Dose escalation with overdose control using a quasi-continuous toxicity score in cancer Phase I clinical trials
Introduction
Phase I cancer clinical trials are one of the most important steps in drug development. In these trials, patients are accrued sequentially and treated with a new agent or combination of existing agents. Dose escalation or de-escalation depends on the DLT status of previously treated patients after one cycle of therapy. For cytotoxic agents, it is known that as the dose of the drug increases, its therapeutic effects usually improve, but often at the cost of increased toxicity. Therapeutic effects are usually not assessed within one cycle of therapy. Consequently, the major goal of a cancer Phase I trial is to determine the MTD of a new drug that is associated with an acceptable level of DLT [5].
The majority of cancer Phase I trials treat DLT as a binary indicator of toxicity. Typically, DLT is defined as grade 3 or 4 non-hematologic and grade 4 hematologic toxicity, see the National Cancer Institute (NCI) common toxicity criteria [6] for the definition of the different grades of toxicity. Patients in these trials can experience different types and grades of toxicities varying from 0 for no toxicity to 5 for death but such information is disregarded when the DLT status is treated as a binary indicator of toxicity. For instance, a patient experiencing multiple grade 2 toxicities would be classified the same as a patient exhibiting a single grade 2 toxicity or a patient exhibiting one or more grade 1 toxicities. Such an approach may be inefficient when controlling the safety of the trial and estimating the MTD. Ideally, one would use all grades of toxicities experienced by the patients when allocating doses to future patients. The goal of this paper is to extend a Bayesian adaptive design known as EWOC [7], [8], [9] by accounting for all grades and types of toxicities experienced by the patients during the trial in order to determine the MTD. A normalized equivalent toxicity score system which was developed by Chen et al. [10] will be used to summarize the information from all toxicities experienced by the patients and to estimate dose allocation and the MTD at the end of the trial. The resulting design is termed EWOC-NETS.
The methodology uses the EWOC algorithm where the indicator of DLT is replaced by a quasi-continuous toxicity score in the Bernoulli likelihood function, which results in a quasi-likelihood function, see for example [11], [12]. After specifying a proper prior distribution for the MTD, we estimate the Phase II dose using the median of the quasi-posterior distribution of the MTD. A similar approach has been implemented in the continuous reassessment method (CRM) by Yuan et al. [13] where the toxicity grades were transformed into a numerical score according to hypothetical severity weights elicited by clinicians. This gives us a simple solution to estimate the MTD by taking into account all grades of toxicities exhibited by all patients. A full likelihood approach can be considered where the corresponding quasi-continuous toxicity score is modeled as a parametric or non-parametric distribution.
The manuscript is organized as follows. In Section 2, we give a brief review of EWOC and the definition of NETS and describe the design EWOC-NETS. Performance of this design including its operating characteristics and comparison with EWOC will be presented in Section 3. The performance of EWOC-NETS compared with that of EWOC is further evaluated with data from a real Phase I clinical trial in Section 4. The article ends with discussion and some concluding remarks.
Section snippets
EWOC-NETS design
In this section, we review EWOC, a Bayesian adaptive design for dose finding based on a binary indicator of DLT; and a normalized equivalent toxicity score NETS system introduced by Chen et al. [10] for designing cancer Phase I trials by taking into account all grades of toxicities experienced by the patients in the trial. Then, we describe EWOC-NETS, a new design for dose finding based on EWOC adapted to a quasi-continuous toxicity score NETS.
Simulation studies
In order to evaluate the performance of EWOC-NETS, simulation studies are carried out comparing EWOC-NETS with EWOC under different scenarios. The primary comparison outcomes are the proportion of trials recommending a given dose level as the correct MTD, average trial sample size, therapeutic effect in terms of percent of patients treated at true MTD, overdose control in terms of percent of patients overdosed and percent of patients with DLT.
Illustration of EWOC-NETS with real trial data
In order to further evaluate the usefulness of EWOC-NETS in real Phase I clinical trials, we generated a total of 5000 pseudo-trials using EWOC-NETS and data from study A09712 in Table 6 of Chen et al. [10]. For the purpose of comparison, we also generated 5000 pseudo-trials using EWOC in a similar way with the data. The dose ranges from 0 to 350 mg/kg and a fixed total number of 9 pre-specified dose levels are tested. There were 41 eligible patients enrolled in the trial and the number of
Discussion
The limitation of using binary toxicity has been studied by several publications. In 1992, Gordon et al. [19] first proposed a multi-grade toxicity scheme which uses all toxicity grades instead of dichotomizing the results as grades < 3 or ≥ 3 and produces more powerful analyses. In 2000, Wang et al. [20] differentiated grade 3 and 4 toxicities in an extended CRM by giving more impact to grade 4 toxicities, thus reducing the chance of selecting the higher dose level as MTD. Yuan et al. [13] made
Acknowledgments
Supported in part by NIH/NCI Grants No. 1 P01 CA116676 (Z.C. and M.T.), P30 CA138292-01 (Z.C. and M.T. and J.K.), 5 P50 CA128613 (Z.C. and M.T.), and CTSI Grant UL1RR033176 (M.T) and the National Center for Research Resources, Grant UL1RR033176, and is now at the National Center for Advancing Translational Sciences, Grant UL1TR000124. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH (M.T).
References (27)
- et al.
A novel toxicity scoring system treating toxicity response as a quasi-continuous variable in Phase I clinical trials
Contemp Clin Trials
(2010) - et al.
Optimal Bayesian-feasibile dose escalation for cancer phase I trials
Stat Probab Lett
(1998) - et al.
Unifying CRM and EWOC designs for phase I cancer clinical trials
J Stat Plann Infer
(2009) - et al.
Phase I trial of preoperative hypofractionated intensity-modulated radiotherapy with incorporated boost and oral capecitabine in locally advanced rectal cancer
Int J Radiat Oncol Biol Phys
(2007) - et al.
A Phase I/II trial combining high-dose melphalan and autologous transplant with bortezomib for multiple myeloma: a dose- and schedule-finding study
Clin Cancer Res
(2010) - et al.
A phase 1 dose escalation study of bortezomib combined with rituximab, cyclophosphamide, doxorubicin, modified vincristine, and prednisone for untreated follicular lymphoma and other low-grade B-cell lymphomas
Cancer
(2012) - et al.
Phase I dose escalation, pharmacokinetic and pharmacodynamic study of naptumomab estafenatox alone in patients with advanced cancer and with docetaxel in patients with advanced non-small-cell lung cancer
J Clin Oncol
(2009) - et al.
Individualized patient dosing in phase I clinical trials: the role of EWOC in PNU-214936
J Clin Oncol
(2004) Phase I. studies of chemotherapeutic agents in cancer patients: a review of the designs
J Biopharm Stat
(2006)Common Toxicity Criteria for Adverse Events v3.0 (CTCAE)
(2003)
Cancer Phase I clinical Trials: efficient dose escalation with overdose control
Stat Med
Flexible Bayesian methods for cancer phase I clinical trials. Dose escalation with overdose control
Stat Med
Dose Finding with Escalation with Overdose Control (EWOC) in Cancer Clinical Trials
Stat Sci
Cited by (0)
- 1
Both first authors.