People.csail.mit.edu

A d a p t i v e R a n d o m i z e d S t u d y o f I d a r u b i c i n a n d C y t a r a b i n e
V e r s u s T r o x a c i t a b i n e a n d C y t a r a b i n e V e r s u s T r o x a c i t a b i n e a n d
I d a r u b i c i n i n U n t r e a t e d P a t i e n t s 5 0 Y e a r s o r O l d e r W i t h
A d v e r s e K a r y o t y p e A c u t e M y e l o i d L e u k e m i a
By Francis J. Giles, Hagop M. Kantarjian, Jorge E. Cortes, Guillermo Garcia-Manero, Srdan Verstovsek, Stefan Faderl, Deborah A. Thomas, Alessandra Ferrajoli, Susan O’Brien, Jay K. Wathen, Lian-Chun Xiao, Donald A. Berry, and Elihu H. Estey Purpose: Troxacitabine has activity in refractory my-
TA stopped after 11 patients. Defining success as complete
eloid leukemia, either as a single agent or when combined
remission (CR) that occurred within 49 days of starting
with cytarabine (ara-C) or with idarubicin. A prospective,
treatment, success rates were 55% (10 of 18 patients) with
randomized study was conducted in patients aged 50 years
IA, 27% (three of 11 patients) with TA, and 0% (zero of five
or older with untreated, adverse karyotype, acute myeloid
patients) with TI. Because three CRs occurred after day 49,
leukemia (AML) to assess troxacitabine-based regimes as
final CR rates were 55% (10 of 18 patients) with IA, 45%
induction therapy.
(five of 11 patients) with TA, and 20% (one of five pa-
Patients and Methods: Patients were randomized to re-
tients) with TI. The probability that TA was inferior to IA
ceive idarubicin and ara-C (IA) versus troxacitabine and
was 70%, with a 5% probability that TA would have a
ara-C (TA) versus troxacitabine and idarubicin (TI). A Bayes-
20% higher CR rate than IA. Survival was equivalent with
ian design was used to adaptively randomly assign patients
all three regimens.
to treatment. Thus, although there was initially an equal
Conclusion: Neither troxacitabine combination was su-
chance for randomization to IA, TA, or TI, treatment arms
perior to IA in elderly patients with previously untreated
with a higher success rate progressively received a greater
adverse karyotype AML.
proportion of patients.
J Clin Oncol 21:1722-1727. 2003 by American
Results: Thirty-four patients were treated. Randomiza-
Society of Clinical Oncology.
tion to TI stopped after five patients and randomization to
ALL NUCLEOSIDE analogues currently approved as anti- agents.4Of87patientstreatedinthisstudy,74patientshadAML cancer agents are in the D configuration.1 The discovery or advanced myelodysplastic syndrome (MDS). Of the patients of lamivudine as a potent inhibitor of human immunodeficiency with either AML or MD, 10 patients (13%) achieved CR and virus 1 (HIV-1) reverse transcriptase led to both the acceptance four patients (5%) had hematologic improvement. Six of 39 that unnaturally configured nucleoside analogs could be metab- patients (15%) with refractory AML or MDS who received olized by humans and to the development of L-enantiomers as troxacitabine and ara-C (TA) achieved CR. Two of 18 patients anticancer agents.2,3 Modification of the structure of lamivudine (11%) with refractory AML or MDS who received troxacitabine resulted in the formation of troxacitabine, which has antileuke- and idarubicin (TI) achieved CR. On a recent analysis of mia activity.4-8 In a phase I study of troxacitabine in patients first-line therapies in a cohort of 1,279 patients with AML or with refractory leukemia, three complete remissions (CRs) and advanced MDS treated at the M.D. Anderson Cancer Center one partial remission (13%) were observed in 30 patients with (Houston, TX) between 1991 and 1999, the idarubicin and ara-C acute myeloid leukemia (AML).5 In a subsequent phase II study, (IA) regimen was at least equivalent, if not superior, to eitherfludarabine and ara-C or topotecan and ara-C regimens.9 We thus two CRs and one partial remission (18%) were observed in 16 conducted a prospective, randomized comparison of IA versus patients with refractory AML.6 Idarubicin, topotecan, and cytar- TA versus TI in patients aged 50 years or older with previously abine (ara-C) are often included in combination regimens for untreated AML and an adverse karyotype.
