|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 26, No 2 (January 10), 2008: pp. 190-195 © 2008 American Society of Clinical Oncology. DOI: 10.1200/JCO.2007.12.7712 Shortening the Timeline of Pediatric Phase I Trials: The Rolling Six Design
From the Division of Clinical Pharmacology and Therapeutics; and Division of Oncology, the Children's Hospital of Philadelphia, Philadelphia, PA Corresponding author: Jeffrey M. Skolnik, MD, the Children's Hospital of Philadelphia, Abramson Research Center 916, 3615 Civic Center Blvd, Philadelphia, PA 19104-4318; e-mail: skolnik{at}email.chop.edu
Purpose To shorten the study conduct timeline of pediatric phase I oncology trials by employing a novel trial design. Methods A comparison of the traditional 3 + 3 patients per cohort, phase I trial design with a novel, rolling six design was performed by using discrete event simulation. The rolling six design allows for accrual of two to six patients concurrently onto a dose level based on the number of patients currently enrolled and evaluable, the number experiencing dose-limiting toxicity (DLT), and the number still at risk of developing a DLT. Clinical trial simulations (n = 1,000) were based on historical data and were performed using SAS 9.1.3 (SAS Institute, Cary, NC). Study timelines and patient numbers were determined for each design, and safety was assessed as a function of the number of DLTs observed. Results In twelve completed historical studies, the median time to study completion was 452 days (range, 220 to 606 days); number of evaluable participants enrolled was 22 (range, 11 to 33), and DLTs occurring per study was three (range, 0 to 5). In 1,000 study simulations, in which the average time to new patient accrual was 10 days, the average ± standard deviation (SD) time to study completion was 294 ± 75 days for the rolling six design versus 350 ± 84 days for the 3 + 3 design, whereas the number of DLTs per study was the same (average ± SD, 3.3 ± 1.1 v 3.2 ± 1.1 for the rolling six and 3 + 3 designs, respectively). Conclusion The rolling six design may significantly decrease the duration of pediatric phase I studies without increasing the risk of toxicity. The design will be tested prospectively in upcoming Children's Oncology Group phase I trials.
The development timeline of new agents for childhood cancer is inherently longer than for adult cancer. Multiple factors contribute to this, including the need to begin pediatric evaluation of new agents usually only after phase I adult trials are completed, the limited number of pediatric patients eligible for study, and the lack of pediatric cancer drug development initiatives undertaken by the pharmaceutical industry.1 Efforts to shorten the overall timeline have increased during the past 5 years and have been impacted most notably by legislative initiatives, including the Best Pharmaceuticals for Children Act. There have not been changes, however, in the timeline associated with actual phase I trial conduct for the last 40 years. Phase I studies in oncology are a critical first step in the evaluation of novel anticancer agents.2 Adult and pediatric phase I studies are similar, in that both determine a recommended phase II dose, often the maximum-tolerated dose (MTD), as their primary objective and both use dose-limiting toxicity (DLT) as the primary end point. Many adult, and almost all pediatric, phase I oncology trials use a modified version of the up-and-down method created in 1948 by Dixon and Mood.3 In the traditional 3 + 3, phase I cancer trial design, a minimum of three participants are studied at each dose level. If none of these three participants experience a DLT, a subsequent three participants are enrolled onto the next highest dose level. If one of three participants at a dose level experiences a DLT, up to three more participants are enrolled. When a DLT is observed in at least two participants in a cohort of three to six, the MTD is exceeded and an additional three participants (up to a total of six) are treated at the next lower dose level. The MTD is defined as the dose level at which none or one of six participants (0% to 17%) experience a DLT, when at least two of three to six participants (33% to 67%) experience a DLT at the next highest dose. In the 3 + 3 design, accrual is suspended after enrollment of each cohort of three patients. When a participant becomes inevaluable for toxicity, most commonly because of early disease progression, the cohort is reopened to a single patient. Key factors that contribute to the overall timeline of a phase I study include delays associated with patient accrual, replacement of inevaluable patients, time to event (ie, DLT), and time associated with data submission and review. One of the primary reasons for the original development of the 3 + 3 design was to limit the number of patients exposed to a potentially toxic or lethal dose of a new drug. Pediatric phase I trials are remarkably safe given the high-risk population being studied. In our prior review,4 we found that, among 1,066 pediatric patients reported from 47 studies, the toxic death rate was 0.5%, which was not different from that observed in adult phase I trials.5 Given the overall safety profile of pediatric phase I trials, the extended periods of time that studies are suspended to accrual, and the observation that the large majority of dose levels are ultimately expanded to accrue six patients,4 we propose a design modification for pediatric phase I studies, termed the rolling six design, that may shorten the timeline of such trials. The rolling six design allows for accrual of two to six patients concurrently onto a dose level. Decisions as to which dose level to enroll a patient are based on the number of patients currently enrolled and evaluable, the number of patients experiencing DLTs, and the number of patients still at risk of developing a DLT at the time of new patient entry. We have compared the performance of the rolling six and 3 + 3 designs by using discrete event simulation.6-8 Our objectives were to define the characteristics of a pediatric phase I design that would reduce the overall time to study completion, to construct a simulation approach to compare the proposed design with the conventional 3 + 3 design, and to evaluate the performance of each design under typical clinical trial conditions.
