Prognostic Index for Adult Patients With Acute Myeloid Leukemia in First Relapse
http://www.100md.com
《临床肿瘤学》
the Dutch-Belgian Hemato-Oncology Cooperative Group (HOVON)
Swiss Group for Clinical Cancer Research Collaborative Group (SAKK)
Department of Hematology and HOVON Data Center, Rotterdam
Department of Hematology, Free University Medical Center Department of Hematology, Academic Medical Center, Amsterdam
Department of Hematology, University Medical Center, Utrecht
Department of Hematology, University Hospital, Groningen, the Netherlands
Department of Hematology, Hospital Gasthuisberg, Leuven, Belgium
Department of Internal Medicine, University Hospital, Zürich, Switzerland
ABSTRACT
PURPOSE: The treatment of acute myeloid leukemia (AML) in first relapse is associated with unsatisfactory rates of complete responses that usually are short lived. Therefore, a clinically useful prognostic index can facilitate therapeutic decision making and evaluation of investigational treatment strategies at relapse of AML.
PATIENTS AND METHODS: A prognostic score is presented based on the multivariate analysis of 667 AML patients in first relapse among 1,540 newly diagnosed non-M3 AML patients (age 15 to 60 years) entered onto three successive Dutch-Belgian Hemato-Oncology Cooperative Group and the Swiss Group for Clinical Cancer Research Collaborative Group trials.
RESULTS: Four clinically relevant parameters are included in this index (ie, length of relapse-free interval after first complete remission, cytogenetics at diagnosis, age at relapse, and whether previous stem-cell transplantation was performed). Using this stratification system, three risk groups were defined: a favorable prognostic group A (overall survival [OS] of 70% at 1 year and 46% at 5 years), an intermediate-risk group B (OS of 49% at 1 year and 18% at 5 years), and a poor-risk group C (OS of 16% at 1 year and 4% at 5 years).
CONCLUSION: The prognostic index estimates the outcome of AML patients in first relapse using four commonly applied clinical parameters and might identify patients who are candidates for salvage and investigational therapy.
INTRODUCTION
Although the outcome of patients with acute myeloid leukemia (AML) has improved because of cytarabine- and anthracycline-based chemotherapy in combination with advanced supportive care and introduction of hematopoietic stem-cell transplantation (SCT), relapse continues to represent the leading cause of death in the majority of patients.1,2 The probability of relapse depends on risk factors such as age, pretreatment cytogenetics, and number of cycles of induction chemotherapy required for attaining the first complete hematologic response (CR).3-6 Insight into the individual role of each of these factors with respect to prognosis may support application of risk-adapted postinduction treatment.4
Treatment of AML in first relapse is associated with relatively low response rates.7 Whenever second CR is attained, the median duration of the second relapse-free interval (RFI) is generally considerably shorter than that of the first RFI.8 Because only a minority of patients who experience relapse will derive durable benefit from current reinduction therapy, it would be practically useful to be able to estimate prognosis at the time of relapse. This insight could then facilitate therapeutic decision making at this stage of the disease and guide individualized and investigational treatment strategies.
Factors predicting outcome of patients with AML in first relapse have been reported, and include RFI, age, and cytogenetics.9-15 However, proposed stratification methods for selecting therapies for patients with relapsed AML have only been based on the duration of RFI,9,10,12,16,17 thus neglecting the influence of other known prognostic factors. Predictive scores for patients with AML in first relapse that include more covariates could be generally applicable, if they would furnish a simple and statistically valid prognostic index. We used the results of an analysis of the outcome of 667 unselected patients with first relapsed non-M3 AML among patients who were enrolled onto three successive Dutch-Belgian Hemato-Oncology Cooperative Group and the Swiss Group for Clinical Cancer Research Collaborative Group phase III trials in adults (age 15 to 60 years) with newly diagnosed AML during the period 1987 to 2001,18,19 and considered four clinically relevant parameters (ie, RFI, age, cytogenetics, and previous transplantation).
PATIENTS AND METHODS
Patients
A total of 1,540 non-M3 AML patients were enrolled during the period 1987 to 2001 onto three successive Dutch-Belgian Hemato-Oncology Cooperative Group and Swiss Group for Clinical Cancer Research phase III trials (AML4, AML4a, AML29).18,19 The studies were for eligible patients (age 15 to 60 years) with newly diagnosed AML. Of these 1,540 patients, 1,269 had attained CR, of whom 444 are alive in CR. Among the complete responders, 158 had died from nonleukemic causes and 667 had recurrence of AML. This report is based on analysis of the cohort of 667 patients with AML in first relapse with a median follow-up from relapse of 56 months (Table 1).
The studies were approved by the ethics committees of participating institutions and were conducted in accordance with the Declaration of Helsinki. All participants gave their informed consent.