patients with either previously untreated or relapsed myeloidleukemias.9,10 A randomized phase I/II study was conducted to establish doses of troxacitabine given in combination with these Patients aged 50 years or older with minimally pretreated (maximum of 3 From the Department of Leukemia and the Department of Biostatistics, days hydroxyurea and/or leukapheresis) AML were eligible if they had an University of Texas M.D. Anderson Cancer Center, Houston, TX. abnormal karyotype other than inv(16), t(8;21), ϪY, or –X. Patients were Submitted November 4, 2002; accepted February 10, 2003. allowed to be randomly assigned treatment on the study before cytogenetic Address reprint requests to Francis J. Giles, MD, Department of Leuke- results were available if they had a blast count greater than 20 ϫ 109/L, mia, Box 428, University of Texas M.D. Anderson Cancer Center, 1515 diffuse intravascular coagulopathy, or organ failure considered to be related Holcombe Blvd, Houston, TX 77030; email: [email protected]. to AML. Other eligibility criteria included serum bilirubin Յ 2.0 mg/dL; 2003 by American Society of Clinical Oncology. AST or ALT levels less than 3 times the upper limit of normal or less than 5 times the upper limit of normal, if considered the result of leukemia; or Journal of Clinical Oncology, Vol 21, No 9 (May 1), 2003: pp 1722-1727 Information downloaded from jco.ascopubs.org and provided by M D ANDERSON HOSP on March 12, 2009 from Copyright 2003 by the American Society of Clinical Oncology. All rights reserved. Table 1. Operating Characteristics of the Adaptive Randomization Design*
NOTE. Arm 0 ϭ IA, arm 1 ϭ TA, and arm 2 ϭ T1.
Abbreviations: IA, idarubicin and cytarabine; TA, troxacitabine and ara-C; TI, troxacitabine and idarubicin.
*Data are based on 1,000 computer-simulated trials of each scenario.
†P0, P1, and P2 are the true probabilities of response for arms 0, 1, and 2, respectively.
‡P (choose arm 0 superior) refers to the probability of choosing arm 0 as superior, on the basis of the interim analyses and the final analysis, given the indicated values of P0, P1, and P2. The number in parentheses gives the probability of choosing arm 0 as superior and of stopping the trial based only on the interim analyses. The valuesfor P (choose arm 1 superior) and P (choose arm 2 superior) are analogous to those for P (choose arm 0 superior).
§n0, n1, and n2 are the mean sample sizes for arms 0, 1, and 2, respectively.
serum creatinine Յ 1.5 mg/dL. The institutional review board approved the The trial proceeded in the following manner. A maximum of 75 patients protocol, and all patients gave signed informed consent indicating that they were to be randomized. Patients were to be randomly assigned to IA (arm 0), were aware of the investigational nature of this study.
TA (arm 1), or TI (arm 2) with probabilities ␲ , ␲ , and ␲ , respectively.
Initially, ␲ ϭ ␲ ϭ ␲ ϭ 1/3. The probability of random assignment to IA (␲ ) remained as 1/3, as long as all three arms remained in the trial. When each new patient entered the trial, q , which was defined as Pr(m Ͻ m Troxacitabine was supplied (Shire Pharmaceutical Development Ltd, data), where k ϭ 1, 2; and r, defined as Pr(m Ͻ m Laval, Quebec, Canada) in vials containing 10 mg of lyophilized drug. The to evaluate the stopping rules and to adapt the randomization probabilities.
