(191) Statistical Evaluation of Surrogate Endpoints to Optimise Trial Design and Speed Up Decision Making Through the Use of Early Outcomes: A Case Study Using Observational Patient-Level Data

Publication Number: 865

Background: Assessing treatment effects on early clinical outcomes in trials can allow more timely intervention in the management of health conditions. However, focus on these early outcomes results in limited information on late patient-relevant outcomes. As such, there should be rigorous assessment that an early outcome is a surrogate for the late outcome of interest, such that a treatment effect on the early outcome translates to an effect on the late outcome.

Objectives: To provide a framework for the statistical evaluation of an early outcome as a surrogate for a late outcome, using patient-level data from multiple sources.

Methods: A case study will be conducted using long-term data from the EXACOS-CV programme, comprising multi-country observational studies for the association between exacerbations of chronic obstructive pulmonary disease and time to first cardiovascular event. The starting point will be to conduct a literature review and collate expert opinion on the biological plausibility of an association between the early and late outcomes. To obtain evidence for an association, we propose starting with descriptive statistics, followed by regression analyses and assessment of the potential role of confounders. Combining results of multiple studies in a meta-analysis will strengthen the evidence for an association. For surrogacy, an additional step would be to evaluate the relationship between treatment effects on early and late outcomes. Counterfactual-based mediation analysis is described, to quantify (i) the difference in late outcomes between treatment groups had treatment not affected the early outcome and (ii) the change in late outcome associated with a change in the early outcome with treatment.

Results: A framework will be developed considering descriptive analyses, association assessment and causal relationships and their interlink for the statistical evaluation of surrogacy.

Conclusions: Using appropriate descriptive and association analyses provides insight into the expected causal relationship between the early and late outcome. Evaluation of the causal relationship requires assumptions to be made that may only be approximations of the truth. Even when full surrogacy may not be proven, our framework will provide insight into the extent to which a change in the early outcome supports a change in the late outcome. This is very important for regulators, and also for health care professionals and patients who are confronted with randomized clinical trials evaluating treatment effects on surrogates.