Treatment Protocols
Treatment in the AML4/4a and AML29 studies involved one cycle of induction with an anthracycline (daunorubicin or idarubicin) in combination with cytarabine (200 mg/m2 for 7 days) and a second cycle of amsacrine with intermediate-dose cytarabine (1,000 mg/m2 every 12 hours for 6 days). If CR occurred, patients in the AML4/4a protocol were randomly assigned after a third cycle of chemotherapy (mitoxantrone and etoposide) to treatment with high-dose busulfan and cyclophosphamide followed by autologous SCT or no additional treatment. In the AML29 study, patients in CR after two cycles of chemotherapy were randomly assigned to treatment with a third cycle of chemotherapy (mitoxantrone and etoposide) without SCT or high-dose chemotherapy with busulfan and cyclophosphamide followed by autologous SCT. When an HLA-identical sibling donor was available, eligible patients proceeded to allogeneic SCT. In the AML29 trial protocol, patients with inv(16), t(16;16), or t(8;21) and WBC count of less than 20 x 109/L at diagnosis were no longer treated with autologous or allogeneic SCT in first CR. In addition, the AML4a study addressed a question of therapy with granulocyte macrophage colony-stimulating factor (molgrastim; Sandoz, Basel, Switzerland) during and/or after chemotherapy.18 The AML29 trial addressed the question of the value of growth factor priming by adding granulocyte colony-stimulating factor (lenograstim; Aventis, Hoevelaken, the Netherlands) to the induction program on the days of chemotherapy.19
Patients who developed a first relapse were treated off protocol at the discretion of the local medical team. In these patients, modality of treatment after first relapse, achievement of second CR, and overall survival (OS) were assessed.
Cytogenetic Analysis
At diagnosis, samples of bone marrow and blood were examined for cytogenetic abnormalities using standard banding techniques and classification according to the International System for Human Cytogenetic Nomenclature.20 Karyotype abnormalities that involved t(16;16)(p13;q22), inv(16)(p13;q22), or t(8;21)(q22;q22) with or without other cytogenetic abnormalities were considered favorable cytogenetics. Monosomies or deletions of chromosomes 5 and 7 (–5, –7, del 5q–, del 7q–); abnormalities of the long arm of chromosome 3(q21;q26), t(6;9)(p23;q34), or t(9;22)(q34;q11); abnormalities involving the long arm of chromosome 11 (abn 11q23); or complex cytogenetic abnormalities (defined as at least four unrelated cytogenetic clones) were considered unfavorable risk factors. Other cytogenetic abnormalities and normal cytogenetics were designated intermediate risk factors.3-6 No cytogenetic data were available from 87 patients.
Statistical Methods
Overall survival after first relapse was the main end point for this analysis. The following factors were analyzed for their association with survival: age at first relapse, French-American-British (FAB) cytologic classification at diagnosis, cytogenetics at diagnosis (cytogenetics at relapse was not available), number of induction cycles of chemotherapy to reach first CR, WBC at diagnosis, whether previous SCT was performed in first CR, and length of RFI after first CR. Actuarial survival probabilities in subgroups were calculated with the method of Kaplan and Meier. Cox regression analysis was used to estimate relative hazard rates and to test for differences or trends in subgroups of each factor. Isotonic regression analysis21 was applied to determine whether continuous factors age and RFI were approximately linear or required transformation before inclusion in the Cox regression analysis. With isotonic regression analysis, a step function is fitted that describes the best monotic relationship between hazard rate and a continuous covariate. If this step function shows marked deviation from linearity, this is an indication for transformation or subdivision in categories of the continuous covariate. Given that there was no evidence of difference in survival between patients with RFI of 4 months and between patients with RFI of 25 months, values of RFI larger than 25 months were shrunk to 25 months and values below 4 months were expanded to 4 months in all regression analyses. Stepwise multivariable Cox regression analysis was applied with backward selection. To prevent overfitting, only factors that showed statistically significant association with survival with P < .01, based on the likelihood ratio test after adjustment for the other factors in the model, were included in the final model. Internal validation with bootstrap (500 replications) was applied to validate the modeling process and the stability of the final model.22,23
RESULTS
Survival After First Relapse
After first relapse, 29% of the patients survived at 12 months and 11% survived at 5 years. The OS probabilities were studied in relation to WBC, FAB classification and cytogenetics at diagnosis, number of induction cycles of chemotherapy required for attaining first CR, RFI after first CR, age at relapse, and whether SCT had been undertaken before first relapse (Table 2 and Fig 1). Patients with favorable cytogenetics [ie, t(16;16), inv(16), or t(8;21)] at diagnosis continued to express more favorable prognosis at relapse. However, patients with t(16;16) or inv(16) had a better prognosis compared with those with t(8;21). There was no difference in OS between the groups with unknown, intermediate-risk, and unfavorable-risk karyotypes. Therefore, these three groups were classified together as one common category of other cytogenetics. Univariate analysis showed that non-M5 FAB classification, favorable cytogenetics, low number of induction cycles toward first CR, longer RFI after first CR, younger age, and no previous SCT predicted for improved OS.