drug was diluted in the vial with 0.9% saline solution to obtain a 2 mg/mL Although all three treatments remained in the trial, the randomization stock solution. To yield the required dose, an appropriate volume of the stock probabilities, ␲ and ␲ , were calculated as ␲ ϭ (2/3)[q 2/(q 2 ϩ q 2)] and solution was further diluted in a polyvinyl chloride infusion bag with 0.9% ␲ ϭ (2/3)[q /(q 2 ϩ q 2)]. If at any time during the trial either q Ͼ 0.85 or saline solution to a total volume of 50 mL, which was administered over 30 q Ͼ 0.85 (ie, the current probability was at least 85% that TA or TI had a minutes. Patients were initially randomly assigned to one of three regimens shorter time to CR than did IA), IA would be dropped from the randomiza- at the following dosages: idarubicin 12 mg/m2 intravenously (IV) daily for 3 tion. If this were to happen and if both investigational arms were still in the days and ara-C 1.5 gm/m2 IV over 2 hours daily for 3 days versus troxacitabine 6 mg/m2 IV daily for 5 days and ara-C 1 gm/m2 IV over 2 hours trial, the randomization probability for arm 1, ␲ r2/[r2 ϩ (1 Ϫ r2)], and daily for 5 days versus troxacitabine 4 mg/m2 IV daily for 5 days and the probability of assignment to arm 2 would become ␲ idarubicin 9 mg/m2 IV daily for 3 days. Patients who achieved CR received 0.15 (ie, TA was being outperformed by IA) or r Ͻ 0.15 (ie, TA the first consolidation course, as per induction therapy, then subsequent was being outperformed by TI), TA would be dropped from randomization.
cycles of the same regimen at reduced doses. Patients received trimethoprim In addition, if q Ͻ 0.15 (TI was being outperformed by IA) or if r Ͼ 0.85 and sulfa, or levofloxacin; fluconazole and itraconazole, or liposomal- (TI was being outperformed by TA), TI would be dropped from randomiza- encapsulated amphotericin; and valacyclovir as antimicrobial prophylaxis.
tion. If at any time during the trial only IA and one investigational arm k Antibacterial and antifungal prophylaxis continued until neutrophil recovery remained, the randomization probability of arm q was set to ␲ ϭ q /[q ϩ was more than 0.5 ϫ 109/L and antiviral prophylaxis continued for 2 weeks (1 Ϫ q )] and the randomization probability for the control was set to ␲ ϭ 1 Ϫ ␲ . Finally, an arm that dropped out could be reopened if information (ie, CR by day 49) became available from patients previously randomly assigned to that arm or if the other arms performed sufficiently poorly,subsequent to closure of the arm in question.
CR was defined as normalization of the blood and bone marrow, Before beginning the study, we used computer simulation to examine the with Յ 5% blasts, normocellular or hypercellular bone marrow, neutro- performance of the above design (its operating characteristics) under various phil count Ն 1 ϫ 109/L, and platelet count Ն 100 ϫ 109/L. Toxicity was scenarios (Table 1). In particular, we were interested in the probability of graded on a scale of 0 to 5, using National Cancer Institute common (correctly) selecting an arm as superior to the other arms if it was truly superior, and conversely, the probability of (incorrectly) selecting an arm thatwas no better than the other arms. For example, assuming that the true probabilities of response with arms 0, 1, and 2 were 0.30, 0.30, and 0.50, Patients were assigned to one of three treatment arms in respectively (Table 1, row 2), the overall probability of (correctly) choosing an adaptive randomized fashion.11 Initially, the randomization was balanced, arm 2 (TI) as superior, on the basis of superiority shown both at interim with a probability of 1 in 3 of random assignment to each of the three arms.
analysis and at the end of the trial, was 0.797. The probability of (incorrectly) As data accrued about efficacy, assignment probabilities shifted in favor of selecting arm 0 (IA) as superior was 0.025, whereas the probability of (incorrectly) selecting arm 1(TA) as superior was 0.178. The probability of The primary efficacy end point (success) was CR without nonhematologic stopping the trial early and declaring arm 2 superior was 0.740, whereas the grade 4 toxicity by 50 days. The comparison of arms in the data analysis and corresponding probabilities for arms 0 and 1 were 0.005 and 0.145, for the adaptive randomization was based on time to success, which we respectively. In this scenario, the expected number of patients to be randomly assumed was exponential, but which was truncated at 50 days. A priori we assigned to arms 0, 1, and 2 were 11, 12, and 17, respectively. This contrasts assumed that the median time to success, m , for each treatment followed an with the numbers of patients (ie, 25, 25, and 25, respectively) that would inverse gamma (2.001, 4.614) distribution.
