Menu Back to Poster-Presentations-Details

S109: Evaluation of Time-To-Event-Endpoints in Oncology Biosimilar Trials

Poster Presenter

Jan-Georg Bohlken

Student
University of Bremen
Germany

Objectives

Comparing time-to-event efficacy estimation methods to assess biosimilarity in immuno-oncology trials with a delayed treatment effect and non-proportional hazards based on published trials from two authorized reference treatments.

Method

In my Master's thesis, I simulated equivalence trials with time-to-event outcomes from published data using a piecewise-exponential model. Treatment arms were compared via hazard ratio, median and restricted mean survival, and survival rates. I evaluated the methods regarding power and type-I error.

Results

This simulation study compared a potential biosimilar product based on published phase III trials from authorized immuno-oncology treatments to its reference product. The piecewise exponential model has shown to be a flexible yet simple model to achieve accurate simulations of time-to-event (TTE) data with a delayed treatment effect. With this approach, we could generate data that closely resembles the reference in all published summary measures, including a delayed separation of the survival curves. A null hypothesis of equivalence and the alternative hypothesis of non-similarity were simulated at varying sample sizes (100-500) and follow-up times (12-24 months). Efficacy differences were estimated using TTE summary measures like the hazard ratio, median survival, restricted mean survival (RMST), and survival rates. The similarity was tested by comparing the confidence intervals of the summary measure with equivalence margins derived from estimations of the active treatment effect. The results indicate that medians of overall survival (OS) and progression-free survival (PFS) are not suited to compare efficacy in the considered scenarios. For OS, the median is reached too late (> three years), requiring unreasonable follow-up times. Although PFS medians occur earlier, the relatively small event rates after treatment separation lead to a high median variance, resulting in wide confidence intervals and narrow equivalence margins. Consequently, the power of an equivalence test based on median survival remains below 80% for all considered simulated scenarios. The RMST has demonstrated an accurate assessment of biosimilarity. A desired power (>= 80%) is reached in all PFS scenarios and some OS scenarios, with ratios of RMST achieving higher power than differences when using RMST-Rations to compare PFS, similar sample sizes as typically used in biosimilar trials (<250) reached the desired power at 24 months or longer follow-up periods.

Conclusion

Biosimilars offer affordable alternatives to existing biologic therapies, potentially increasing access to life-saving cancer treatments for a broader range of patients. Equivalence trials aim to demonstrate no meaningful differences between a biosimilar and a reference product, ensuring the safety and efficacy of a new biosimilar. The delayed treatment effect in immune-oncology treatments poses challenges for efficacy estimation based on time-to-event data due to non-proportional hazards. Appropriate statistical methods must be employed to ensure an accurate assessment of similarity. Our findings indicate that median survival for overall survival (OS) and progression-free survival (PFS) is inappropriate for comparing treatment efficacy in biosimilar trials due to late occurrence and high variance. Conversely, the restricted mean survival time (RMST) has shown to be a potential measure for assessing biosimilarity, achieving the desired statistical power in all PFS and select OS scenarios, thus offering a more reliable and interpretable method for biosimilar evaluations. All results are concluded solely following statistical considerations, excluding all clinical aspects that in an applied setting would affect, for example, the size of the equivalence margins. These results might help to further develop biosimilar efficacy assessment strategies. More research needs to be done that uses other TTE models, recruitment patterns, and censoring mechanisms.