A method for utilizing co-primary efficacy outcome measures to screen regimens for activity in two-stage Phase II clinical trials

Abstract

Background Most Phase II clinical trials utilize a single primary end point to determine the promise of a regimen for future study. However, many disorders manifest themselves in complex ways. For example, migraine headaches can cause pain, auras, photophobia, and emesis. Investigators may believe that a drug is effective at reducing migraine pain and the severity of emesis during an attack. Nevertheless, they could still be interested in proceeding with the development of the drug if it is effective against only one of these symptoms. Such a study would be a candidate for a clinical trial with co-primary end points. Purpose The purpose of the article is to provide a method for designing a single arm, two-stage clinical trial with dichotomous co-primary end points of efficacy that has the ability to detect activity on either response measure with high probability when the drug is active on one or both measures, while at the same time rejecting the drug with high probability when there is little activity on both dimensions. The design enables early closure for futility and is flexible with regard to attained accrual. Methods The design is proposed in the context of cancer clinical trials with tumor response and progression-free survival (PFS) status after a certain period. Both end points are assumed to be distributed as binomial random variables, and uninteresting probabilities of success are determined from historical controls. Given the necessity of accrual flexibility, exhaustive searching algorithms to find optimum designs do not seem feasible at this time. Instead, critical values are determined for realized sample sizes using specific procedures. Then accrual windows are found to achieve a design’s desired level of significance, probability of early termination (PET), and power. Results The design is illustrated with a clinical trial that examined bevacizumab in patients with recurrent endometrial cancer. This study was negative by tumor response but positive by 6-month PFS. The procedure was compared to modified procedures in the literature, indicating that the method is competitive. Limitations Although the procedure allows investigators to construct designs with desired levels of significance and power, the PET under the null hypothesis is smaller than for single end point studies. Conclusions The impact of adding an additional end point on the sample size is often minimal, but the study gains sensitivity to activity on another dimension of treatment response. The operating characteristics are fairly robust to the level of association between the two end points. Software is available online.