pertain if no interim analyses had been done. Note in the above scenario that Information downloaded from jco.ascopubs.org and provided by M D ANDERSON HOSP on March 12, 2009 from Copyright 2003 by the American Society of Clinical Oncology. All rights reserved. despite the fact that IA and TA had the same true response rates, the was given TA. At this time, the probability of random assign- probability of incorrectly selecting IA as the best arm is 0.005, whereas the ment to TI became 0.0 (ie, the TI arm dropped out), whereas the probability of incorrectly selecting TA is 0.178. This indicates that the design probability of random assignment to IA became 0.87, and was more protective of the investigational arms than of the standard arm(IA). This can be further appreciated by examining Table 1, row 4. Although probability of random assignment to TA became 0.13. The final all three arms here have the same true probability of response, the probability patient (ie, patient 34) was randomly assigned to treatment on of (incorrectly) selecting arm 2 or 3 as superior is 0.899 (0.449 ϩ 0.450). In November 1, 2001. At this time, success rates were seven of 12 sum, the design reflected our willingness to tolerate a relatively high patients (58%) with IA, three of eight patients (37%) with TA, probability of falsely declaring TA or TI superior (when they were not) to and zero of five patients with TI. Responses remained unknown have a relatively high probability of selecting these arms when they weretruly superior. This reflected the unsatisfactory response rate associated with at days 27, 23, 22, 9, and 3 in patients 28, 29, 30, 31, and 33, the standard IA treatment arm. The methods used for logistic regression and respectively, who were given IA; and at days 39 and 30 in for model criticism (goodness-of-fit analyses) were as previously described.9 patients 26 and 32, respectively, who were given TA. Theprobabilities of random assignment were 0.96 for IA and 0.04 for TA, and the TA arm was dropped. Because success was defined Between the randomization dates of the first and last patients as CR without nonhematologic grade 4 toxicity by 50 days, the (April 3, 2001 and November 1, 2001, respectively), 34 patients final success rates were 10 of 18 patients (55%) with IA, three of were randomly assigned to treatment arms. Two of the 34 11 patients (27%) with TA, and zero of five patients with TI.
patients (9%) had a normal karyotype, but they were randomly Two CRs occurred after day 49 in the TA group (patients 11 assigned to treatment arms because their clinical condition did and 26) and one CR occurred in the TI group (patient 1).
not permit waiting for cytogenetic results to become available Accordingly, the final CR rates were 10 of 18 patients (55%) and because of the probability that, given their ages (61 and 77 with IA, five of 11 patients (45%) with TA, and one of five years), they would have abnormal cytogenetics. The 34 ran- patients with TI. Using a beta distribution with a noninformative domly assigned patients had a median age of 66 years (range, 50 prior (0.5,0.5),13 the probability, given these data, that the CR to 78 years). Twelve patients (35%) had a Zubrod performance rate would be lower with TA than with IA was 70%; the score of 2 or 3 at presentation. Eighteen patients (53%) had probability that the CR rate would be 20% higher with TA than monosomies of chromosomes 5 and/or 7 or deletions of the long with IA was 5%. Corresponding values for TI were 92% and 1%.
arms of these chromosomes (Ϫ5/Ϫ7); four patients had trisomy Among patients achieving CR, recurrence rates by treatment 8, 3, 11q deletions; seven patients had one or two miscellaneous arm were seven of 10 patients (70%) with IA, four of five abnormalities; and two patients were cytogenetically normal, as patients (80%) with TA, and one of one patient (100%) with TI.
noted above. Thus, using the Medical Research Council classi- For IA, times to relapse were 6, 10, 11, 12, 25, 32, and 52 weeks, fication system, 18 patients had a worse prognosis and 16 with remissions ongoing in three patients at 15, 15, and 34 patients had an average prognosis, as determined by karyotype.12 weeks. Corresponding times for TA were 19, 21, 22, and 40 Fifteen patients (44%) had a documented abnormality in blood weeks, with one remission ongoing at 46 weeks; the only patient cell count for at least 1 month before diagnosis of AML achieving CR after TI relapsed 12 weeks later. No patient died in presentation (antecedent hematologic disorder [AHD]), and in 10 CR. Therefore, there was no significant difference among pa- of these patients the duration of AHD exceeded 3 months.
