Probabilistic sensitivity analysis (PSA) is a method to account for uncertainty in cost-effectiveness analysis. The degree of correlation between input parameters is not well reported and is often overlooked in PSA. This means PSA results could be mis-estimating uncertainty. This study aimed to develop a simple model to explore the impact of input correlation on the incremental cost-effectiveness ratio (ICER) and the reported likelihood of cost-effectiveness.
A Markov model was developed with 3 different approaches to correlation: no correlation, partial correlation, and perfect correlation. A hypothetical case study was used to explore the impact of each correlation option on the intervention’s likelihood of cost-effectiveness. Scenario analyses were also used to investigate whether the findings were consistent across different scenarios.
The ICER was comparable across the correlation options. In all scenarios, the no-correlation option had the most certain decision outcomes, and the perfect-correlation option had the least certain likelihood. The proximity of the ICER to the willingness-to-pay threshold influenced the impact of correlation on the PSA results.
This study suggests that the approach toward modeling parameter correlation in PSA has a substantial impact on the level of certainty in model outputs. By ignoring this, the level of certainty of cost-effectiveness could be over or underestimated. Therefore, researchers and decision makers should be careful to consider the potential impact of inter-parameter correlation.