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

A surrogate endpoint S in a clinical trial is an outcome that may be measured earlier or more easily than the true outcome of interest T. In this work, we extend causal inference approaches to validate such a surrogate using potential outcomes. The causal association paradigm assesses the relationship of the treatment effect on the surrogate with the treatment effect on the true endpoint. Using the principal surrogacy criteria, we utilize the joint conditional distribution of the potential outcomes T, given the potential outcomes S. In particular, our setting of interest allows us to assume the surrogate under the placebo, S ( 0 ) , is zero-valued, and we incorporate baseline covariates in the setting of normally distributed endpoints. We develop Bayesian methods to incorporate conditional independence and other modeling assumptions and explore their impact on the assessment of surrogacy. We demonstrate our approach via simulation and data that mimics an ongoing study of a muscular dystrophy gene therapy.

Divining responder populations from survival data

BACKGROUND

Biomarkers that predict treatment response are the foundation of precision medicine in clinical decision-making and have the potential to significantly improve the efficiency of clinical trials. Such biomarkers may be identified before clinical testing but many trials enroll unselected populations. We hypothesized that time-varying treatment effects in unselected trials may result from identifiable responder subpopulations that may have associated biomarkers.

MATERIALS AND METHODS

We first simulated scenarios of clinical trials with biomarker populations of varying prevalence and prognostic and predictive associations to illustrate the impact of subgroup-specific effects on overall population estimates. To show a real-world example of time-dependent treatment effects resulting from a prognostic and predictive biomarker, we re-analyzed data from a published clinical trial (RTOG, Radiation Therapy Oncology Group, 9402). We then demonstrated a quantitative framework to fit survival data from clinical trials using statistical models incorporating known estimates of biomarker prevalence and prognostic value to prioritize predictive biomarker hypotheses.

RESULTS

Our simulation studies demonstrate how biomarker subgroups that are both predictive and prognostic can manifest as time-dependent treatment effects in overall populations. RTOG 9402 provides a representative example where 1p/19q co-deletion and IDH mutation biomarker-specific effects led to time-varying treatment effects and a considerable deviation from proportional hazards in the overall trial population. Finally, using biomarker data from The Cancer Genome Atlas, we were able to generate statistical models that correctly identified and prioritized a commonly used biomarker through retrospective analysis of published clinical trial data.

CONCLUSIONS

Biomarkers that are both predictive and prognostic can result in characteristic changes in survival results. Retrospectively analyzing survival data from clinical trials may highlight potential indications for which an underlying predictive biomarker may be found.