Development of a Prognostic Score for AML Patients in First Relapse
Our aim was to develop a prognostic score suitable for broad clinical use. We applied multivariate Cox regression analysis with stepwise backward selection. Initially, all seven factors studied (ie, age, RFI, cytogenetics, previous SCT, FAB classification, WBC, number of cycles required to reach first CR) were included in the model. Factors that showed no or only limited statistically significant association (P > .01) with OS adjusted for the remaining factors in the model, were deleted from the model. The final model included RFI, age, previous SCT, and cytogenetics (Table 3). On the basis of the variables and associated regression coefficients of this final model, a prognostic score was derived with the formula: 0.016 x (age in years) – 0.068 x (RFI in months) – 0.50 x [t(8;21), no = 0, yes = 1] – 1.24 x [t(16;16) or inv(16), no = 0, yes = 1] + 0.43 x (previous SCT, no = 0, yes = 1).
The stability of the model was verified by 500 bootstrap replications of the complete backward selection process. In nearly all bootstrap replications (95%) the four factors were reproducibly selected, whereas non-M5 FAB classification was included in the model in 46% of the patients. This validated the choice of the four parameters. The factors WBC and number of cycles required to reach first CR were selected in only 13% and 1% of the patients, respectively (Table 3). For each bootstrap replication a prognostic score was calculated based on the variables included in the model fitted for that replication and the associated regression coefficients. Correlations between these scores and the prognostic score for the final model were all high, with a mean correlation of 0.96 (range, 0.78 to 1.00). This confirms that the amount of overfitting in the final model is limited.
Simplified Prognostic Score for AML Patients in First Relapse
the model presented in the previous section, a simplified prognostic score for adult patients with AML in first relapse was derived that might be applied in clinical practice. For this simplified score, we fitted the Cox regression model in which continuous factors age and RFI were replaced by indicator variables for different ranges. Subdivision of age in this simplified score in three intervals was based on calculation of hazard ratios for each 5-year age interval, and grouping together of intervals with lowest, highest, and intermediate hazard ratios. The same was done for the subdivision of RFI with subdivision in 6-month intervals. Estimated coefficients for this model are shown in Table 4. Division of these coefficients by 0.25 and rounding to the nearest integer led to the score points in Table 4.
The variables with the highest impact on the score are RFI and cytogenetics at diagnosis (both 0 to 5 points), followed by age at relapse (0 to 2 points), and previous SCT (2 points). This score has a theoretical range between 0 and 14, for which a low score corresponds with relatively favorable prognosis (ie, high OS). In contrast, a high score correlates with poor prognosis (ie, low OS). For all 667 patients in the study, the prognostic score was calculated with a range of 1 to 14. Patients with the same score were classified together and the mean OS was determined for each score outcome. We observed a gradual decline in OS with increasing score. Subsequently, three groups were distinguished on the basis of OS probabilities. Patients with scores from 1 to 6 were associated with favorable outcome (1-year OS, 70%; 5-year OS, 46%). A second group of patients with scores from 7 to 9 had comparatively less favorable outcome (1-year OS, 49%; 5-year OS, 18%). For patients with scores from 10 to 14, an adverse prognosis was derived (1-year OS, 16%; 5-year OS, 4%; Fig 2). Composition of these three risk groups reflects the contribution of different factors to the score (Table 5). Favorable group A (score, 1 to 6; n = 57; 9% of all patients) has the lowest mean age, the longest mean RFI, the lowest proportion of patients with previous SCT, and the most patients with favorable cytogenetics. Patients with intermediate prognosis (group B; score, 7 to 9; n = 165; 25% of all patients) differ from those in group A, particularly with regard to higher age and greater proportion of patients treated with SCT before first relapse. The majority of patients in the poor prognostic group C (score, 10 to 14; n = 445; 67% of all patients) had a notably short RFI and were (on average) of older age. In addition, the poor prognostic group contained few patients with favorable cytogenetics, whereas 29% of the patients had been treated with SCT before first relapse.
To compare predictions on the basis of the fitted model in Table 3 with those based on the simplified model, a subdivision was made from the range of the full score in three groups of the same size as the simplified score classes. Ninety-one percent of the patients were classified in the same group and the survival curves of the three groups from the fitted model and the simplified model were almost identical, indicating that almost no discriminatory information was lost. When the three prognostic groups were compared for patients in the AML4/4a and AML29 trial separately, nearly identical prognostic groups, both in size and OS, could be identified. A test for interaction between trial and prognostic group with respect to OS was not statistically significant.
