|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Originally published as JCO Early Release 10.1200/JCO.2005.02.913 on June 27 2005 © 2005 American Society of Clinical Oncology.
Enrichment Designs: Efficiency in Development of Cancer TreatmentsCenter for Drug Evaluation and Research, US Food and Drug Administration, Rockville, MD The analysis of an enrichment study design by Friedlin and Simon1 in this issue of the Journal of Clinical Oncology addresses the critical drug development issue of efficiency. It has long been recognized that it is possible to identify patient subpopulations selected from the general population in which it is more likely to show a drug benefit than it would be in an unselected population. The selection of such populations has been called enrichment2 and ranges from practical, variance-decreasing maneuvers (choosing people likely to stay in study and follow directions, people whose blood pressure remains elevated after a period of observation, or people with stable exercise tests); to pathophysiologic selection of people with potentially responsive disease (breast cancer patient with Her2-positive tumors to test trastuzumab); to selection of people likely to have an event that is being measured (people with a past history of an acute infarction, very high cholesterol, or elevated C-reactive protein to test a treatment to prevent reinfarction, because they are more likely to have another infarction; people with rapidly increasing prostate-specific antigen, who are more likely to die of prostate cancer, for a survival trial of a prostate cancer adjuvant treatment3). All these selection maneuvers are reasonably well established and not controversial, except insofar as the generalizability of the study results to the general population may arise.
A less common enrichment design is to use early response to the test drug to choose the people who will be randomly assigned into a long-term study. A classic example of this design was the Cardiac Arrhythmia Suppression Trial (CAST),4 which randomly assigned postinfarction patients to a long-term survival study of encainide/flecainide compared with placebo only if they had Friedlin and Simon1 explore a study design for cytostatic drugs that is similar in concept to that used in the CAST, although it is different in detail because of different circumstances. Although the CAST investigators could simply look for the desired short-term VPB response, withdraw the drug, and randomly assign the responders to the long-term treatment, it would not be satisfactory in patients with malignancy to withdraw everyone from treatment after a 16-week period to examine response, await some return of tumor growth, and then randomly assign the patients to drug or placebo. Instead, they examined a randomized withdrawal design in which after a 16-week treatment period, some of the treated patients (the ones doing reasonably well) would be randomly assigned to continued treatment or receive placebo, and considered how this approach compared with ordinary upfront randomization with respect to efficiency (sample size needed to attain adequate power) in demonstrating an effect on progression rates. In addition to efficiency, of course, the design is attractive to patients and physicians because everyone has a chance to receive the new treatment. Not surprisingly, but now shown quantitatively, the efficiency of the randomized withdrawal design depends on the behavior of the tumor. If the main determinant of response is the inherent growth rate of the tumor (they respond if their growth rate is below some threshold), upfront randomization is generally more efficient. But if there is a responder subpopulation (independent of growth rate), and patients chosen for randomization are those with stable disease (< 10% growth in tumor diameter at 16 weeks), the randomized withdrawal design is considerably more efficient, especially when the responder population is relatively low (such as < 30%). As Friedlin and Simon point out, development of cytostatic agents cannot in most cases rely on the single-arm phase II trials usually used for cytotoxic drugs, as response rates may be very modest and may understate the value of the drug. The randomized trials needed to evaluate these agents can be daunting, and demonstration of effects on progression-free survival or survival will be difficult to show in an unselected population if only a small fraction of the population responds to the drug. Where identification of responders is possible, this problem is greatly reduced, as even preliminary data on erlotinib reveal.5 In an unselected population of 731 patients with non–small-cell lung cancer randomly assigned 2:1 to erlotinib and placebo, a survival effect (median difference, 2 months) was shown. A retrospective analysis of 238 patients whose epidermal growth factor receptor (EGFR) expression status was known, however, showed a large difference between EGFR-positive and EGFR-negative patients. In the former (n = 127), there was a highly significant 7-month survival difference (hazard ratio, 0.65), with little if any effect seen in the 110-patient EGFR-negative group (hazard ratio, 1.01), although there were too few patients in the group to rule out a possible modest effect. It may be that genomics and target receptor biology will rapidly establish responder populations for many or most treatments, but while we await this, it is still necessary to consider cases where they have not done so and design efficient trials for those cases. Friedlin and Simon have shown, for what is surely a particularly difficult case where responders cannot be identified until 4 months have elapsed, that an enrichment approach can be useful. Earlier response assessment, such as an early assessment of effect on nutrient uptake by a tumor or on a circulating tumor marker, might be more efficient still, allowing rapid screening of the population to be randomly assigned. Whether accomplished by sophisticated genetic or receptor analyses or by empirical observations of response to treatment, there is growing recognition that people are not all the same in the way they respond to treatment and that groups that might respond differently should be studied, a change from the established wisdom of conducting trials with broad entry criteria while eschewing subset analyses.6 This is becoming obvious in oncology but is rapidly becoming recognized in cardiovascular medicine, where a variety of circulating proteins are being used to predict outcomes7 and, conceivably, may one day predict responses as well. Author's Disclosures of Potential Conflicts of Interest The author indicated no potential conflicts of interest. REFERENCES
1. Freidlin B, Simon R: Evaluation of randomized discontinuation design. J Clin Oncol 23: 5094-5098, 2005 2. Temple RJ: Special study designs: Early escape, enrichment, studies in non-responders. Commun Stat Theory Meth 23: 499-531, 1994
3. Amico AV, Chen MH, Roehl KA, et al: Preoperative PSA velocity and the risk of death from prostate cancer after radical prostatectomy. N Engl J Med 351: 125-135, 2004 4. Echt DS, Liebson PR, Mitchell LB, et al: Mortality and morbidity in patients receiving encainide, flecainide, or placebo. N Engl J Med 324: 781-788, 1991[Abstract] 5. Tarceva (erlotinib) Tablets Package Insert. Melville, NY, OSI Pharmaceuticals Inc, December 2004 6. Simon R: An agenda for clinical trials: Clinical trials in the genomic era. Clinical Trials 1: 468-470, 2004
7. Heeschen C, Dimmeler S, Fichtlscherer S, et al: Prognostic value of placental growth factor in patients with acute chest pain. JAMA 291: 435-441, 2004
Related Article
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2005 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
70% reduction of ventricular premature beat (VPB) rates during a prerandomization trial period. The rationale for this design was that if a drug had no effect on VPB rate, the postinfarction signal of risk, it could not affect survival. Despite the imaginative design, the result of CAST greatly enhanced nervousness about reliance on surrogate end points in cardiovascular medicine, as it showed a substantially increased mortality in the treated patients, despite their excellent VPB response. 
