OBJECTIVES: Probabilistic sensitivity analysis (PSA) in health technology assessment involves simulating a large number of realisations as inputs to economic models to appropriately characterise parameter uncertainty and its consequences for decision uncertainty. If two variables are believed to be related in such a manner that one is greater than another then using standard sampling approaches may result in inappropriate PSA. This research aims to propose a method, the ‘Difference Method’ (DM), for generating PSA samples where the constraint that one value is greater than another is maintained and which also satisfies both clinical and statistical validity.

METHODS: The DM approach samples the target variables via a difference parameter. If the target variables are bounded, it involves transforming the variables so that they are unbounded and then sampling via the difference parameter. The DM approach was compared with two commonly applied methods (independent sampling and sampling using a common random number generator) using two examples.

RESULTS: The DM-generated PSA samples have summary statistics that were similar to the given values in our examples whilst maintaining the constraint that one value was greater than another. It also implies plausible correlation between the two target variables. We have developed an Excel workbook to implement the method.

CONCLUSIONS: Failure to account for constraints between parameter values may result in PSA values that do not accurately characterise the uncertainty present in a decision problem. This could result in decisions made on the allocation of scarce health care resources being sub-optimal. The proposed excel-implemented DM approach provides a solution to overcome the problem with naïve sampling methods and should be considered in PSA.