tients receiving IA, TA, and TI in terms of time to treatment Table 2 lists the changes in randomization probabilities as the trial progressed. As noted above, the chance of randomization to A fundamental reason to distinguish between CRs occurring IA remained 0.33 until either the TA or TI arm dropped out. The before the start of therapy and those occurring 49 days after the first patient was randomly assigned to arm TI. The second patient start of therapy is the hypothesis that the latter are essentially presented for random assignment 8 days later, and because the cosmetic (see Discussion). Disease reappeared (at 22 and 40 first patient had yet to achieve CR, there was a trivial increase in weeks from CR date) in both patients who achieved CR after the probability of assignment to TA (0.34) rather than to TI more than 49 days from the start of TA therapy; however; there (0.32), with the probability of assignment to IA remaining at were no differences in time to treatment failure between patients 0.33. The first patient was assigned to TA on June 6, 2001. By given TA who took less than 50 days to achieve CR and patients this time, the success rates were one of two patients with IA and who took more than 50 days to achieve CR. All patients who zero of two patients with TI (CR in patient 1 occurred on day achieved CR with IA therapy did so within 49 days after starting 50), whereas in an additional three patients given IA and in an therapy. The only CR with TI occurred 50 days after beginning additional two patients given TI, responses remained unknown therapy. All of the above results suggest that there are too little data (at days 31, 35, and 45 in the IA group and at days 9 and 21 in to test the hypothesis of cosmetic CR in this study. Death rates were the TI group). This led to probabilities of randomization to IA, 11 of 18 patients (61%) with IA, seven of 11 patients (64%) with TA, and TI of 0.33, 0.42, and 0.24, respectively. When patient 25 TA, and five of five patients (100%) with TI. Time to death was presented for randomization on September 12, 2001, success equivalent in all three regimens. The failure of the higher CR rate rates were five of nine patients (55%) with IA, three of seven with IA to translate into a superior survival, even when compared patients (43%) with TA, and zero patients with TI. Responses with TI, seems attributable to the brevity of the IA-induced CR.
were unknown at days 21 and 12 in patients 23 and 24, The number of patients randomly assigned to treatment was respectively, who were given IA and at day 44 in patient 20, who sufficiently small that imbalances in the distribution of important Information downloaded from jco.ascopubs.org and provided by M D ANDERSON HOSP on March 12, 2009 from Copyright 2003 by the American Society of Clinical Oncology. All rights reserved. Table 2. Application of Adaptive Randomization Design
Abbreviations: IA, idarubicin and cytarabine; TA, troxacitabine and ara-C; TI, troxacitabine and idarubicin; CR, complete response.
*For purposes of random assignment of subsequent patients to treatment, response in these patients was considered to have occurred on †This patient was randomized on 9/12/01 and died 12 days later without having received treatment because of intercurrent problems; for purposes of random assignment of subsequent patients to treatment, this patient was considered to have experienced treatment failureon day 12.
prognostic covariates between treatment arms could have arisen.
AHD) and than the TI group (with respect to AHD). In contrast, Thus, the IA group tended to have somewhat poorer performance performance status (less favorable in IA patients) and age status, but more favorable cytogenetics than the TA or TI groups.
(similar in IA, TA, and TI groups) was not considered for The IA group also had an AHD less frequently and they were inclusion. The fitted model (Table 4) indicates that treatment more frequently treated in HEPA-filtered rooms (Table 3). Given with IA (rather than TA), TA (rather than with TI), and these data, logistic regression was performed to determine cytogenetics (other than the Ϫ5/Ϫ7 or complex) were indepen- whether a treatment effect was present after accounting for dent predictors of CR, but that none achieved statistical signif- covariates not related to treatment. Two considerations moti- vated our approach. First, there were too few patients for which to examine the independent effects of all the covariates shown inTable 3.14,15 Second, as noted in the discussion of the study Induction therapy for AML is unsatisfactory, particularly for design, we wished not to reject TA or TI, even at the expense of elderly patients and/or for those with an adverse karyotype.9 rejecting IA. These desiderata led us to examine the following Troxacitabine is a novel nonnatural nucleoside analog with for inclusion in a logistic model predicting CR (at any time): significant activity as a single agent in patients with refractory treatment arm (IA v TA v TI), cytogenetics (Ϫ5/Ϫ7 or complex AML.5-7 On a phase I/II randomized study, troxacitabine com- v other), and AHD (no v yes). Table 3 shows that the IA group bined with ara-C or idarubicin achieved CR in patients with was better than the TA group (with respect to cytogenetics and refractory AML, including patients who had failed prior high- Information downloaded from jco.ascopubs.org and provided by M D ANDERSON HOSP on March 12, 2009 from Copyright 2003 by the American Society of Clinical Oncology. All rights reserved. Table 3. Distribution of Covariates According to Treatment Arm
Abbreviations: IA, idarubicin and cytarabine; TA, troxacitabine and ara-C; TI, troxacitabine and idarubicin; AHD, antecedent hematologic disorder.