Treatment After Relapse
We classified the various therapeutic approaches applied after first relapse as no further intensive treatment, intensive chemotherapy, intensive therapy followed by autologous SCT, intensive therapy followed by allogeneic SCT, donor lymphocyte infusion following allogeneic SCT in first- or second-line treatment, anti-CD33 antibody treatment, or treatment unknown. Eighty-one percent of all patients received some form of treatment. Of these, 46% achieved a second CR (Table 6). Patients treated with autologous or allogeneic SCT in first CR had a lower second CR rate (33%), compared with patients not treated with SCT in first CR (49%).
Although the majority of patients in any of the prognostic groups received salvage treatment, the percentage of second CR in the favorable prognostic group A was higher (85% of the patients that received treatment) than those in intermediate and poor prognostic groups B and C (60% and 34% of the patients that received treatment, respectively). Because patients who developed first relapse were treated off protocol at the discretion of the local medical team, analysis of probability of OS in relation to treatment strategy was obviously hampered by unavoidable selection bias. Therefore, data relating to salvage treatment have mainly descriptive value. For instance, treatment with SCT after relapse was offered to 220 patients. Most of these transplantations (60%) were given as consolidation treatment in second CR, but a considerable proportion of transplantations (40%) were given while the patient was still in first relapse and reinduction treatment with chemotherapy had not resulted in second CR. As a result, 47 AML patients in first relapse who received a transplantation never reached a second CR.
To reduce selection bias, we restricted analysis of outcome in relation to treatment to patients who actually reached a second CR (Table 7). The 1-year OS of patients from prognostic groups A and B was comparable for the three main treatment modalities (ie, chemotherapy, autologous SCT, or allogeneic SCT; 1-year OS, 64% to 100%). Patients in the poor prognostic group C had a lower probability of second CR, and OS of these patients with second CR was lower compared with the more favorable prognostic groups A and B. Although in all prognostic groups the best long-term survival was observed in patients who could be treated with allogeneic SCT, the possibility of selection bias prevents any definite conclusions.
DISCUSSION
In this study, we present a prognostic score for patients (age 15 to 60 years) with AML in first relapse based on multivariate analysis of 667 patients using four clinically relevant parameters (ie, RFI, cytogenetics, age, and previous SCT). Three prognostic subsets were defined. Patients from favorable group A are more likely to attain second CR and show an almost 50% OS rate at 5 years. Intermediate prognostic group B contains patients with less favorable prognosis. Although a significant number of patients reach second CR, the proportion of patients with long-term OS is less (ie, probability of 18% OS at 5 years). Patients with a poor risk index (group C) have a distinctly dismal prognosis. When we analyzed only patients who reached second CR, differences in OS between favorable and poor prognostic groups were still apparent. This difference was present in all three salvage treatment modalities. These data together are consistent with the powerful prognostic impact of the predictive score independent of subsequent therapy. They are most likely determined by intrinsic disease (eg, cytogenetics) and host (eg, age) characteristics.
To verify the presented prognostic score with four parameters, analysis with 500 bootstrap replications was performed. In 95% of the bootstrap replications, the four factors were reproducibly selected. Subsequently, a high correlation of 0.96 was shown between the calculated prognostic scores from the bootstraps and the presented prognostic score, confirming that overfitting was limited in this model. An approach of splitting the data set in a training set and a validation set was not performed because data splitting has the significant disadvantage of reducing sample size for both model development and model testing. This would have created a less stable model, as has been discussed by Harrell.23 The prognostic score merits validation in future studies involving independent data sets.
Four statistically significant parameters generated a prognostic score for AML patients in first relapse that is easy to apply in clinical practice. Thus, integration of four prognostic factors, which were previously individually applied to patients with relapsed AML, yields one simple score. Although the presented prognostic system contains four different covariates that are partially dependent on each other, the analysis suggests that each individual factor has statistical significance (Table 3). Addition of other possible relevant parameters (ie, FAB classification, WBC, number of cycles of induction chemotherapy to reach first CR) did not enhance the prognostic index.
Previously, RFI has been considered as the major factor determining prognosis after relapse,9,10,12 and risk stratification methods for AML in first relapse were based on length of RFI only.16,17 When we compared the prognostic index presented here with a method solely based on length of RFI, the present prognostic index appears to identify patients with relatively favorable prognosis more accurately. For this comparison, we distinguished the population of 667 AML patients in first relapse of this study into three groups according to length of RFI, as previously proposed (group 1, RFI of > 24 months; group 2, RFI from 13 to 24 months; group 3, RFI of 12 months).17 Group 1, with the longest RFI, contained only patients from our favorable and intermediate prognostic groups A and B (Table 8). However, intermediate group 2 (114 patients) contained 39 patients from poor prognostic group C with limited prognosis. Conversely, group 3 with the shortest RFI (505 patients) included 99 patients with favorable and intermediate prognosis according to our analysis. This comparison confirms that the factors age, cytogenetics, and previous SCT add relevant prognostic value to the factor RFI.