Historical Phase I Data Retrospective data from a series of 14 completed, Phase I oncology trials conducted by the Children's Oncology Group (COG) Phase I and Pilot Consortium between 2000 and 2006 were used to examine study conduct and timelines. From this series, we selected a subset of completed trials that performed in accordance with the traditional 3 + 3 methodology. Primary end points for analysis included the time to study completion, the number of participants per study, the number of DLTs per study, and the MTD dose level. Secondary end points included the number of participants per dose level, the number of dose levels per study, and the time to completion of the dose level. Outcomes included the occurrence of DLTs, becoming inevaluable for toxicity (IE), or completing a cycle without a DLT or without becoming inevaluable (termed PASS); outcomes and the time of an event were summarized from individual patient event records. Additional information collected included administrative time, which represented time spent per trial not actively recruiting patients because of trial closure or suspension. Descriptive statistics were generated for the historical studies.
Simulating Patient Populations for Evaluation To evaluate the effect of changes in interpatient arrival time and the evaluation period (cycle length) on design comparison, populations of simulated patients were generated for interpatient arrival time distribution means of 5, 10, 20, and 100 days and for evaluation periods of 21, 28, and 35 days. The effect of changing inevaluability rates on study timelines was studied by examining rates that ranged from 0% to 30%. The effect of increased risk of toxicity was evaluated by shifting the starting dose level to a risk of DLT of 30%. To verify that the simulated event probability (IE or DLT) conformed to the design logic, the number of events were summed across trials by using SAS/Proc Freq (SAS Institute Inc), and the distributions of the time to event variables (eg, time to DLT, interpatient arrival time) were verified using SAS/Insight (SAS Institute Inc). The independence of event times and outcomes were also confirmed for each study simulation.
Comparing Study Design Logic
In the rolling six design, up to six patients were concurrently enrolled onto study. Accrual to the study was only suspended when awaiting data from six patients. Decisions as to whether to enroll a new participant onto the current, next highest, or next lowest dose level were made based on available data at the time of new participant enrollment. Dose level assignment was based on the number of participants currently enrolled in the cohort, the number of DLTs observed, and the number of participants at risk for developing a DLT (ie, participants enrolled but who were not yet evaluable for toxicity). For example, when three participants were enrolled onto a dose cohort, if toxicity data was available for all three when the fourth participant entered and there were no DLTs, the dose was escalated and the fourth participant was enrolled to the subsequent dose level. If data was not yet available for one or more of the first three participants, or if one DLT had been observed, the new participant was entered at the same dose level. Lastly, if two or more DLTs had been observed, the dose level was de-escalated. The process was repeated for participants five and six. In place of suspending accrual after every three participants, accrual was only suspended when a cohort of six was filled. When participants were inevaluable for toxicity, they were replaced with the next available participant if escalation or de-escalation rules had not been fulfilled at the time the next available participant enrolled onto the study. Because participants were assigned an interpatient arrival time, the selection of participants from simulated participant cohorts was not linear but rather was chronologic; if the study was not accruing new participants at the time assigned to an individual, that individual was skipped and the next available participant was assessed for arrival time, in accordance with real-life study conduct. The same patients were used for both the rolling six and the 3 + 3 designs in each pair of 1,000 simulated trials (intrasimulation), and separate groups of patients were used for each new simulation (intersimulation).
Historical Phase I Database Of the 14 COG studies completed, 12 were found to provide acceptable study characteristics and timeline data for analysis (Table 2). In two of these 12 studies, analysis was limited to cohorts that followed the traditional 3 + 3 methodology. The total number of patients in all studies was 249. The median number of patients per study was 22 (range, 11 to 33), dose levels evaluated were four (range, 2 to 8), and the dose level at which the MTD was defined was three (range, 1 to 7). Of the 12% of patients (30 of 249) who experienced a DLT, the median time to DLT was 14 days (range, 0 to 40 days). The median time to dose level and study completion was 77 days (range, 33 to 274 days) and 452 days (range, 220 to 606 days), respectively. Of note, studies were suspended to accrual at a median of 185 days (range, 86 to 430 days), which represented more than 50% of total study time.
