Investigating the impact of oncology phase II trial design parameters on their ability to successfully screen new treatments

Introduction Phase III oncology trials have significantly high attrition rates, where many treatments fail to show efficacy over standard treatments. Design of phase II trials contribute to these inefficiencies and there is much debate regarding optimal phase II design. The effect of the relationship between phase II and III trial endpoints, randomisation, using one-stage or two stages and the operating characteristics of phase II trials in oncology, on the efficiency of the phase II and III process, are all investigated in this thesis.
Methods Evaluation of design parameters was based on simulating multiple phase II and III trials until a successful phase III trial is observed, assuming many treatments are available for testing. Phase II and III trials were conducted assuming the true effect of each treatment was drawn from a standard normal distribution. Phase III trials were assumed to be randomised, with a continuous primary endpoint, 80% power and 5% significance level. Specific design scenarios were considered. The effect of the correlation between the phase II and III trial endpoints was explored analytically, by ranging the variance of the true treatment effect, while randomisation, number of stages and operating characteristics of phase II trials were explored using simulations. The number of phase II and III patients required to lead to the first successful phase III trial was used to measure efficiency of design parameters.
Results For the scenarios considered, the number of patients required to lead to the first successful phase III trial decreased from 3200 to 1000 patients, on average, as the correlation between endpoints increased from 0 to 1. Randomised single-stage phase II trials required 730 patients to lead to the first successful phase III, while Jung’s randomised two-stage design required 554. A’hern’s exact single-arm single-stage design required 463 phase II and III patients while Simon’s single-arm two-stage design required 438. The type I error, α, and power, 1-β, significantly affected the efficiency of phase II trials. Less stringent α= 0.1,0.15 and 0.2 combined with powers 1-β = 0.4, 0.45, 0.5, 0.55, 0.6 yielded 417 phase II and III patients on average, while stringent α= 0.01 or 0.05 combined with any choice of power required 555 phase II and III patients to lead to the first successful phase III trial.
Discussion Understanding the impact of differing design parameters on the efficiency of phase II trials better equips us with the tools needed to improve their design. Based on the scenarios considered Simon’s single-arm two-stage design with a less stringent type I error and small type II error yielded the greatest efficiency in phase II trials.