Uncertainty and Sensitivity Analysis in Health Economic Evaluation: Principles, Tools, and Application
On this page
Introduction
In health economics, decisions are rarely made under conditions of perfect certainty. Clinical outcomes, cost estimates, and utility values used in economic models often rely on assumptions or incomplete evidence. To ensure that policy conclusions drawn from such models are reliable and transparent, uncertainty and sensitivity analyses are indispensable. These techniques allow researchers to explore the robustness of their findings and inform decision-makers about the confidence they can place in cost-effectiveness results.
This article explains the types of uncertainty that arise in health economic modeling, introduces the methods used to assess their impact, and demonstrates how these methods guide policy under uncertainty.
I. Understanding Uncertainty in Economic Models
Nature of Uncertainty
Uncertainty refers to a lack of precise knowledge about inputs or structural elements within a model. Key types include:
- Parameter Uncertainty: Arises from imprecision in data inputs such as treatment effectiveness, event probabilities, or healthcare costs.
- Model Structure Uncertainty: Reflects ambiguity in how the model is built—e.g., which health states are included, or how transitions are defined.
- Methodological Uncertainty: Pertains to choices about model perspective, time horizon, or discount rate.
In practice, parameter uncertainty is the most commonly assessed, given its quantifiable nature and availability of statistical distributions.
II. Sensitivity Analysis: A Systematic Response to Uncertainty
Sensitivity analysis investigates how variation in input values affects model outputs. It acts as a diagnostic tool to identify the most influential parameters and provides transparency about the stability of conclusions.
1. One-Way Sensitivity Analysis (Univariate)
This approach changes one parameter at a time while holding others constant to assess its specific influence on results. It helps answer the question: Which variable, if misestimated, could reverse the policy conclusion?
Visual Tool:
- Tornado diagrams rank parameters by the extent to which their variation alters outcomes. Parameters with the longest bars have the most influence on the incremental cost-effectiveness ratio (ICER).
Example: Imagine assessing the cost-effectiveness of a new anticoagulant. By varying the drug cost ±20%, one could see whether this alone shifts the ICER above or below a willingness-to-pay (WTP) threshold.
2. Multi-Way Sensitivity Analysis
Here, two or more parameters are changed simultaneously, often as scenario analyses. For example, both utility values and treatment costs could be varied in a combined “best-case” or “worst-case” scenario.
III. Probabilistic Sensitivity Analysis (PSA): A Holistic Simulation
Unlike one-way sensitivity analysis, PSA considers joint uncertainty across all parameters. It simulates thousands of scenarios by drawing input values from predefined probability distributions.
1. Distributions Used for Inputs
Different data types require distinct distribution forms:
| Parameter Type | Distribution | Rationale |
|---|---|---|
| Probabilities/Proportions | Beta | Bound between 0 and 1 |
| Utilities (0 = death; 1 = full health) | Beta | Captures uncertainty in quality of life |
| Relative Risks or ORs | Log-normal | Reflects ratio metrics with skewness |
| Costs | Gamma | Skewed right, no negative values |
2. Monte Carlo Simulation
Each simulation run samples a value from each parameter’s distribution and computes outcomes like ICER. Repeating this process 1,000 to 10,000 times yields a cloud of results that reflects uncertainty across all variables simultaneously.
Visual Output:
- Cost-Effectiveness Scatter Plot: Dots spread across a plane of incremental cost vs. incremental effectiveness (e.g., QALYs).
- Cost-Effectiveness Acceptability Curve (CEAC): Summarizes the probability that an intervention is cost-effective across varying WTP thresholds.
IV. From Scatterplots to CEAC: Interpreting Probabilistic Results
While scatterplots illustrate variability, CEACs convey decision-relevant probabilities. They answer the question: At a given WTP per QALY, what is the chance that this intervention is a good investment?
Example:
- At a WTP of $5,000/QALY, a program might be cost-effective in 48% of simulations.
- At $15,000/QALY, this rises to 76%, indicating increasing confidence as the threshold grows.
Such curves help policymakers visualize the tradeoff between economic risk and health gain, especially when budgets are constrained.
V. Applying Uncertainty Analysis to Real Policy Scenarios
Case Application: Revaccination Policy
A model comparing two vaccination strategies for measles revaccination in adolescents showed that reducing the assumed exposure probability (from 20% to 15%) led to fewer deaths averted. This illustrates how a single input can shift public health outcomes and potentially change recommendations.
Case Application: Pharmacogenomics
In evaluating HLA-B*5801 genetic screening to prevent severe drug reactions, uncertainty analysis revealed how variation in test cost, disease incidence, and adverse outcome probabilities influenced cost-effectiveness. Results presented through tornado diagrams, scatterplots, and CEACs offered decision-makers a robust picture of risk and value.
VI. Communicating Uncertainty Transparently
While technical results are crucial, communication is equally important. Uncertainty analysis should be:
- Visually intuitive: Use graphs that simplify probabilistic data.
- Clinically contextualized: Translate findings into real-world implications (e.g., lives saved, hospitalizations avoided).
- Policy-attuned: Align results with WTP thresholds or budget impact considerations relevant to the country or payer.
Conclusion
Uncertainty and sensitivity analysis are not mere statistical accessories—they are essential for credible, defensible health economic evaluation. By revealing how assumptions and data variability influence outcomes, these methods enhance transparency, guide resource allocation, and safeguard against misguided conclusions. For any intervention being considered for inclusion in national health benefits or formularies, the ability to handle uncertainty robustly is as important as the base-case result itself.
Key Takeaways
- Uncertainty is inherent in economic evaluation—handling it openly is crucial.
- One-way sensitivity analysis identifies influential parameters; PSA evaluates overall robustness.
- Distribution choice must match the nature of the data (probability, cost, utility).
- CEACs convey policy-relevant insight: What’s the probability this intervention is worth funding?
- Visualization and clarity in communication make complex uncertainty accessible to non-technical stakeholders.