One-Way Sensitivity Analysis for Probabilistic Cost-Effectiveness Analysis: Conditional Expected Incremental Net Benefit

Abstract

Although probabilistic analysis has become the accepted standard for decision analytic cost-effectiveness models, deterministic one-way sensitivity analysis continues to be used to meet the need of decision makers to understand the impact that changing the value taken by one specific parameter has on the results of the analysis. The value of a probabilistic form of one-way sensitivity analysis has been recognised, but the proposed methods are computationally intensive. Deterministic one-way sensitivity analysis provides decision makers with biased and incomplete information whereas, in contrast, probabilistic one-way sensitivity analysis (POSA) can overcome these limitations, an observation supported in this study by results obtained when these methods were applied to a previously published cost-effectiveness analysis to produce a conditional incremental expected net benefit curve. The application of POSA will provide decision makers with unbiased information on how the expected net benefit is affected by a parameter taking on a specific value and the probability that the specific value will be observed.

Probabilistic One-Way Sensitivity Analysis with Multiple Comparators: The Conditional Net Benefit Frontier

“…Under these circumstances, decision makers are interested in whether a parameter could take a value that would change the most cost-effective strategy. As McCabe et al described, one of the key differences between deterministic one-way sensitivity analysis and probabilistic one-way sensitivity analysis (POSA) is that POSA tells the decision maker not only whether such a value is possible, but also the probability that such a value will be observed [ 2 ].…”

“…Generating the data required to construct the cNBF is identical to the processes described for conducting POSA and estimating cINB [ 2 ]. Having chosen a parameter for which POSA is required, a set of values is selected from the distribution for that parameter, covering the full range of possible values.…”

“…Decision makers are often interested in whether changes in specific parameters in the evidence base could change the assessment of whether the strategy being considered represents good or poor value for money. McCabe et al [ 2 ] describe how the conventional approach to addressing this question, deterministic one-way sensitivity analysis, provides decision makers with biased results and hence harms the quality of decision making. They go on to describe the ‘conditional incremental net benefit’ (cINB) approach for one-way sensitivity analysis; this approach respects non-linearities in cost-effectiveness models and correlations between model parameters in order to provide less biased results.…”

The Application and Implications of Novel Deterministic Sensitivity Analysis Methods

“…Current DSA methods do not provide reliable insight into the changes to the outcome of the model due to individual parameter uncertainty. Limitations to the classic DSA approach have been well known for years [10][11][12], and include that (i) the parameter ranges included in DSAs are often chosen arbitrarily and no insight in effects at the margin is provided, (ii) non-linearities in models are not visible, (iii) results of DSAs do not provide insight into the likelihood of the reported parameter values, (iv) correlations between parameters are not taken into account and (v) DSAs are usually reported in the incremental costeffectiveness ratio (ICER), the mathematical properties of which bring several limitations [1,[12][13][14][15][16].…”

A Computationally Efficient Method for Probabilistic Parameter Threshold Analysis for Health Economic Evaluations

“…In addition, the relationship between the uncertain input parameters and the corresponding uncertainty around the cost-effectiveness of a health care strategy can be assessed using a range of statistical methods, including value of information analysis. 6 McCabe et al 7 proposed a probabilistic threshold analysis based on a 2-level Monte Carlo approach. In complex health economic evaluations, a 2-level Monte Carlo simulation can be computationally demanding.…”