科研之友
微信
新浪微博
Facebook
分享链接
摘要
Background: To increase power to detect a treatment effect, trials may combine multiple endpoints such as survival, myocardial infarctions, and strokes. When such trials are used to define the uncertainty associated with input parameters in an economic model, the output uncertainty will depend on the way in which the dependency between individual endpoints is modeled. Objective: To develop a flexible approach to model the interrelationship between individual components of a combined endpoint. Methods: A standard independent Dirichlet approach is compared with a dependent Dirichlet approach and logistic approaches. The logistic approaches use a link between the various endpoints by either an observed clinical variable (cholesterol) or a latent one. The logistic and Dirichlet methods are compared using 6 statin trials including 5 endpoints: myocardial infarction (MI), stroke, fatal MI, fatal stroke, and other cardiovascular death. The results are compared using the point estimates and uncertainty in a simplified cardiovascular model to calculate point estimates and uncertainty of estimated incremental life-years when applying probabilistic sensitivity analysis. The influence of the link between endpoints is tested by changing the prior in the logistic approaches. Results: The dependent Dirichlet approach reduces uncertainty up to 29% and changes the point estimate by up to 28%. The logistic approach with uninformative priors does not affect the uncertainty and point estimates. When strong priors are used, the uncertainty margins get smaller (up to 49%) and point estimates vary more. Including information about cholesterol has limited impact. Conclusions: The logistic approaches offer a flexible way to reflect one's beliefs about the interrelationships between individual endpoints, potentially decreasing uncertainty margins. The approach works equally well with and without data concerning the underlying disease process.
- 出版日期2014-4
