Mobile Menu

Surrogate Endpoint Evaluation Methodology

Robust identification of clinical trial surrogate end points can help to accelerate the development of pharmacotherapies for diseases that are traditionally evaluated using true endpoints associated with prolonged follow-ups. The meta-analysis-based surrogate endpoint evaluation (SEE) integrates data from multiple smaller trials to statistically confirm a surrogate endpoint. To test the applicability of SEE when only one, larger trial is available, this study analysed the cardiovascular survival endpoint from the larger multinational LEADER trial.

Surrogate endpoints

Many trials assess well-established true endpoints such as survival or other hard outcomes, for which a robust and regulatory acceptable evaluation requires the accrual of a pre-defined and sizeable number of events of a usually relatively infrequent occurrence.

Surrogate endpoints are outcomes that represent a proxy for another outcome, and which may help accelerate the evaluation and approval of drugs. The FDA recognises several biomarkers and other measures as surrogate endpoints. While the identification of surrogate endpoints has been attempted with some success for diseases such as diabetes, it is not a straightforward process. Moreover, once a candidate surrogate endpoint has been discovered, the robust and true confirmation of its surrogacy has not been well-established.

Surrogate endpoint evaluation (SEE)

One procedure for confirming surrogate endpoints that is gaining traction and acceptance is the surrogate endpoint evaluation (SEE) methodology. SEE integrates endpoint data to statistically identify a surrogate endpoint as a potentially robust proxy for a true endpoint. A SEE analysis is typically a meta-regression-based analysis in which the effect of the treatment on both the surrogate and true endpoint is assessed for each included trial. This will allow for the evaluation of the trial-level association between the treatment effects on the surrogate endpoint and the true endpoint. However, in many cases where SEE could help to confirm a surrogate endpoint, only a few trials or a single trial will be available. To overcome this, it has been hypothesised that in these circumstances, data from the existing trial could be split into subsets for by a unit of analysis (e.g., country), thereby satisfying the meta-analytic premise of SEE.

Addressing this hypothesis, the researchers used SEE on the LEADER trial data, which is a single global multicentre cardiovascular outcome trial. The researchers tested the applicability of dividing the trial dataset in subgroups by country, trial site or region followed by merging the subgroups with few occurrences of the true outcome to allow for reliable assessment of treatment effects.

Using the LEADER trial to assess surrogate endpoint evaluation

In this study, the LEADER trial data was grouped by country, ensuring at least 30 cardiovascular deaths in each of the nine resulting country groups. In a two-step SEE on the grouped dataset, the researchers first fitted the group-specific Cox proportional hazard models. Following that, on the trial-level, the researchers regressed the estimated hazard ratio of the true endpoints (cardiovascular death: 497 events, or all-cause death: 828 events) on the hazard ratio of the surrogate endpoint (major cardiovascular adverse event [MACE]: 1302 events) and derived the group-specific R2 and its 95% confidence interval.

The researchers found that group-level surrogacy of MACE was supported for cardiovascular death, but not for all causes of death, with R2 values of 0.85 and 0.23, respectively. Sensitivity analyses using different grouping approaches, such as region, verified the robustness of the surrogate endpoints as well as the appropriateness of the data-grouping approaches.


This study derived a specific grouping approach to successfully apply SEE on data from a single trial. This supports the hypothesis that SEE can still be applied to large monolithic outcome trials to identify surrogate endpoints. This is particularly pertinent for drug development where multiple outcome trials are not available, such as for diabetes.

Image credit: drobotdean – FreePik

Share this article