Abstract:Motivated by increasing pressure for decision makers to shorten the time required to evaluate the efficacy of a treatment such that treatments deemed safe and effective can be made publicly available, there has been substantial recent interest in using an earlier or easier to measure surrogate marker, $S$, in place of the primary outcome, $Y$. To validate the utility of a surrogate marker in these settings, a commonly advocated measure is the proportion of treatment effect on the primary outcome that is explained by the treatment effect on the surrogate marker (PTE). Model based and model free estimators for PTE have also been developed. While this measure is very intuitive, it does not directly address the important questions of how $S$ can be used to make inference of the unavailable $Y$ in the next phase clinical trials. In this paper, to optimally use the information of surrogate S, we provide a framework for deriving an optimal transformation of $S$, $g_{opt}(S)$, such that the treatment effect on $g_{opt}(S)$ maximally approximates the treatment effect on $Y$ in a certain sense. Based on the optimally transformed surrogate, $g_{opt}(S)$, we propose a new measure to quantify surrogacy, the relative power (RP), and demonstrate how RP can be used to make decisions with $S$ instead of $Y$ for next phase trials. We propose nonparametric estimation procedures, derive asymptotic properties, and compare the RP measure with the PTE measure. Finite sample performance of our estimators is assessed via a simulation study. We illustrate our proposed procedures using an application to the Diabetes Prevention Program (DPP) clinical trial to evaluate the utility of hemoglobin A1c and fasting plasma glucose as surrogate markers for diabetes.

Sufficient conditions for concluding surrogacy based on observed data.

Criteria for surrogate end points

CRITERIA FOR MULTIPLE SURROGATES

Surrogacy Beyond Prognosis: The Importance of “Trial-Level” Surrogacy

Quantifying proportion of treatment effect by surrogate endpoint under heterogeneity

On the role of surrogates in the efficient estimation of treatment effects with limited outcome data

Estimating Treatment Effects using Multiple Surrogates: The Role of the Surrogate Score and the Surrogate Index

A new proportion measure of the treatment effect captured by candidate surrogate endpoints

Incorporating baseline covariates to validate surrogate endpoints with a constant biomarker under control arm

Novel Criteria to Exclude the Surrogate Paradox and Their Optimalities

Criteria for surrogate end points based on causal distributions

Continuous Treatment Effects with Surrogate Outcomes

Assessing heterogeneity in surrogacy using censored data

Propensity score calibration in the absence of surrogacy

Flexible evaluation of surrogacy in Bayesian adaptive platform studies

Using a Surrogate with Heterogeneous Utility to Test for a Treatment Effect

Long-Term Effect Estimation with Surrogate Representation

Asymptotically Pivotal Statistic for Surrogate Testing with Extended Hypothesis

Evaluating Surrogate Marker Information using Censored Data

Towards Optimal Use of Surrogate Markers to Improve Power

On the Evaluation of Surrogate Markers in Real World Data Settings