The presented data are in agreement with previously reported smaller studies that have shown a positive relation between younger age and longer duration of RFI before first relapse with OS.9-15 In addition, this study contains 62 AML patients who experienced relapse and had favorable cytogenetics [ie, t(16;16), inv(16), or t(8;21)] at diagnosis. The presence of favorable cytogenetics at diagnosis continued to express favorable prognostic value relating to survival after relapse. This is also confirmed by Kern et al,24 who show that karyotype instability between diagnosis and relapse does not influence the prognostic effect of favorable cytogenetics. Interestingly, AML patients with t(16;16) or inv(16) have better prognosis compared with AML patients with t(8;21). The analysis reveals a difference in prognosis between patients who experienced relapse with or without previous SCT. Reasons for the poor outcome of patients who have had previous SCT may relate to the cumulative toxicity of prior cytotoxic therapy and lack of additional therapeutic options. Alternatively, it is also possible that leukemia that relapses after high-dose therapy and SCT is a priori of a more aggressive type.
The prognostic index of adult patients (age 15 to 60 years) with AML in first relapse provides insight into the considerable variation in prognosis of patients with AML at the point of first relapse, and leads to a practical framework for therapeutic decision making. Patients with relatively good prognosis regarding survival are patients with a prognostic score below 10, which assigns them to favorable- and intermediate-risk groups A and B. These are the best candidates for additional intensive salvage treatment. In contrast, patients from poor-risk group C have highly unfavorable perspectives at first relapse. Many of the latter could be considered for experimental or palliative approaches. In fact, available conventional, experimental, and palliative treatment options can be considered in the perspective of a quantitative prognostic estimate.
Appendix
The following centers and investigators participated in the study: the Netherlands: VU University Medical Center, Amsterdam (P.C. Huijgens, G.J. Ossenkoppele); University Medical Center, Utrecht (L.F. Verdonck, A.W. Dekker); University Hospital, Groningen (E. Vellenga, S.M.J.G. Daenen); Erasmus University Medical Center and Daniel den Hoed Cancer Center, Rotterdam (B. Lwenberg, G.E. de Greef, P. Sonneveld, W.L.J. van Putten, D.A. Breems); Academic Medical Center, Amsterdam (J. Van der Lelie); University Hospital, Maastricht (H.C. Schouten); Leijenburg Hospital, The Hague (P.W. Wijermans); Sophia Hospital, Zwolle (M. van Marwijk Kooy); Hospital Eemland, Amersfoort (S. Wittebol); Medical Center Twente, Enschede (M.R. Schaafsma); and Antoni van Leeuwenhoek Hospital, Amsterdam (J.W. Baars); Belgium: Hospital Gasthuisberg, Leuven (M.A. Boogaerts, G. Verhoef); Cliniques Universitaires Saint-Luc, Brussels (A. Ferrant); Cliniques Universitaires de Mont-Godinne, Yvoir (A. Bosly); and Hpital de Jolimont, Haine-St. Paul (A. Delannoy); Switzerland: University Hospital, Zürich (E. Jacky, J. Gmür); University Hospital, Bern (M.F. Fey, A. Tobler); University Hospital, Lausanne (T. Kovacsovics); University Hospital, Basel (A. Gratwohl, A. Tichelli); Kantonsspital, Aarau (M. Wernli); Hospital St. Giovanni, Bellinzona (G. Marini, L. Leconcini); Hpital Cantonal Universitaire, Geneva (B. Chapuis); Kantonsspital, St. Gallen (U. Hess); University Hospital, Neuchtel (D. Piguet); and Kantonsspital, Winterthur (T. Kroner); Germany: Johannes Gutenberg University Hospital, Mainz (M. Theobald, J. Beck); and Nordwest Hospital, Frankfurt am Main (A. Knuth). Cytogenetic review: A. Hagemeijer (Leuven, Belgium), S.L. Bhola (Leiden, the Netherlands), and M. Jotterand-Bellomo (Lausanne, Switzerland). Central data management: M. van Os, A.J.M. Meurisse, C. van Hooije (HOVON Cooperative Group Data Center, Rotterdam, the Netherlands).
