Sensitivity Analysis in Economic Evaluation

Sensitivity analysis in economic evaluation is the set of methods for examining how robust a cost-effectiveness conclusion is to uncertainty in its inputs and assumptions. Because the costs, effects, and structural choices that drive a model are estimated rather than known with certainty, analysts vary them—singly, jointly, or probabilistically—to show how the decision might change and how confident a conclusion can be.

Definition

Sensitivity analysis in economic evaluation is the systematic exploration of how the results of a cost-effectiveness analysis change when uncertain parameters, structural assumptions, or methodological choices are varied, ranging from one-way deterministic analysis to fully probabilistic analysis that propagates joint parameter uncertainty.

Scope

The entry covers deterministic approaches (one-way and multi-way analysis, scenario analysis, threshold analysis) and probabilistic sensitivity analysis, together with the tools used to summarise uncertainty for decision-makers, such as the cost-effectiveness plane, cost-effectiveness acceptability curves, and the expected value of information. It is methodological reference material and does not advise on any specific intervention. This node is distinct from the epidemiology entry on sensitivity analysis for unmeasured confounding, which it cross-links as a neighbour.

Core questions

• How much does the cost-effectiveness conclusion depend on any single uncertain input?
• How is the joint uncertainty in all parameters propagated to the result?
• How should uncertainty be summarised for a decision-maker?
• What is the value of collecting more evidence before deciding?

Key concepts

• Deterministic (one-way and multi-way) sensitivity analysis
• Scenario and threshold analysis
• Probabilistic sensitivity analysis
• Parameter, structural, and methodological uncertainty
• Cost-effectiveness plane
• Cost-effectiveness acceptability curve
• Monte Carlo simulation
• Expected value of information

Mechanisms

Deterministic sensitivity analysis varies one or a few parameters across plausible ranges to identify the drivers of the result, while threshold analysis finds the input value at which the decision would switch. Probabilistic sensitivity analysis instead assigns probability distributions to uncertain parameters and uses Monte Carlo simulation to draw many parameter sets, producing a distribution of incremental costs and effects. These are displayed on the cost-effectiveness plane and summarised as cost-effectiveness acceptability curves showing the probability that each option is cost-effective at different thresholds. Value-of-information analysis goes further, quantifying the expected benefit of reducing uncertainty through additional research (Briggs et al., 2012; Fenwick et al., 2004; Claxton et al., 2005; Van Hout et al., 1994).

Clinical relevance

Uncertainty analysis tells decision-makers how secure a cost-effectiveness recommendation is and whether more evidence is needed before committing resources, so it shapes the confidence placed in health technology assessment outputs. This topic describes uncertainty methodology and how economic evidence is appraised; it is not guidance for individual clinical or treatment decisions.

Evidence & guidelines

The ISPOR-SMDM good-practice report on parameter estimation and uncertainty sets out expectations for characterising and propagating uncertainty, and bodies such as NICE require probabilistic sensitivity analysis in technology assessment; methodological papers established the cost-effectiveness acceptability curve and the cost-effectiveness plane as standard summaries (Briggs et al., 2012; Claxton et al., 2005; Fenwick et al., 2004; Van Hout et al., 1994).

History

Early economic evaluations relied on simple one-way sensitivity analyses, but the recognition that parameters are jointly uncertain led, through the 1990s, to probabilistic methods using Monte Carlo simulation. Van Hout and colleagues introduced graphical tools on the cost-effectiveness plane in 1994, cost-effectiveness acceptability curves became the standard way to present uncertainty, and by the mid-2000s reimbursement bodies were requiring probabilistic sensitivity analysis as a default rather than an option (Van Hout et al., 1994; Fenwick et al., 2004; Claxton et al., 2005).

Debates

Is probabilistic sensitivity analysis essential or optional?

Some argued deterministic analyses suffice, but because parameters are jointly uncertain and cost-effectiveness results are non-linear, influential work held that probabilistic sensitivity analysis is necessary to characterise decision uncertainty correctly; this position became the norm for major assessment bodies.

Key figures

• Andrew Briggs
• Karl Claxton
• Mark Sculpher
• Elisabeth Fenwick
• Ben van Hout

Seminal works

• briggs-2012-uncertainty
• fenwick-2004-ceac
• vanhout-1994

Frequently asked questions

What is the difference between deterministic and probabilistic sensitivity analysis?

Deterministic analysis varies one or a few inputs by hand across chosen ranges to see how the result moves, whereas probabilistic analysis assigns probability distributions to all uncertain inputs and uses simulation to capture their joint effect on the result.

What does a cost-effectiveness acceptability curve show?

It plots, across a range of cost-effectiveness thresholds, the probability that an intervention is the cost-effective choice given the uncertainty in the analysis, giving decision-makers a direct read on how secure the conclusion is.