Parameter Uncertainty

Uncertainty arising from imperfect knowledge of the true value of an input parameter, such as a transition probability or unit cost.

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

Parameter uncertainty stems from imperfect knowledge of a true underlying value, such as a transition probability or unit cost, which is fixed but not precisely known. Because it reflects a gap in knowledge rather than genuine variation, gathering more or better data narrows it, and with enough evidence it could in principle be removed. This distinguishes it from variability between individuals, which more data describes but cannot eliminate. Its reducibility is what makes further research potentially worthwhile. Briggs and colleagues (2006) draw this distinction.

Source: Briggs et al. 2006

Parameter uncertainty is represented by assigning each uncertain parameter a probability distribution that reflects the range and likelihood of its plausible true values, given the evidence, and then sampling from these distributions in probabilistic sensitivity analysis. Running the model over many samples produces a distribution of results conveying how parameter uncertainty propagates to the output. The distributions are chosen to suit each parameter's nature and set from its data, so that the represented uncertainty reflects the precision with which each parameter is known.

Source: Briggs, Claxton & Sculpher 2006

Parameter uncertainty, imperfect knowledge of the true parameter values, differs from first-order or individual-level variability, which is the random variation between individuals present even if parameters were known, and from methodological or structural uncertainty, which concerns choices of methods and model structure. Parameter uncertainty can be reduced by more data and is represented by distributions on the parameters, whereas individual variability averages out for a population decision and methodological uncertainty is explored by scenario analysis. Distinguishing these clarifies what drives decision uncertainty and how each is handled.

Source: Drummond et al. 2015

Parameter uncertainty matters because it determines how confident one can be in a model's results and hence in the decision, since imprecise parameter values lead to uncertain outputs that may span different conclusions. Characterising it through probabilistic analysis shows the probability that an intervention is cost-effective and where the uncertainty is greatest, informing both the decision and whether further research is worthwhile. Ignoring parameter uncertainty would give a false sense of precision, so representing it is central to a credible economic evaluation.

Source: Briggs, Claxton & Sculpher 2006

Parameter uncertainty can be reduced by collecting more or better evidence on the uncertain parameters, since it stems from limited data, so larger studies or additional research narrow the distributions around the true values. Value-of-information analysis helps identify which parameters are worth researching further, by valuing the reduction in uncertainty in terms of better decisions. Focusing data collection on the influential, uncertain parameters is the efficient way to reduce parameter uncertainty, in contrast to individual variability, which reflects genuine variation and is not reduced by more data.

Source: Briggs, Claxton & Sculpher 2006