Adherence to guidelines for sensitivity analysis: cost-effectiveness analyses of dual oral antiplatelet therapy
Abstract Objective: Cost-effectiveness analyses of new treatments for cardiovascular disease frequently require input parameters whose values are known with uncertainty due to limited data. The objective of this paper is to examine the extent to which published sensitivity analyses addressing this uncertainty adhere to Health Technology Assessment (HTA) guidelines. Research design and methods: A systematic review of published cost-effectiveness analyses was performed for an example drug treatment scenario, dual oral antiplatelet therapy compared with aspirin alone following acute coronary syndromes and/or percutaneous coronary intervention. The following medical literature databases were searched for articles published from January 1997 to June 2007: PubMed, Cochrane Collaboration, EMBASE and the Health Economic Evaluation Database (HEED). Evidence tables were created to show the sensitivity of the cost-effectiveness estimates to changes in the input parameter values, as well as the data sources used for the reference-case and sensitivity analysis input parameter values. The extent to which the sensitivity analyses adhered to HTA guidelines were also examined. Results: Cost-effectiveness ratios were most sensitive to changes in the efficacy of dual antiplatelet therapy and reference-case model assumptions about costs beyond the trial period. Although alternative values tested in the sensitivity analysis for some input parameters were based on observed ranges or distributions, alternative values tested for many other input parameters were assumed without justification. Conclusions: Sensitivity analyses in the cost-effectiveness studies of dual oral antiplatelet therapy were not fully adherent with HTA guidelines. In particular, long-term costs and benefits were not always included in the sensitivity estimates, the impact of differential effects on death and myocardial infarction was not explored, and justification for the alternative parameter values tested was not always provided.