Survival trial design and monitoring using historical controls

Abstract

In this paper, we propose a multistage group sequential procedure to design survival trials using historical controls. The formula for the number of events required for historical control trial designs is derived. Furthermore, a transformed information time is proposed for trial monitoring. An example is given to illustrate the application of the proposed methods to survival trial designs using historical controls.

A two‐stage drop‐the‐losers design for time‐to‐event outcome using a historical control arm

“…Historical control group is mis-specified in the sense that an exponential survival model was used in the one-sample log rank test while data were simulated according to a log-gamma distribution T A B L E 6 Family-wise error rate (FWER) and empirical power (theoretical power in brackets) over the simulated scenarios (S1-S9) of the two stage drop-the-losers (3:1) design with misspecification of historical data according to total sample sizes a (N) from 40 to 200 patients at stage 1 and stage 2 for a nominal FWER of 0.05. Historical control group is mis-specified in the sense that an exponential survival model was used in the one-sample log rank test while survival data of the experimental arms were simulated according to a generalized gamma distribution with a report from Wu et al 23 The more patients were included, the greater the bias: this explains why the FWER and the empirical power increased with sample size. When compared with the correct specification, the greatest differences in FWER and empirical power were observed for large sample sizes and small treatment effects.…”

“… 19 , 20 Several phase II clinical trial designs with an historical control arm for survival outcomes were proposed: the phase II design with Edgeworth expansion of Sun et al, 21 the two-stage or optimal two-stage design by Kwak et al, 22 or the group sequential design of Wu et al. 23 The Kwak’s optimal two-stage design is based on the one-sample log-rank test (OSLRT) 24 , 25 with a procedure restricted to the exponential distribution. Wu et al provided a modified one sample log-rank test (MOSLRT) 23 which allows the use of more flexible survival distributions.…”

“… 23 The Kwak’s optimal two-stage design is based on the one-sample log-rank test (OSLRT) 24 , 25 with a procedure restricted to the exponential distribution. Wu et al provided a modified one sample log-rank test (MOSLRT) 23 which allows the use of more flexible survival distributions. They showed that the MOSLRT was more efficient than the OSLRT and protected well the type I error rate (the OSLRT tended to be conservative even with relatively large sample sizes).…”

“…For patients with a rare life‐threatening disease, the use of a control arm may be unethical or unfeasible 19,20 . Several phase II clinical trial designs with an historical control arm for survival outcomes were proposed: the phase II design with Edgeworth expansion of Sun et al, 21 the two‐stage or optimal two‐stage design by Kwak et al, 22 or the group sequential design of Wu et al 23 . The Kwak's optimal two‐stage design is based on the one‐sample log‐rank test (OSLRT) 24,25 with a procedure restricted to the exponential distribution.…”

Trial Design Challenges and Approaches for Precision Oncology in Rare Tumors: Experiences of the Children’s Oncology Group

“…This data set has a combined sample size of 212 participants, with only 10 who were right censored during follow-up. Using the methodology described in Wu and Xiong 64 and the empirical Kaplan-Meier estimate to model the combined OS distribution (Fig 3B ), a reasonable phase II design will require 45 patients to target an HR of 0.60 with 80% power and maintain less than a 5% type I error rate using a one-sided log-rank test. This HR translates to an approximately 18% net improvement in 1-year OS rate (42% to 60%).…”