Uncertainty Quantification

The formal process of measuring how much confidence should be placed in a model's results, given the sources of uncertainty affecting it.

Source: Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford University Press; 2006. doi:10.1093/oso/9780198526629.001.0001.

Simply noting that a result is uncertain tells a decision-maker little about how much to worry. Uncertainty quantification goes further and measures the uncertainty, expressing it as ranges, distributions, or probabilities, so that its size and its effect on the decision can be judged. This lets a firm result be distinguished from a fragile one, and shows whether the uncertainty is large enough to change what should be done. Turning a vague caveat into a measured quantity is what makes it useful. Briggs and colleagues (2006) describe this.

Source: Briggs et al. 2006

Uncertainty quantification is carried out by identifying the sources of uncertainty, parameter, structural, and methodological, representing them appropriately, such as assigning probability distributions to uncertain parameters, and propagating them through the model, typically by probabilistic sensitivity analysis that samples the distributions and produces a distribution of results. Structural and methodological uncertainties are explored through scenario and structural analyses. The outputs, such as acceptability curves, express the confidence in the results. This process turns the identified uncertainties into a measured account of the reliability of the conclusions.

Source: Briggs, Claxton & Sculpher 2006

Uncertainty quantification is important because decisions based on a model depend on how reliable its results are, and quantifying the uncertainty shows the probability that each option is best and where confidence is limited, preventing a false impression of precision. It also identifies which uncertainties most affect the decision, guiding further research through value-of-information analysis. By measuring rather than ignoring uncertainty, it supports honest, informed decisions and efficient allocation of research effort, making uncertainty quantification a core part of credible economic evaluation.

Source: Drummond et al. 2015

Uncertainty quantification produces a characterisation of the uncertainty in a model's results, such as the distribution of incremental costs and effects, the probability that an intervention is cost-effective at given thresholds through acceptability curves, and an indication of which parameters or assumptions most drive the uncertainty. It may also feed value-of-information measures quantifying the value of reducing uncertainty. These outputs convey the confidence that can be placed in the conclusions and where it is limited, informing both the decision and the case for further research.

Source: Briggs, Claxton & Sculpher 2006

Uncertainty quantification depends on how well the uncertainties are identified and represented, so unrecognised uncertainties, or poorly chosen distributions and scenarios, can leave it incomplete or misleading, and it more readily captures parameter uncertainty than structural or methodological uncertainty, which involve discrete choices. It can be computationally demanding and its outputs may be misread if the assumptions are not understood. These limitations mean uncertainty quantification is conducted carefully, with its assumptions made explicit and its coverage of all relevant sources of uncertainty considered.

Source: Briggs, Claxton & Sculpher 2006