Decision modeling to inform decision making: Seeing the wood for the trees

Abstract

We have made 2 recommendations for the conduct of decision modeling to inform decisions by 3rd-party payers who are interested in the health benefits that their health dollars buy. First, modelbased estimates of cost-effectiveness should incorporate costs and benefits over an appropriate time horizon (e.g., lifetime) for all patients included in a model. This requirement is not achieved using the AP modeling approach, which includes only costs and benefits in the 1st year of treatment for incident cases in the final year of the model. There may be information value to 3rd-party payers in presenting cost estimates based on the AP approach, but these estimates should be presented in addition to the IC-based estimates of costeffectiveness. Second, separate cost-effectiveness analyses of relevant identifiable patient groups should be presented, rather than broad analyses that combine such groups, where relevancy refers to the ability of clinicians to differentiate between diagnoses (e.g., stable angina and acute myocardial infarction). There is a risk that actual use of a drug may not be cost-effective, even if it has been demonstrated to be cost-effective in a particular population. Leakage is a well-known problem with the implementation of reimbursement policies,7 but the level of leakage will vary from country to country and with the policies adopted. Analyses of cohorts that include all potentially eligible patient groups assume 100% leakage across all patient groups. This is a worstcase scenario, which may be of interest to decision makers (i.e., if an intervention remains cost-effective in the worst-case leakage scenario, we can have more confidence in our reimbursement decision). However, such confidence is not necessarily required to support a positive reimbursement decision for a patient subgroup.