Simulated Studies: Performance of 3 + 3 Versus Rolling Six Design The rolling six design outperformed the 3 + 3 design for key performance metrics (Table 3). The distribution of elapsed time to complete 1,000 simulated trials, with mean interpatient arrival times of 5, 10, or 20 days, uniformly favored the rolling six design (Fig 1). For an average interpatient arrival time of 10 days—the most commonly observed interpatient arrival time for studies—the mean ± standard deviation (SD) time to complete a pediatric phase I study was 294 ± 75 days for the rolling six design versus 350 ± 84 days for the 3 + 3 design. With the rolling six design, 13%, 52%, 29%, and 2.5% of studies reached the MTD at dose levels 1, 2, 3, or 4, respectively; with the 3 + 3 design, 12%, 52%, 30%, and 3% of studies reached the MTD at those dose levels, respectively. On average, the rolling six design enrolled three more patients per study than the 3 + 3 design (Fig 2) with an overall mean ± SD patient accrual of 20 ± 5 and 17 ± 4 for the rolling six and 3 + 3 designs, respectively. Importantly, there was no difference in the distribution of DLTs between designs (Fig 3); a mean ± SD of 3.3 ± 1.1 participants developed a DLT in the rolling six design, and 3.2 ± 1.1 participants developed a DLT in the 3 + 3 design.
Changes in Simulation Parameters Changes in study parameters, including changes in the interpatient arrival time, the probability of a DLT, and the cycle length, did not change the performance of the rolling six design (Table 3). Changes in the evaluation period from 21 to 28 or 35 days did not change the outcome of simulations; in all instances, the rolling six design outperformed the 3 + 3 design with respect to total study length. When interpatient arrival time was increased to a mean of 100 days, the rolling six design slightly outperformed the 3 + 3 design. The impact of increasing the rate of inevaluability was greater with the 3 + 3 design than with the rolling six design; when the rate increased to 30%, the mean ± SD time to study completion with the 3 + 3 design was 98 ± 85 days longer than with the rolling six design. When the probability of a DLT at the starting dose level increased from 5% to 30%, the rolling six design was favored in terms of elapsed time to complete the trial (mean ± SD, 185 ± 39 v 242 ± 59 days); there was no difference in the incidence of DLT, but, as expected, fewer patients were required per study (mean ± SD, 13 ± 2 v 12 ± 3).
By using discrete event simulation, we found that the rolling six method has the potential to significantly decrease the duration of pediatric phase I studies. There was a small increase in the total number of patients enrolled when using the rolling six design (n = 3), but such an increase is readily acceptable because an experience of 18 to 24 pediatric patients is typical before proceeding to phase II trials. Importantly, there was no clinically significant increase in the incidence of DLTs when using the new design. The primary reason the rolling six design shortens the overall duration of study conduct is that the number of times a study is suspended to accrual is significantly decreased compared with the 3 + 3 method. The rolling six design also decreases the likelihood that a patient who is eligible to enroll onto a pediatric phase I study is unable to do so because of study suspension to accrual. The rolling six design performance is not sensitive to key metrics, such as increases in risk of DLTs or changes in evaluation period. At the lowest extremes of interpatient arrival times, the rolling six continued to outperform the 3 + 3 design; when interpatient arrival time was zero days, the rolling six study was on average 26 days shorter than the corresponding 3 + 3 study (data not shown). In studies with a protracted interval between patient enrollment, which occurs infrequently in pediatric phase I solid tumor studies, the rolling six performs similarly to the 3 + 3 design (Table 3). In adult phase I trials, the most commonly used alternative method to the traditional 3 + 3 design is the continual reassessment method (CRM)10 or its modifications,11,12 which attempt to minimize the number of patients enrolled below a biologically active dose without increasing the number of DLTs. As pediatric trials are conducted only after completion of adult phase I trials, the potential dose range studied is far narrower than in adult trials,4 which minimizes the potential benefit of using a CRM approach to define the pediatric MTD. A fundamental difference between the rolling six and CRM methods is that the former is geared towards shortening the duration of a study and not towards refining the estimate of the MTD. Our study utilized real-world data to generate simulated populations of patients. Unlike past studies in adults, which did not include actual study conduct data,13,14 we derived baseline population metrics (eg, interpatient arrival time, time to an event, risk of a DLT, risk of IE) from both a large set of published Phase I pediatric oncology studies,4 and a newer set of 12 recently completed, pediatric phase I studies. This qualifies our approach and allows a real-time experience with the simulations that few phase I simulations have shared in the past. With an increasing number of new agents entering the clinical pipeline, the creation of a study design that is less sensitive to changes in metrics, such as in interpatient arrival time, probability of DLT, and cycle length, could have a significant impact on pediatric cancer drug development. Future phase II studies are likely to center on randomized phase II selection designs15-17; having several new phase I agents ready for phase II evaluation in shorter time periods will facilitate the design and conduct of these studies. We plan to prospectively evaluate the rolling six design in the upcoming generation of trials conducted by the COG Phase I and Pilot Consortium.
