Probabilistic sensitivity analysis examines the parameter uncertainty of a health economic model by assigning probability distributions to uncertain input values. Repeated random draws generate a distribution of incremental costs and effects. A Cost-Effectiveness Acceptability Curve can represent these results as the probability that an option is cost-effective at different willingness-to-pay thresholds.
Modeling Uncertainty as a Distribution
A base case uses a single value for each model parameter, such as transition probabilities, costs, utility values, or relative effects. Probabilistic sensitivity analysis does not treat these values as point estimates. It models their statistical uncertainty using appropriate distributions and simulates many joint draws. This allows multiple parameters to change simultaneously, which can be relevant for the uncertainty of an overall model.
The results are often shown in a cost-effectiveness plane. Each simulation run provides a point with incremental costs and incremental effects compared to the comparator. The distribution of points reveals whether the model uncertainty predominantly points in one direction or whether it includes both additional costs and lower effects. In stochastic models, Monte Carlo draws can also affect the random component of the model.
From Point Cloud to Acceptability Curve
A Cost-Effectiveness Acceptability Curve summarizes, for different willingness-to-pay thresholds, the proportion of simulations in which the intervention achieves the highest expected net benefit. It does not answer whether a threshold is appropriate. It merely shows how sensitive the model decision is to a given value for an additional effect or a QALY.
According to the IQWiG methods, the point cloud of the probabilistic sensitivity analysis can be mapped as an acceptability curve or supplemented with a confidence ellipse. The German cost-benefit assessment does not use a universal willingness-to-pay threshold. A CEAC is therefore an instrument for representing uncertainty and not an automatic rule for reimbursement decisions.
Distinction: Deterministic Analysis and Clinical Sensitivity
In a univariate deterministic sensitivity analysis, one parameter is varied at a time; a tornado diagram typically ranks the resulting changes by their influence. A multivariate deterministic analysis changes several fixed values. In contrast, probabilistic sensitivity analysis varies parameters according to distributions and generates a joint distribution of results.
The existing entry ‘sensitivity-analysis’ may refer to statistical sensitivity analyses in clinical trials, such as alternative analysis populations or assumptions about missing data. The methods discussed here belong to health economic modeling. Their subject is costs, effects, utility values, and model parameters, not the confirmation of a primary clinical endpoint.
The distributions must match the parameters. Probabilities, for example, are represented differently from costs, which are often right-skewed, or relative risks, which cannot be negative. If unsuitable distributions are used, the simulations can generate values that are clinically or mathematically impossible. It is also important to distinguish whether a parameter was estimated from a study, whether multiple parameters share the same dataset, and whether structural uncertainty can even be captured by a parameter distribution.
The CEAC must not be confused with a probability that a therapy is effective for an individual patient. It describes the model uncertainty of an economic decision under an assumed threshold. A high curve can therefore arise despite insufficient clinical evidence if the model assumptions are narrow. The statement must always be read in conjunction with the model structure, data quality, and deterministic scenarios.
The number of simulations should be sufficient so that the point cloud and the curve do not merely reflect random fluctuations of a small simulation run. Reproducible settings and a traceable seed strategy facilitate technical review.
For the review of a PSA, a table of the parameter distributions used is therefore useful. It should contain means, dispersion, sources, and justifications so that the relationship between evidence and simulated results can be understood.
A plausible point cloud therefore does not exempt from a review of the model. Unjustified extrapolations, unsuitable comparators, or missing patient-relevant data are merely repeated by many simulations, not methodologically corrected.
Relevance for clinical trials
Study results provide distributions and uncertainty measures for key model inputs. Confidence intervals for treatment effects, distributions of utility values, and fluctuations in resource consumption should be documented in a way that allows them to be appropriately translated into a probabilistic analysis. The dependencies between parameters deserve special attention, because an independent draw of correlated values can create an unrealistic picture of uncertainty.
Full-service CROs like Mediconomics support by planning utility- and resource-relevant data collections and by preparing parameter estimates for economic models. Biostatistics can justify distributions and correlations, while Health Economics teams translate simulations, scatter plots, and acceptability curves into comprehensible report sections.
Frequently Asked Questions (FAQ)
Does a probabilistic sensitivity analysis replace the base case?
No. The base case provides the main calculation; the probabilistic analysis describes how parameter uncertainty changes its result distribution.
Does a Cost-Effectiveness Acceptability Curve determine a reimbursable price?
No. It shows model-based probabilities at given thresholds but does not decide on price or reimbursement.
Why is a tornado diagram not part of probabilistic sensitivity analysis?
It usually visualizes deterministic variations of single or fixed parameter combinations, not random simulation runs.
Regulatory References
- IQWiG, General Methods Version 8.0 – requires PSA in addition to deterministic analyses.
- Paragraph 35b SGB V – establishes the legal framework for the IQWiG mandate for cost-benefit analysis.
- G-BA Rules of Procedure, Chapter 5 – situates drug evaluations within the G-BA procedure.