Methods for Evaluation of Surrogate Endpoints for Health Technology Assessment Decision Making: A Good Practices...

Surrogate endpoints are frequently used as primary outcomes in clinical trials. This is appropriate when they are validated for their ability to predict clinical benefit measured on patient-relevant target outcome(s). Such validation is often lacking, thus increasing uncertainty in the decision-making process of regulatory bodies, health technology assessment agencies and payers. This ISPOR Task Force Report provides recommendations on best practices for surrogate endpoint evaluation for health technology assessment decision making. It covers methods that address the 3 levels of evidence for surrogate endpoint validation described in several methodological guidelines: (1) association between treatment effects on the surrogate and the target outcome, (2) association between the surrogate and the target outcome, and (3) biological plausibility. Statistical methods for surrogate endpoint evaluation include meta-analytic approaches using individual participant data or aggregate data. Multivariate meta-analytic models are recommended because they account for the within-study correlation and estimation errors. Issues with limited data and generalizability might be addressed through Bayesian approaches for information sharing from different treatments, treatment classes or indications. Real-world data can complement randomized controlled trial data, especially in rare diseases, but require careful consideration of underlying bias. For plausibility of health economic modeling, the surrogacy analysis and the health economic model should be aligned. The modeled time course of surrogate and target outcomes per treatment arm, as well as the modeled relative effects, should be reported to assess plausibility. Parameter and structural uncertainty in surrogate relationships can be explored through scenario analyses, probabilistic sensitivity analyses, value of information analyses, and threshold analysis techniques.