Assessing uncertainties surrounding combined endpoints for use in economic models

Area of Science:

• Health economics
• Biostatistics
• Clinical trial design

Background:

• Combined endpoints in clinical trials enhance statistical power for treatment effect detection.
• Modeling endpoint dependencies is crucial for accurate uncertainty estimation in economic models.

Purpose of the Study:

• To develop a flexible method for modeling interrelationships between components of combined endpoints.
• To compare different approaches for modeling endpoint dependencies in economic evaluations.

Main Methods:

• Comparison of independent Dirichlet, dependent Dirichlet, and logistic approaches for modeling endpoint relationships.
• Utilized 6 statin trials with 5 cardiovascular endpoints (myocardial infarction, stroke, fatal outcomes, cardiovascular death).
• Assessed impact on point estimates and uncertainty in a cardiovascular economic model via probabilistic sensitivity analysis.

Main Results:

• The dependent Dirichlet approach reduced uncertainty by up to 29% and altered point estimates by up to 28%.
• Logistic approaches with uninformative priors showed minimal impact on uncertainty and point estimates.
• Strong priors in logistic models significantly reduced uncertainty (up to 49%) and increased point estimate variability; cholesterol data had limited influence.

Conclusions:

• Logistic approaches provide a flexible framework to incorporate prior beliefs about endpoint interrelationships, potentially reducing uncertainty.
• These methods are effective whether or not data on underlying disease processes are available.