Fig 1. Survival of patients treated with idarubicin and cytarabine (ara-C; IA),
troxacitabine and ara-C (TA), or troxacitabine and idarubicin (TI).
dose ara-C.4 We thus conducted a prospective, randomized studyof these troxacitabine-based regimens versus IA in an elderlycohort of patients with poor prognosis AML. In terms of early therapy are transient.16 Second, we have observed that subse- CR, IA was superior to both troxacitabine combinations. When quent survival in patients who are in CR after one course (but CRs are compared at any time, it seems unlikely that TA would who require Ͼ 49 days to do so) more closely resembles that be superior to IA in these patients. Overall survival with all three seen in patients who live at least 49 days (but never achieve CR) study regimens was equivalent (Fig 1). Randomization in this than that seen in patients who are in CR by day 49 of study was carried out in an adaptive Bayesian fashion.13 This first-induction therapy.17 Thus, CR attained in course 2 or only randomization process was used in an attempt to align two after 49 days of a first-induction course have been cosmetic, somewhat conflicting major issues (ie, the reluctance of inves- motivating the criterion chosen here. It can be contended, tigators to randomly assign patients to standard or control however, that this formulation derives from data in patients regimens that were known to be highly unsatisfactory and the given IA and that it may not be applicable to regimens, including demand for truly randomized studies to generate plausible data) novel agents, such as troxacitabine. As noted above, there is in the conduct of randomized studies in patients with AML.
insufficient information to examine this possibility in this study.
Some elaborations are necessary. First, we addressed the However, when all CRs achieved at any time are included in the possibility that the statistical design prevented us from identify- analysis, it still seems unlikely that TA or TI are superior to IA ing the activity of the TA or TI regimens by computing the in the patient population with AML studied in this protocol. The design’s operating characteristics (Table 1). As noted, the design survival data (Fig 1) lend support to this view.
was intentionally more protective of TA and TI than of IA.
Another difficulty stems from the possibility of imbalances in Furthermore, the design allowed us to reach a conclusion after the distribution of important prognostic covariates (Table 3).
treating 34 patients. Equally important, as a result of the adaptive This difficulty stems from the small number (n ϭ 34) of patients randomization, 18 patients (53%) in the study received the randomized. Even though there was no suggestion that TA or TI seemingly superior IA regimen, whereas only 11 patients (33%) produced higher CR rates than IA, even when the analysis was would have received this regimen if random assignment to done in a manner that might have been expected to favor the treatment had not been done adaptively.
former regimens (Table 4), there still may have been imbalances An important issue is what is meant by superiority. In in latent unobserved covariates. Whether this possibility, which particular, was it reasonable to use CR obtained by day 50 of is inherent in any adaptively randomized design, is sufficient to course 1 as the criterion of success? Our rationale in defining outweigh the medical advantages consequent to the use of success in this way was two-fold. First, it is well known that adaptive randomization may vary depending on circumstances.
most remissions attained only after a second course of induction Finally, it should be emphasized that these data do not address the relative efficacy of troxacitabine-based regimens in other Table 4. Logistic Regression Model for CR
important subsets of patients with either de novo or relapsed AML. In vitro data indicate that troxacitabine has activity against ara-C–resistant tumor cells.18-20 Troxacitabine, either as a single Cytogenetics ϭ Ϫ5/Ϫ7 or complex rather agent or when combined with ara-C, has activity in patients with AML who have failed high-dose ara-C therapy.7 Thus, troxacitabine-based regimens merit further investigation in the relapsed AML setting. However, within the limits dis- Abbreviations: IA, idarubicin and cytarabine; TA, troxacitabine and ara-C; TI, cussed above, IA remains the least unsatisfactory induction troxacitabine and idarubicin; CR, complete response.
regimen we have investigated to date in elderly patients with *A negative value indicates that the covariate has an unfavorable independent effect on the probability of CR; a positive value denotes the converse.
