OBJECTIVES: In probabilistic sensitivity analysis (PSA), it is typical to see distributions assigned to most parameters in a model. However, attention is only usually paid to estimating covariance or interactions between only a small number of parameters. This paper explores the impact of interaction assumptions on the outcomes of PSA, and whether including extra detail in models can actually create falsely confident results.

METHODS: An eight-state Markov model was developed, with corresponding input parameters for transition probabilities, costs and utilities for all health states. Alternative approaches to parameter correlation were taken, ranging from zero correlation to extreme cases such as ‘full dependent interaction’. These were applied to a range of different structural assumptions in the model (e.g., rather than a single input parameter for 'monthly cost of health state X', individual parameters were created for 'cost of physician visits', 'cost of tests', 'cost of drugs', 'cost of hospital visits', etc.) and all were varied independently. The impact of all permutations on the shapes of the PSA scatter plot and CEAC was recorded.

RESULTS: The analysis demonstrates that, if a specific input parameter is broken down into several components which are varied independently, then it is likely that the random variation in each parameter will cancel out the effect of the changes in the other parameters, suggesting a false level of certainty in the PSA's results. In the example used, the likelihood of an intervention being cost-effective varied from 53% to 84% depending on the approach to correlation, even when identical input parameters and confidence intervals were used in each case.

CONCLUSIONS: This analysis demonstrates that modellers can, counterintuitively, create false confidence in PSA results by including more parameters or increasing the granularity of other inputs. A number of recommendations are provided for the critical appraisal of probabilistic model outputs.