PS28: Role of Covariate adjustment in Oncology Trials – Key learnings and opportunities

Over the last few decades of oncology drug development, the field has seen significant innovation in novel therapies that have led to very meaningful improvements in patient outcomes both in terms of disease control and overall survival. However, such improved prognosis has made it more challenging for newer therapies to demonstrate superior outcomes with adequate precision and assurance when compared to the current standard of care. Historically, oncology Phase III trials have primarily focused on the marginal estimand of Progression Free Survival (PFS)/Disease Free Survival (DFS) or Overall Survival (OS) focusing only on the marginal effect of experimental therapy versus control while ignoring other sources of variation based on patient baseline characteristics. The role of such baseline covariates has largely remained limited to the selection of stratification factors during the design stage and subgroup analyses to assess the overall consistency of the treatment benefit. In recent years, covariate adjustment methods have gained increasing prominence as a principled approach to improving statistical efficiency and precision in the estimation of treatment effects, without inflating type I error. This growing interest was further reinforced by the U.S. Food and Drug Administration's May 2023 final guidance, "Adjustment for Covariates in Randomized Clinical Trials for Drugs and Biological Products," which has elevated the visibility and practical relevance of covariate adjustment across drug development programs. The ICH E9 guidance also encourages the identification of covariates expected to have an important influence on primary variables in a pre-specified manner in the statistical analysis to improve precision and compensate for any imbalance between the treatment groups. However, unlike continuous or binary outcomes, survival outcomes pose unique challenges due to non-collapsible, nonlinear estimators. Lu and Tsiatis (2008, Biometrika) first addressed this for unconditional hazard ratios via Cox models and log-rank tests. Ye et al. (2024, Biometrika) extended this framework to stratified randomization and analysis methods, establishing guaranteed efficiency gains and universal applicability. A key feature of these works is the separation of estimand from analysis method, ensuring that covariate adjustment improves efficiency without altering the inferential target. When proportional hazards assumptions fail-common in immunotherapy versus chemotherapy comparisons-restricted mean survival time (RMST) offers a clinically interpretable alternative. Karrison and Kocherginsky (2018, Clinical Trials) showed that adjusting for prognostic covariates improves RMST precision. Li et al. (2023) proposed pseudo-value regression approaches for covariate-adjusted RMST that maintain type I error control while enabling reduced sample sizes. Despite this progress, adoption of covariate adjustment methods in oncology remains uneven. While well established for continuous or binary endpoints, these methods are far less familiar-and less frequently implemented-for survival outcomes. Regulatory guidance on their acceptability for time-to-event endpoints remains limited, and distinct statistical challenges persist: defining appropriate estimands, ensuring adjustment does not alter the inferential target, and addressing implications for interim monitoring under group sequential designs. This session is timely as the oncology community seeks to translate FDA guidance into practice. The session will address both educational and practical aspects of covariate adjustment for time-to-event endpoints in oncology trials. While covariate adjustment is intuitive for continuous outcomes, its application to survival data is less straightforward-it is not performed through familiar models such as Cox regression, and it may be unclear how adjustment can improve efficiency without altering the estimand. Presentations will clarify these conceptual foundations while highlighting recent methodological developments, including covariate-adjusted log-rank–type tests. The session will also address key design considerations specific to group sequential and other adaptive designs common in oncology, covering practical topics such as selection and construction of prognostic covariates, estimation of attainable efficiency gains from historical data, and integration of covariate adjustment with interim monitoring while preserving error control. Industry case studies will illustrate strategic and regulatory considerations, including covariate pre-specification and the impact of adjustment on power, sample size, and decision-making in late-phase trials. The session will feature speakers from academia and industry. In addition, a senior speaker from the FDA will provide broad commentary on the proposed methods and their practical implications, including need for a strong control of type I error in group sequential design, challenges with interpretability and validity of derived covariates, handling of missing data in covariates, and tread off between simplicity of unadjusted approach vs added complexity of covariate adjusted testing strategies specifically developed for statistical inference in time to event analyses. References : FDA (2023). Adjusting for Covariates in Randomized Clinical Trials for Drugs and Biological Products: Guidance for Industry. Karrison T, Kocherginsky M (2018). Restricted mean survival time: Does covariate adjustment improve precision in randomized clinical trials? Clinical Trials, 15(2):178-188. Lu X, Tsiatis AA (2008). Improving the efficiency of the log-rank test using auxiliary covariates. Biometrika, 95(3):679-694. Tsiatis AA, Davidian M, Zhang M, Lu X (2008). Covariate adjustment for two-sample treatment comparisons in randomized clinical trials: A principled yet flexible approach. Statistics in Medicine, 27(23):4658-4677. Ye T, Shao J, Yi Y (2024). Covariate-adjusted log-rank test: Guaranteed efficiency gain and universal applicability. Biometrika, 111(2):691-705. Ye T, Shao J, Yi Y, Zhao Q (2023). Toward better practice of covariate adjustment in analyzing randomized clinical trials. Journal of the American Statistical Association, 118(544):2370-2382.