Adaptive dose finding for time-to-event outcomes with adaptive choice of patient number based on response rate
Yevgen Ryeznik (1, 2), Oleksandr Sverdlov (3), Andrew C. Hooker (2)
(1) Department of Mathematics, Uppsala University, (2) Department of Pharmaceutical Biosciences, Uppsala University, (3) EMD Serono
Objectives: Many clinical trials use time-to-event (TTE) outcomes as primary measures of efficacy or safety. D-optimal and adaptive D-optimal designs for dose finding clinical trials with censored TTE outcomes were developed in  as efficient designs in order to determine dose-response relationship. It was shown that in presence of censoring TTE observations  these designs are more efficient than classical uniform (balanced) design. However, the accuracy of the dose-response estimation depends on how much data is censored, i.e. more patients are needed if the response rate is low. In this work we present a stopping criterion to resolve this issue.
Methods: We consider an accelerated failure time (AFT) model  assuming a quadratic dose-response model for log-transformed TTE outcomes with a Weibull distribution that are subject to right censoring with a fixed or random censoring time. For implementing optimal designs in practice a multi-stage adaptive design is applied. We propose a maximum allowed value of the relative standard error (RSE) for all parameter estimates as a stopping criterion for the study. Patients are randomized in cohorts, and after each cohort responds we estimate model parameters and check the RSE of estimated parameters. If they are less than some predefined value, we stop the randomization. Otherwise, we randomize a new, optimized, cohort.
Results: The proposed stopping criteria allows the adaptive choice of number of patients involved in a clinical trial. For high response rates we need fewer patients, while the number increases for a low response rate. The number of patients needed with this adaptive design is much less than required with a standard uniform design. In fact, even with a very large patient numbers, the standard uniform design provides biased estimations of the model parameters. The adaptive designs provide the same variability of estimated parameters as D-optimal (non-adaptive) designs if the number of randomized patients is 25% larger than the corresponding number for the D-optimal design (assuming that the D-optimal design has perfect information about model parameters BEFORE the start of the experiment, an unrealistic situation).
Conclusion: The proposed stopping criteria can improve efficiency of clinical trials with time-to-event outcomes by adapting the number of recruited patients based on response rate.
 Ryeznik Y, Hooker AC, Sverdlov O Adaptive designs for dose finding clinical trials with time-to-event outcomes. PAGE 24 (2015) Abstr 3608 [www.page-meeting.org/?abstract=3608]
 Lawless JF. Statistical Models and Methods for Lifetime Data, 2nd Edition, Wiley, New York (2003).
 Wei LJ. The accelerated failure time model: A useful alternative to the Cox regression model in survival analysis. Statistics in Medicine 11(14-15):1871-1879 (1992).