Phase II multi-step planning methods in oncology: comparison, recommendations and potential applications

Phase II clinical trials in oncology are commonly used to determine whether a new treatment has a sufficient response rate to be compared with the best standard therapy in phase III. The multi-step planning methods such as Fleming's procedure and the triangular test adapted for phase II trials were elaborated to terminate phase II trials early. We compared these methods considering a fixed number of steps (4 or 5) on their statistical properties: overall type II error and observed type II error, average number of subjects necessary to conclude, probability to conclude at each step and maximum number of subjects necessary to conclude. Fleming's procedure has an actual power similar to the theoretical power for low to high response rate. In contrast, triangular test has an actual power different from the theoretical power in a few situations: it was very high in case of a very low minimum response rate and very low (<20%) in case of a very high minimum response rate (>80%). In these situations, Fleming's procedure and triangular test were not compared and triangular test cannot be recommended. For intermediate response rate, triangular test required a lower average number of subjects to conclude, a larger number of subjects at each step and thus a larger maximum number of subjects. It provided a larger probability to conclude during the first steps. These advantages should be balanced with the risk to have to include a larger number of patients. Multi-step methods, when correctly used are useful for cytotoxic development when response rate is the end point. They can also be used in trials where toxicity is the end point, and could be of great interest for cytostatic development for example with biological surrogate endpoints. An example using real phase II data is also presented.