The authors indicated no potential conflicts of interest.
Conception and design: Jeffrey M. Skolnik, Jeffrey S. Barrett, Peter C. Adamson Collection and assembly of data: Jeffrey M. Skolnik, Jeffrey S. Barrett, Bhuvana Jayaraman, Dimple Patel Data analysis and interpretation: Jeffrey M. Skolnik, Jeffrey S. Barrett, Bhuvana Jayaraman, Dimple Patel, Peter C. Adamson Manuscript writing: Jeffrey M. Skolnik, Jeffrey S. Barrett, Peter C. Adamson Final approval of manuscript: Jeffrey M. Skolnik, Jeffrey S. Barrett, Peter C. Adamson
We thank Elizabeth O'Connor of the Children's Oncology Group (Arcadia, CA) for her assistance with data collection for the historical phase I data.
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
1. Adamson PC, Weiner SL, Simone JV, et al: Making Better Drugs for Children with Cancer. Washington, DC, The National Academies Press, 2005 2. Smith M, Bernstein M, Bleyer WA, et al: Conduct of phase I trials in children with cancer. J Clin Oncol 16:966-978, 1998[Abstract] 3. Reiner E, Paoletti X, O'Quigley J: Operating characteristics of the standard phase I clinical trial design. Comput Stat Data Anal 30:303-315, 1999[CrossRef] 4. Lee DP, Skolnik JM, Adamson PC: Pediatric phase I trials in oncology: An analysis of study conduct efficiency. J Clin Oncol 23:8431-8441, 2005 5. Ross L: Phase I research and the meaning of direct benefit. J Pediatr 149:S20-S24, 2006[CrossRef][Medline] 6. Connelly LG, Bair AE: Discrete event simulation of emergency department activity: A platform for system-level operations research. Acad Emerg Med 11:1177-1185, 2004[CrossRef][Medline] 7. Le Lay A, Despiegel N, Francois C, et al: Can discrete event simulation be of use in modelling major depression? Cost Eff Resour Alloc 4:19, 2006[CrossRef][Medline] 8. Shechter SM, Bryce CL, Alagoz O, et al: A clinically based discrete-event simulation of end-stage liver disease and the organ allocation process. Med Decis Making 25:199-209, 2005 9. Barrett JS, Jayaraman B, Patel D, et al: Discrete Event Simulation (DES) as a Technique to Study Decision Rule Efficiency in Event-Driven Clinical Designs. American College of Clinical Pharmacology 46:1092, 2006 10. O'Quigley J, Pepe M, Fisher L: Continual reassessment method: A practical design for Phase I clinical trials in cancer. Biometrics 46:33-48, 1990[CrossRef][Medline] 11. Goodman SN, Zahurak ML, Piantadosi S: Some practical improvements in the continual reassessment method for phase I studies. Stat Med 14:1149-1161, 1995[Medline] 12. Moller S: An extension of the continual reassessment methods using a preliminary up-and-down design in a dose finding study in cancer patients, in order to investigate a greater range of doses. Stat Med 14:911-922; discussion 923, 1995[Medline] 13. He W, Liu J, Binkowitz B, et al: A model-based approach in the estimation of the maximum tolerated dose in phase I cancer clinical trials. Stat Med 25:2027-2042, 2006[CrossRef][Medline] 14. Loke YC, Tan SB, Cai Y, et al: A Bayesian dose finding design for dual endpoint phase I trials. Stat Med 25:3-22, 2006[CrossRef][Medline] 15. Rosner GL, Stadler W, Ratain MJ: Randomized discontinuation design: Application to cytostatic antineoplastic agents. J Clin Oncol 20:4478-4484, 2002 16. Morita S, Sakamoto J: Application of an adaptive design to a randomized phase II selection trial in gastric cancer: A report of the study design. Pharm Stat 5:109-118, 2006[CrossRef][Medline] 17. Estey EH, Thall PF: New designs for phase 2 clinical trials. Blood 102:442-448, 2003 Submitted May 23, 2007; accepted September 27, 2007.
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2008 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
TOP
ABSTRACT
INTRODUCTION
