With public health policy increasingly relying on mathematical models to provide insights about the impacts of potential policy options, the demand for uncertainty and sensitivity analyses that explore the implications of different assumptions in a model continues to expand. Although analysts continue to develop methods to meet the demand, most modelers rely on a single method in the context of their assessments and presentations of results, and few analysts provide results that facilitate comparisons between uncertainty and sensitivity analysis methods.
Methods: vary in their degree of analytical difficulty and in the nature of the information that they provide, and analysts should communicate results with a note that not all methods yield the same insights. The authors explore several sensitivity analysis methods to test whether the choice of method affects the insights and importance rankings of inputs from the analysis. They use a dynamic cost-effectiveness model of a hypothetical infectious disease as the basis to perform 1-way and multi-way sensitivity analyses, design of experiments, and Morris' method. They also compute partial derivatives as well as a number of probabilistic sensitivity measures, including correlations, regression coefficients, and the correlation ratio, to demonstrate the existing methods and to compare them. The authors find that the magnitudes and rankings of sensitivity measures depend on the selected method(s) and make recommendations regarding the choice of method depending on the complexity of the model, number of uncertain inputs, and desired types of insights from the sensitivity analysis.