Designing phase II trials in cancer: a systematic review and guidance

Over recent years there has been the emergence of a wide range of new cancer therapies, with highly variable modes of action. The pathway of development of novel therapeutic agents is well established (Figure 1). It typically takes up to a decade from preclinical development to new drug approval, and is estimated to cost hundreds of millions of dollars, although the average amount is difficult to accurately estimate (Collier, 2009). Only a small number of newly licensed therapeutic agents are related to oncology each year, and cancer drug attrition rates are significantly higher than in other therapeutic areas, with success in 5% of cancer drug developments compared with 20% in cardiovascular drug development, and improvements are urgently required (DiMasi and Grabowski, 2007; Walker and Newell, 2009). Figure 1 Phases of drug development. These depressing statistics have led to increased attention on the drug development process, aiming to identify ways of reducing the attrition rate. Phase II trials act as screening tools to assess whether a treatment has sufficient activity to warrant further investigation in large costly phase III trials. In this respect, the term ‘activity' is used to describe the ability of an investigational treatment to produce an impact on a short-term or intermediate clinical outcome measure; in oncology, this is commonly either response rate or proportion of patients progression-free at a specific time point, but in the era of targeted therapies this may be measured as a change in biomarker level. With increasing costs of clinical trial design and implementation, including the significant costs associated with phase II and phase III studies, it is essential that resources are appropriately streamlined. Phase II studies must be designed, performed and reported to allow accurate interpretation of results and to obtain the best quality data to facilitate and inform unbiased decisions regarding the subsequent development of the drug(s) under study in the phase III setting. It is important for researchers designing phase II trials (clinicians and statisticians alike) to understand the elements of phase II design. Given the number and variety of designs available, it is also essential that the most appropriate design is selected to accommodate the specific requirements of a trial, and to allow the most informative data to be obtained. Several reviews have discussed the current use of statistical designs in published phase II studies in cancer (Mariani and Marubini, 2000; Perrone et al, 2003; Thezenas et al, 2004; Lee and Feng, 2005). The review by Mariani and Marubini (2000) was restricted to articles cited on MEDLINE that were published in 1997, aiming to identify which statistical methodologies were in current use in cancer-specific phase II studies. Of 295 single-arm trials using objective response rate, only 58 (20%) reported an identifiable statistical design. The most common statistical approach was the two-staged hypothesis testing method, with Simon's methods (optimal and minimax) being the preferred methods. Gehan's two-stage design was the most frequently referenced design. The low frequency of an identifiable statistical design was acknowledged as a remarkable weakness of the studies reviewed, as well as the poor quality of reporting of statistical design where it was identifiable, and the low level of compliance with the design in the study conduct and result interpretation. It was concluded that there was a need for better understanding of available statistical designs for phase II studies in cancer, to increase the quality of studies, and also the need for better reporting. These results have been borne out in other reviews. Lee and Feng (2005) published a review of statistical designs utilised between 1986 and 2002 (n=266), and reported that statistical design was only documented in 46% of cases, with only 23% providing sufficient data to characterise the design used. A similar exercise reported in breast cancer phase II studies published between 1995 an 1999 in seven oncology journals (impact factor ⩾2), again found limited reporting of statistical design (51out of 145, 35.5% Perrone et al, 2003). Although guidance exists for specific elements of phase II design, such as when to incorporate randomisation, choice of outcome measures and broad design categories (Mariani and Marubini, 1996; Booth et al, 2008; Seymour et al, 2010), there is little information to assist trialists and clinicians in choosing between specific statistical designs or adaptations/extensions to commonly used designs. None of the available literature provides a contemporary practical guide to phase II design. A systematic literature review was performed of phase II clinical trial methodology to produce a detailed library of available phase II designs applicable to cancer trials. This provided the background information to facilitate the development of a phase II trial guidance manual aiming to generate a structured and systematic process to categorise and incorporate the principles behind the different designs, assisting researchers in their choice of phase II design. We describe the methodology and results of the systematic literature review and categorisation, and briefly introduce the content and layout of the subsequent phase II trial design guidance manual. The manual is freely available to researchers upon request via the University of Leeds Clinical Trials Research Unit (CTRU) website (http://ctru.leeds.ac.uk/phaseII).