OBJECTIVES: In health technology assessment (HTA), decision-makers face increased pressures to make earlier decisions. However, in early stages of drug development, data on effectiveness of new health technologies measured by the final outcome is often limited, especially when measuring the effectiveness of new interventions requires long follow-up time. Therefore, shorter-term surrogate endpoints play an important role in HTA. Candidate surrogate endpoints do not always proove to be perfect. However, when more than one of such endpoints exists, they may jointly fully mediate the treatment effect on the final outcome. This study presents methodology for evaluation of multiple surrogate endpoints as predictors of the treatment effect on the final outcome.

METHODS:

Meta-analytical methods for the evaluation of multiple surrogate endpoints are developed. The modelling techniques, developed in Bayesian framework, take into account measurement errors of the treatment effects on all outcomes and the correlations between them. Methods developed are applied to a case study in multiple sclerosis (MS) where  the relapse rate (RR) and the number of active MRI lesions (MRI) are the candidate surrogate endpoints and the final outcome is the disability progression (DP). Surrogate endpoints are evaluated by assessing their predictive value in the cross-validation procedure.

RESULTS: Applying bivariate model showed a good association between effects on RR and DP. Extending to trivariate case to include the effect on MRI increased the precision of the association and reduced the heterogeneity. The cross validation gave better predictions, by reducing the intervals on average by 14%,  when including multiple surrogate endpoints.

CONCLUSIONS: The methods used for combining evidence on multiple surrogate outcomes can lead to more precise predictions of the effect on final outcome. Inclusion of multiple surrogate endpoints may lead to a more substantial gain in precision in other disease areas, hence leading to faster HTA decisions.