OBJECTIVES : In a CE model, ordered parameters which are numerically close often lose their order while running a PSA. This risks misinterpretation of resulting analysis and Incremental cost effectiveness ratios (ICERs) generated may become uninterpretable. To overcome this, difference method has been proposed, but its impact on CE model across several modelling scenarios have not been fully explored.

METHODS : We investigated the impact of difference method on a previously published CE model which assessed treatment efficacy of three arms – treatment1, treatment2 and standard therapy in total hip replacement (THR). PSA was performed using both conventional and difference method for three scenarios where we varied the difference between the utilities of parameters, successful Primary THR (0.85 in all scenarios), successful revision THR (0.60-0.80), and cost of 'success' state only after revision (£0-£5000). Cost of revision THR state (£5294) was assumed to be same in all three scenarios. 10,000 samples were simulated for both the costs and utility parameters for PSA. Relative change in ICER was calculated across all three scenarios for all treatment combinations to determine the impact of difference method versus conventional approach.

RESULTS : The difference in incremental costs and QALY varied between 9% to 22% across all scenarios, but due to incremental QALYs being of small magnitude, the resulting change in ICER was much higher (4% to 332%). Additionally, using conventional PSA method, Treatment2 was favored in deterministic, but not in probabilistic analysis, whereas using the difference method, Treatment2 was favored in both deterministic and probabilistic results.

CONCLUSIONS : In CE models, uncertainty analysis plays an important role in determining the cost-effectiveness of any intervention. However, ordered parameters with small relative differences pose a challenge in such assessments. Difference method approach can improve the robustness of the analysis and help reduce the generation of anomalous ICERs due to conventional PSA methods.