ABSTRACT
The current clinical pathway for the development and establishment of safety and efficacy for therapeutic agents and regimens, particularly in oncology, has been set since approximately 1960. First, phase I trials of approximately 15-30 patients are used to determine a safe dose for further testing, either the maximal tolerated dose or, in the case of agents with little toxicity (molecularly targeted or biologic agents), a biologically effective dose. These studies are generally done in broadly defined patient populations with sufficiently advanced disease to have few other appropriate treatment options. Then phase II studies of approximately 30-100 patients are used to initially indicate the presence or absence of promising clinical effect, often compared to historical controls. These studies are generally done in more narrowly defined patient populations, for which the agent or regimen is expected to show promise. Finally, randomized phase III trials of hundreds or thousands of patients are done to definitively establish whether or not the agent or regimen is superior to what is currently available for a particular patient population. These studies are generally done in patient populations defined by specific disease, stage of disease, and sometimes other prognostic factors, so that the control treatment is generally uniform. General reviews of the statistical issues involving cancer clinical trials are given, in brief, byRubinstein [1] and, in more detail, by Simon [2]. Very recently, a new form of study, the phase 0 trial, has been added to the beginning of the clinical trials sequence [3]. Phase 0 studies of approximately 15 patients are used to determine whether a particular agent has the anticipated specific biologic effect, which is generally measured by a pharmacodynamic assay. These studies may be done in broadly defined patient populations, with varying disease and disease stage. They are useful in very quickly discovering ineffective agents or in prioritizing among agents designed to affect the same molecular target. The endpoints used in phase III trials of agents or regimens designed to address
serious diseases, particularly in oncology, are almost always time-to-event endpoints.
This is because the outcome of primary interest is usually either overall survival (OS) or progression-free survival (PFS-the minimum of time-to-death or time-to-disease progression). While these endpoints are occasionally measured in binomial fashion, as the percentage of patients who survive, or survive progression-free, at a particular point in time (for example, 1 year or 5 years), this is generally not done, since comparisons by means of the logrank test are much more statistically efficient. In Section 2.3, we discuss, in detail, issues concerning the logrank test, its statistical derivation, and the characteristics of its various uses in clinical trials design, analysis, and monitoring. Those readers who are unfamiliar with the logrank test may wish to read quickly through Section 2.3.1.1 to better understand some of the references to the logrank test in Section 2.2. The endpoint used in phase II trials, in oncology, has been primarily the rate of
objective tumor response across patients, until recently. With the advent of molecularly targeted agents, many of which are anticipated to prolong the time-to-disease progression, rather than cause outright tumor shrinkage, this is beginning to change, and randomized phase II trials using the logrank test to compare the time-to-event endpoint of PFS across patients are beginning to be used. In Section 2.2, we discuss randomized phase II studies using the logrank test, as well as other forms of phase II studies, to put the former in context and make it clear what role they have in the wider phase II setting. We do not discuss phase I studies at all in this chapter, since the endpoints have not been time-to-event variables, nor is there any expectation that they will be in the foreseeable future. Readers who wish to learn more about phase I trials for the purpose of rounding out their understanding of the clinical trials process are referred to the chapters cited above [1,2]. In Section 2.4 we briefly discuss the statistical issues concerning phase 0 trials. It
is likely, in the immediate future, that the endpoints in phase 0 trials will remain as pharmacodynamic assays taken at a particular time, rather than time-to-event endpoints based on a progression of such assays, but this restriction may disappear soon. At any rate, phase 0 trials promise to be an important emerging tool, so we felt it important, in this introductory chapter, to discuss them and place them in the context of the larger clinical trials process. Both authors of this chapter have worked primarily in oncology clinical trials, as
have many biostatisticians involved in the development, refinement, and application of time-to-event statistical methodology. The structure, emphases, and examples in this chapter reflect that. However, we feel that the material certainly applies to clinical trials, in general.
