes to the CE conclusion are identified based on quadrant changes in the CE plane, and parameters are categorized as having the potential to change the conclusion both positively and negatively, only positively, only negatively, or not at all. Within these categories, a secondary ranking based on costs and health outcomes is used. Furthermore, visualization techniques anchored in the CE plane help assess whether conclusion changes are due primarily to changes in health, changes in costs, or both. This research demonstrates that the conclusions-based ranking algorithm works in more general settings than the traditional tornado diagram format. The conclusions-based approach is a powerful method that provides a more complete picture of the impact of parameter uncertainty in economic evaluations.