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Lwenberg B, Downing JR, Burnett A: Acute myeloid leukemia. N Engl J Med 341:1051-1062, 1999
Burnett AK: Acute myeloid leukemia: Treatment of adults under 60 years. Rev Clin Exp Hematol 6:26-45, 2002
Grimwade D, Walker H, Oliver F, et al: The importance of diagnostic cytogenetics on outcome in AML: Analysis of 1612 patients entered into the MRC AML 10 trial. Blood 92:2322-2333, 1998
Wheatley K, Burnett AK, Goldstone AH, et al: A simple, robust, validated and highly predictive index for the determination of risk-directed therapy in acute myeloid leukaemia derived form the MRC AML 10 trial. Br J Haematol 107:69-79, 1999
Slovak ML, Kopecky KJ, Cassileth PA, et al: Karyotypic analysis predicts outcome of preremission and postremission therapy in adult acute myeloid leukemia: A Southwest Oncology Group/Eastern Cooperative Oncology Group study. Blood 96:4075-4083, 2000
Byrd JC, Mrozek K, Dodge RK, et al: Pretreatment cytogenetic abnormalities are predictive of induction success, cumulative incidence of relapse, and overall survival in adult patients with de novo acute myeloid leukemia: Results from Cancer and Leukemia group B (CALBG 8461). Blood 100:4325-4336, 2002
Leopold LH, Willemze R: The treatment of acute myeloid leukemia in first relapse: A comprehensive review of the literature. Leuk Lymphoma 43:1715-1727, 2002
Lee S, Tallman MS, Oken MM, et al: Duration of second complete remission compared with first complete remission in patients with acute myeloid leukemia. Leukemia 14:1345-1348, 2000
Kantarjian HM, Keating MJ, Walters RS, et al: The characteristics and outcome of patients with late relapse acute myelogenous leukemia. J Clin Oncol 6:232-238, 1988
Mortimer J, Blinder MA, Schulman S, et al: Relapse of acute leukemia after marrow transplantation: Natural history and results of subsequent therapy. J Clin Oncol 7:50-57, 1989
Keating MJ, Kantarjian H, Smith TL, et al: Response to salvage therapy and survival after relapse in acute myelogenous leukemia. J Clin Oncol 7:1071-1080, 1989
Uhlman DL, Bloomfield CD, Hurd DD, et al: Prognostic factors at relapse for adults with acute myeloid leukemia. Am J Hematol 33:110-116, 1990
Angelov L, Brandwein JM, Baker MA, et al: Results of therapy for acute myeloid leukemia in first relapse. Leuk Lymphoma 6:15-24, 1991
Thalhammer F, Geissler K, Jager U, et al: Duration of second complete remission in patients with acute myeloid leukemia treated with chemotherapy: A retrospective single-center study. Ann Hematol 72:216-222, 1996
Kern W, Schoch C, Haferlach T, et al: Multivariate analysis of prognostic factors in patients with refractory and relapsed acute myeloid leukemia undergoing sequential high-dose cytosine arabinoside and mitoxantrone (S-HAM) salvage therapy: Relevance of cytogenetic abnormalities. Leukemia 14:226-231, 2000
Hiddemann W, Martin WR, Sauerland CM, et al: Definition of refractoriness against conventional chemotherapy in acute myeloid leukemia: A proposal based on the results of retreatment by thioguanine, cytosine arabinoside, and daunorubicin (TAD9) in 150 patients with relapse after standardized first line therapy. Leukemia 4:184-188, 1990
Estey E, Kornblau S, Pierce S, et al: A stratification system for evaluating and selecting therapies in patients with relapsed or primary refractory acute myelogenous leukemia. Blood 88:756, 1996
Lwenberg B, Boogaerts MA, Daenen SMGJ, et al: Value of different modalities of granulocyte-macrophage colony-stimulating factor applied during or after induction therapy of acute myeloid leukemia. J Clin Oncol 15:3496-3506, 1997
Lwenberg B, Van Putten W, Theobald M, et al: Effect of priming with granulocyte-colony-stimulating factor on the outcome of chemotherapy for acute myeloid leukemia. N Engl J Med 349:743-752, 2003
Mitelman F ICSN 1995: An international system for human cytogenetic nomenclature. Basel, Switzerland, Karger, 1995
Barlow RE, Bartholomew DJ, Bremner JM, et al: Statistical Inference Under Restrictions. New York, NY, Wiley, 1972
Efron B, Tibshirani R: An Introduction to the Bootstrap. New York, NY, Chapman and Hall, 1993
Harrell FE: Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression and Survival Analysis. New York, NY, Springer, 2001
Kern W, Haferlach T, Schnittger S, et al: Karyotype instability between diagnosis and relapse in 117 patients with acute myeloid leukemia: Implications for resistance against therapy. Leukemia 16:2084-2091, 2002(Dimitri A. Breems, Wim L.)