Information downloaded from jco.ascopubs.org and provided by M D ANDERSON HOSP on March 12, 2009 from Copyright 2003 by the American Society of Clinical Oncology. All rights reserved. 1. Tsimberidou AM, Alvarado Y, Giles FJ: Evolving role of ribonucleo- 11. Berry DA, Eick SG: Adaptive assignment versus balanced ran- side reductase inhibitors in hematologic malignancies. Expert Rev Antican- domization in clinical trials: A decision analysis. Stat Med 14:231-246, 2. Balzarini J, Wedgwood O, Kruining J, et al: Anti-HIV and anti-HBV 12. Grimwade D, Walker H, Oliver F, et al: The importance of diagnostic activity and resistance profile of 2Ј, 3Ј-dideoxy-3Ј-thiacytidine (3TC) and its cytogenetics on outcome in AML: Analysis of 1,612 patients entered into the arylphosphoramidate derivative CF 1109. Biochem Biophys Res Commun MRC AML 10 trial—The Medical Research Council Adult and Children’s Leukaemia Working Parties. Blood 92:2322-2333, 1998 3. Grove KL, Guo X, Liu SH, et al: Anticancer activity of beta-L- 13. Berry DA: Statistics: A Bayesian Perspective. Belmont, CA, Wads- dioxolane-cytidine, a novel nucleoside analogue with the unnatural L configuration. Cancer Res 55:3008-3011, 1995 14. Harrell FE Jr, Lee KL, Matchar DB, et al: Regression models for 4. Giles FJ, Faderl S, Thomas DA, et al: A randomized phase I/II study of prognostic prediction: Advantages, problems, and suggested solutions. Can- troxacitabine combined with cytarabine, idarubicin, or topotecan in patients with refractory myeloid leukemias. J Clin Oncol (in press) 15. Harrell FE Jr: Predicting outcomes: Applied survival analysis 5. Giles FJ, Cortes JE, Baker SD, et al: Troxacitabine, a novel dioxolane and logistic regression. Durham, NC, Duke University Medical Center, nucleoside analog, has activity in patients with advanced leukemia. J Clin 16. Keating MJ, Smith TL, Gehan EA, et al: Factors related to length of 6. Giles FJ, Garcia-Manero G, Cortes JE, et al: Phase II study of complete remission in adult acute leukemia. Cancer 45:2017-2029, 1980 troxacitabine, a novel dioxolane nucleoside analog, in patients with refrac- 17. Estey EH, Shen Y, Thall PF: Effect of time to complete remission on tory leukemia. J Clin Oncol 20:656-664, 2002 subsequent survival and disease-free survival time in AML, RAEB-t, and 7. Giles FJ: Troxacitabine-based therapy of refractory leukemia. Expert 18. Gourdeau H, Bibeau L, Ouellet F, et al: Comparative study of a novel 8. Alvarado Y, Kantarjian HM, Cortes JE, et al: Troxacitabine activity in nucleoside analogue (Troxatyl, troxacitabine, BCH-4556) and AraC against extramedullary myeloid leukemia. Hematology 7:179-185, 2002 leukemic human tumor xenografts expressing high or low cytidine deami- 9. Estey EH, Thall PF, Cortes JE, et al: Comparison of idarubicin ϩ nase activity. Cancer Chemother Pharmacol 47:236-240, 2001 ara-C-, fludarabine ϩ ara-C-, and topotecan ϩ ara-C-based regimens in 19. Gourdeau H, Clarke ML, Ouellet F, et al: Mechanisms of uptake treatment of newly diagnosed acute myeloid leukemia, refractory anemia and resistance to troxacitabine, a novel deoxycytidine nucleoside ana- with excess blasts in transformation, or refractory anemia with excess blasts.
logue, in human leukemic and solid tumor cell lines. Cancer Res 10. Beran M, Estey E, O’Brien S, et al: Topotecan and cytarabine is an 20. Galmarini CM, Mackey JR, Dumontet C: Nucleoside analogues: active combination regimen in myelodysplastic syndromes and chronic Mechanisms of drug resistance and reversal strategies. Leukemia 15:875- myelomonocytic leukemia. J Clin Oncol 17:2819-2830, 1999 Information downloaded from jco.ascopubs.org and provided by M D ANDERSON HOSP on March 12, 2009 from Copyright 2003 by the American Society of Clinical Oncology. All rights reserved.

Source: http://people.csail.mit.edu/mrosenblum/Teaching/adaptive_designs_2010/GilesKontarjian.pdf