ABSTRACT
The continual reassessment method (CRM) as originally proposed assumes that the outcome is binary. However, in practice, toxicity and adverse event data are captured across all body systems and symptoms on a grade from 0 to 5, with 0 being no toxicity, 1 a mild toxicity, 2 a moderate toxicity, 3 a severe toxicity, 4 a life-threatening toxicity, and 5 being death. To be able to use the CRM, the multidimensional toxicity data are summarized into a single binary endpoint in the presence or absence of a dose-limiting toxicity (DLT), which is generally defined as the presence of a grade 3 or higher nonhematologic toxicity or the presence of a grade 4 or higher hematologic toxicity in the first cycle of therapy. While using this summary measure from the first cycle alone is simple and advantageous for rapid assignment of doses in sequentially treated patients, and for timely study completion when using conventional methods to determine the maximum tolerated dose (MTD), this approach assumes the absence of late-onset toxicities and ignores cumulative toxicity. Moreover, summarizing a wide range of adverse events into a single binary endpoint, though simple, also comes with weaknesses because it does not differentiate between lower grades of toxicity and between types of toxicity. Thus, a lot of information is potentially lost. Several methods have been proposed for summarizing toxicities into a continuous or ordinal outcome and for estimating the MTD associated with such outcomes. Among the proposed methods that work with a continuous or ordinal endpoint directly, several, such as the latent variable approach by Bekele and Thall [1], have already been described in the previous chapter. In this chapter, we will review the CRM with multiple toxicity constraints in more detail [2, 3] and describe the quasi-continual reassessment method (quasi-CRM) [4, 5] and the unified approach by Ivanova and Kim [6]. While the CRM with multiple constraints defines the MTD as the dose that satisfies various prespecified toxicity thresholds of a continuous or ordinal outcome, the other methods define the MTD as the dose associated with a target level of the continuous or ordinal outcome. Thus, the selection of a target level can be challenging, given that generally little information is known about the summary toxicity measure that would make it possible to specify a relevant target. This is particularly important given that when using these methods, the MTD is defined as the dose associated with the specified target level of the continuous outcome. Each of these methods will be discussed in detail in this chapter. Section 3.2 discusses various methods used for summarizing toxicities into continuous measures and the differences among them. An illustration of a dose-finding clinical trial using the CRM with multiple constraints is presented in Section 3.3. The mathematical framework and the dose-escalation algorithm are presented for the CRM with multiple constraints in Section 3.4, for the quasi-CRM is presented in Section 3.5.1, and for the unified approach by Ivanova and Kim in Section 3.5.2.