Swiss Group for Clinical Cancer Research Collaborative Group (SAKK)
Department of Hematology and HOVON Data Center, Rotterdam
Department of Hematology, Free University Medical Center Department of Hematology, Academic Medical Center, Amsterdam
Department of Hematology, University Medical Center, Utrecht
Department of Hematology, University Hospital, Groningen, the Netherlands
Department of Hematology, Hospital Gasthuisberg, Leuven, Belgium
Department of Internal Medicine, University Hospital, Zürich, Switzerland
ABSTRACT
PURPOSE: The treatment of acute myeloid leukemia (AML) in first relapse is associated with unsatisfactory rates of complete responses that usually are short lived. Therefore, a clinically useful prognostic index can facilitate therapeutic decision making and evaluation of investigational treatment strategies at relapse of AML.
PATIENTS AND METHODS: A prognostic score is presented based on the multivariate analysis of 667 AML patients in first relapse among 1,540 newly diagnosed non-M3 AML patients (age 15 to 60 years) entered onto three successive Dutch-Belgian Hemato-Oncology Cooperative Group and the Swiss Group for Clinical Cancer Research Collaborative Group trials.
RESULTS: Four clinically relevant parameters are included in this index (ie, length of relapse-free interval after first complete remission, cytogenetics at diagnosis, age at relapse, and whether previous stem-cell transplantation was performed). Using this stratification system, three risk groups were defined: a favorable prognostic group A (overall survival [OS] of 70% at 1 year and 46% at 5 years), an intermediate-risk group B (OS of 49% at 1 year and 18% at 5 years), and a poor-risk group C (OS of 16% at 1 year and 4% at 5 years).
CONCLUSION: The prognostic index estimates the outcome of AML patients in first relapse using four commonly applied clinical parameters and might identify patients who are candidates for salvage and investigational therapy.
INTRODUCTION
Although the outcome of patients with acute myeloid leukemia (AML) has improved because of cytarabine- and anthracycline-based chemotherapy in combination with advanced supportive care and introduction of hematopoietic stem-cell transplantation (SCT), relapse continues to represent the leading cause of death in the majority of patients.1,2 The probability of relapse depends on risk factors such as age, pretreatment cytogenetics, and number of cycles of induction chemotherapy required for attaining the first complete hematologic response (CR).3-6 Insight into the individual role of each of these factors with respect to prognosis may support application of risk-adapted postinduction treatment.4
Treatment of AML in first relapse is associated with relatively low response rates.7 Whenever second CR is attained, the median duration of the second relapse-free interval (RFI) is generally considerably shorter than that of the first RFI.8 Because only a minority of patients who experience relapse will derive durable benefit from current reinduction therapy, it would be practically useful to be able to estimate prognosis at the time of relapse. This insight could then facilitate therapeutic decision making at this stage of the disease and guide individualized and investigational treatment strategies.
Factors predicting outcome of patients with AML in first relapse have been reported, and include RFI, age, and cytogenetics.9-15 However, proposed stratification methods for selecting therapies for patients with relapsed AML have only been based on the duration of RFI,9,10,12,16,17 thus neglecting the influence of other known prognostic factors. Predictive scores for patients with AML in first relapse that include more covariates could be generally applicable, if they would furnish a simple and statistically valid prognostic index. We used the results of an analysis of the outcome of 667 unselected patients with first relapsed non-M3 AML among patients who were enrolled onto three successive Dutch-Belgian Hemato-Oncology Cooperative Group and the Swiss Group for Clinical Cancer Research Collaborative Group phase III trials in adults (age 15 to 60 years) with newly diagnosed AML during the period 1987 to 2001,18,19 and considered four clinically relevant parameters (ie, RFI, age, cytogenetics, and previous transplantation).
PATIENTS AND METHODS
Patients
A total of 1,540 non-M3 AML patients were enrolled during the period 1987 to 2001 onto three successive Dutch-Belgian Hemato-Oncology Cooperative Group and Swiss Group for Clinical Cancer Research phase III trials (AML4, AML4a, AML29).18,19 The studies were for eligible patients (age 15 to 60 years) with newly diagnosed AML. Of these 1,540 patients, 1,269 had attained CR, of whom 444 are alive in CR. Among the complete responders, 158 had died from nonleukemic causes and 667 had recurrence of AML. This report is based on analysis of the cohort of 667 patients with AML in first relapse with a median follow-up from relapse of 56 months (Table 1).
The studies were approved by the ethics committees of participating institutions and were conducted in accordance with the Declaration of Helsinki. All participants gave their informed consent.
