Step 5: Complete the Economic Evaluation and Sensitivity Analysis
Economic evaluations are complex and often require assumptions regarding costs and effectiveness where gaps exist. Complexities are compounded by lack of guidance regarding key assumptions. Sensitivity analysis is used to determine whether the results are robust to differences that arise from uncertainty (i.e., conclusions do not change when different assumptions are used in the analysis) and for this reason sensitivity analyses are a fundamental component of any health economic evaluation. Examples of uncertainty include uncertainty regarding the true value of a cost or effectiveness parameter, uncertainty regarding the generalizability of effectiveness results, and uncertainty regarding the structure of the model (Hunink & Glasziou, 2001). Various methods exist for testing uncertainty; however, in cost- effectiveness analyses, hypothesis testing is normally not done because negative and positive ICERs can have two meanings. For example, in Figure 18.2, an ICER is negative when it is in Southwest or Northeast regions. Yet, the interpretation of the ICER is different in Southwest region compared to Northeast region. Furthermore, the ICER is a ratio statistic and ratios are not normally distributed.
One-way sensitivity analysis is the most common type of sensitivity analysis. In a one-way sensitivity analysis, an individual parameter is varied between a low and high value and all other parameters are held constant. In turn, this generates a low and high value for the ICER (in the case of a cost-effectiveness analysis) and indicates the relative impact of a parameter. The range can be based on a confidence interval, clinical opinion, or a best- and worst-case scenario. A threshold analysis is a form of a one-way sensitivity analysis. In a threshold analysis, a parameter is changed until a condition is met. For example, a parameter may be changed to find the point at which the ICER rises above $50,000 per QALY. One-way sensitivity analysis will generally understate uncertainty if there is uncertainty in multiple parameters. A two-way sensitivity analysis can be used to vary two parameters at the same time. Finally, a probabilistic sensitivity analysis can be used to account for multiple parameters changing simultaneously. A probabilistic sensitivity analysis takes into account uncertainty by assuming there is a distribution around each parameter. The distribution around parameters in turn can be used to generate a distribution of ICERs. Distributions of ICERs can be generated from bootstrapping (if parameters are based on participant-level data) or Monte Carlo simulation (if a decision analytic model is used). The distribution of ICERs can then be used to determine the probability that an ICER will be less than a given cost-effectiveness threshold.