Phase II clinical trials - traditional approaches

Hello my name is Fei Ye, I'm a faculty member from Vanderbilt University Medical Center. This is an introductory lecture on phase II clinical trial designs. In this lecture I will introduce some traditional methods which are based on Frequentist approach in designing a phase II clinical trial.

Suppose we have already established a reasonably safe targeted dose or a series of doses from phase I trials. Then the next step is to see if the hazard treatment has any efficacy or biological activity. The overall objective of a phase II clinical trial is to evaluate both efficacy and safety. We would like not only to continue to monitor safety for things like toxicity, adverse effects and other risks as we usually do in phase I trials, but also to find out how well the treatment actually works. Phase II trials are often single arm but it can also be controlled with a standard treatment arm or a placebo arm. Phase II trials can be single stage or have multiple stages and the sample size typically ranges between 20 and 200.

A phase II trial, can be designed using a Frequentist approach or a Bayesian approach. Until today, most of the phase II trials that have been proved and conducted used a Frequentist method. Although in recent years the number of Bayesian trials has been increasing very rapidly. I'll talk about Bayesian trials a bit later and for now let's focus on Frequentist designs. Frequentist designs can be further categorized to non-adaptive and adaptive trials. In each category, we have a number of designs that I would like to cover today. They differ either in number of stage, number of outcomes, number of arms, et cetera. The goal of phase II trial from a statistical point of view is to estimate the treatment effect, in other words, the efficacy of the treatment. We not only calculate the point estimate with the Frequentist approach, we also provide the level of certainty about that point estimate by constructing confidence intervals usually. For example, we can calculate the proportion of patients who had either a complete response or partial response to be, say 35 percent and then we can calculate the 95 percent confidence interval to be, say, from 15 percent to 53 percent. So, basically we are saying that we are 95 percent confident that the true response rate is between 15 percent and 53 percent. Now, that's quite a wide confidence interval, this will be very different from observing the same response rate of 35 percent but with a much narrower confidence interval say, from 33 percent to 38 percent. With the latter we are much more certain about our estimation of the response rate which is again 35 percent and that level of precision depends on a number of things. First one's, sample size, second is, level of significance or type one error rate and also type two error rate or power and also the amount of variation in the data. We always need to plan for the sample size that is, a sufficient minimum sample size required in a Frequentist trial, at the planning stage of a trial. Even for adaptive trials where we allow sample size adjustments, we're still needed to pre-specify adaptation rules at the planning stage. Now, the sufficient minimum samples has again depends on four things: type one error rate, power, the amount of the variation in the data or standard deviation or variance, and the minimum effect size that you would like to detect, and that is very clinically meaningful to you. And as we know the smaller type one error rate you have, the more patients you need to enroll. The greater power you want your study to achieve, the larger sample size you need and the larger amount of variation in the data, in other words, more noise in your data then the more patients you would also need to recruit. And the last thing is the smaller difference you want to detect, you will also need a larger sample size. And all the above is that when everything else is being fixed of course.