Assessing surrogacy using restricted mean survival time ratio for overall survival in liver cancer immunotherapy studies.

e16222 Background: Recently, immunotherapy has played a crucial role in treating liver cancer, one of the major cancers that contributes to global cancer burden. Overall survival (OS) is widely applied in cancer trials to evaluate the treatment effects of new therapies. However, it requires more patients and longer follow-up time comparing with progression-free survival (PFS). In addition, while assessing the treatment effects of cancer immunotherapy, proportional hazard (PH) assumption is often violated due to issues such as delayed treatment effects. Restricted Mean Survival Time (RMST) ratio is increasingly used for treatment effect evaluation when the PH assumption is violated. Such change prompts an important question whether the surrogacy value of PFS will be affected when RMST ratio is used to characterize treatment effect. The aim of this study is to examine the feasibility of using PFS as a surrogate endpoint for OS when the treatment effect is measured using hazard ratio (HR) versus RMST ratio. Methods: The surrogacy of PFS on OS was evaluated through examining the association between PFS and OS using HR and RMST ratio. Seven immunotherapy studies published between 2000 and 2021 were included (Table). Information of examined studies such as treatment arms information, OS and PFS were collected. RMST ratio for PFS and OS were calculated based on the Kaplan-Meier plots extracted from each article using WebPlotDigitizer 4.4. The weighted least square regression lines and R^2 between OS and PFS for HR and RMST ratio were calculated. Results: Among 7 immunotherapy studies, 4 gave placebos to the control arms as treatments. 2 provided Sorafenib and 1 assigned the same drug as treatment arm but with different schedule. All 7 studies had OS and PFS as the endpoints. 2 studies violated the PH assumptions. Based on the data extracted from examined articles, a moderate correlation (0.52) between PFS and OS was observed for HR while low correlation (0.10) was observed for RMST ratio. Conclusions: The R^2 values differ greatly depending on whether HR or RMST ratio was used for assessing surrogacy. Our finding may have important implications for the design of future immunotherapy liver cancer trials. For future work, increasing the number of included studies for a more comprehensive analysis is needed. Moreover, trial-level surrogacy analysis should be conducted to complement the study-level investigation.[Table: see text]

Duke Scholars

Author Xiaofei Wang Biostatistics & Bioinformatics, Division of Biostatistics