Treatment Protocols
Treatment in the AML4/4a and AML29 studies involved one cycle of induction with an anthracycline (daunorubicin or idarubicin) in combination with cytarabine (200 mg/m2 for 7 days) and a second cycle of amsacrine with intermediate-dose cytarabine (1,000 mg/m2 every 12 hours for 6 days). If CR occurred, patients in the AML4/4a protocol were randomly assigned after a third cycle of chemotherapy (mitoxantrone and etoposide) to treatment with high-dose busulfan and cyclophosphamide followed by autologous SCT or no additional treatment. In the AML29 study, patients in CR after two cycles of chemotherapy were randomly assigned to treatment with a third cycle of chemotherapy (mitoxantrone and etoposide) without SCT or high-dose chemotherapy with busulfan and cyclophosphamide followed by autologous SCT. When an HLA-identical sibling donor was available, eligible patients proceeded to allogeneic SCT. In the AML29 trial protocol, patients with inv(16), t(16;16), or t(8;21) and WBC count of less than 20 x 109/L at diagnosis were no longer treated with autologous or allogeneic SCT in first CR. In addition, the AML4a study addressed a question of therapy with granulocyte macrophage colony-stimulating factor (molgrastim; Sandoz, Basel, Switzerland) during and/or after chemotherapy.18 The AML29 trial addressed the question of the value of growth factor priming by adding granulocyte colony-stimulating factor (lenograstim; Aventis, Hoevelaken, the Netherlands) to the induction program on the days of chemotherapy.19
Patients who developed a first relapse were treated off protocol at the discretion of the local medical team. In these patients, modality of treatment after first relapse, achievement of second CR, and overall survival (OS) were assessed.
Cytogenetic Analysis
At diagnosis, samples of bone marrow and blood were examined for cytogenetic abnormalities using standard banding techniques and classification according to the International System for Human Cytogenetic Nomenclature.20 Karyotype abnormalities that involved t(16;16)(p13;q22), inv(16)(p13;q22), or t(8;21)(q22;q22) with or without other cytogenetic abnormalities were considered favorable cytogenetics. Monosomies or deletions of chromosomes 5 and 7 (–5, –7, del 5q–, del 7q–); abnormalities of the long arm of chromosome 3(q21;q26), t(6;9)(p23;q34), or t(9;22)(q34;q11); abnormalities involving the long arm of chromosome 11 (abn 11q23); or complex cytogenetic abnormalities (defined as at least four unrelated cytogenetic clones) were considered unfavorable risk factors. Other cytogenetic abnormalities and normal cytogenetics were designated intermediate risk factors.3-6 No cytogenetic data were available from 87 patients.
Statistical Methods
Overall survival after first relapse was the main end point for this analysis. The following factors were analyzed for their association with survival: age at first relapse, French-American-British (FAB) cytologic classification at diagnosis, cytogenetics at diagnosis (cytogenetics at relapse was not available), number of induction cycles of chemotherapy to reach first CR, WBC at diagnosis, whether previous SCT was performed in first CR, and length of RFI after first CR. Actuarial survival probabilities in subgroups were calculated with the method of Kaplan and Meier. Cox regression analysis was used to estimate relative hazard rates and to test for differences or trends in subgroups of each factor. Isotonic regression analysis21 was applied to determine whether continuous factors age and RFI were approximately linear or required transformation before inclusion in the Cox regression analysis. With isotonic regression analysis, a step function is fitted that describes the best monotic relationship between hazard rate and a continuous covariate. If this step function shows marked deviation from linearity, this is an indication for transformation or subdivision in categories of the continuous covariate. Given that there was no evidence of difference in survival between patients with RFI of 4 months and between patients with RFI of 25 months, values of RFI larger than 25 months were shrunk to 25 months and values below 4 months were expanded to 4 months in all regression analyses. Stepwise multivariable Cox regression analysis was applied with backward selection. To prevent overfitting, only factors that showed statistically significant association with survival with P < .01, based on the likelihood ratio test after adjustment for the other factors in the model, were included in the final model. Internal validation with bootstrap (500 replications) was applied to validate the modeling process and the stability of the final model.22,23
RESULTS
Survival After First Relapse
After first relapse, 29% of the patients survived at 12 months and 11% survived at 5 years. The OS probabilities were studied in relation to WBC, FAB classification and cytogenetics at diagnosis, number of induction cycles of chemotherapy required for attaining first CR, RFI after first CR, age at relapse, and whether SCT had been undertaken before first relapse (Table 2 and Fig 1). Patients with favorable cytogenetics [ie, t(16;16), inv(16), or t(8;21)] at diagnosis continued to express more favorable prognosis at relapse. However, patients with t(16;16) or inv(16) had a better prognosis compared with those with t(8;21). There was no difference in OS between the groups with unknown, intermediate-risk, and unfavorable-risk karyotypes. Therefore, these three groups were classified together as one common category of other cytogenetics. Univariate analysis showed that non-M5 FAB classification, favorable cytogenetics, low number of induction cycles toward first CR, longer RFI after first CR, younger age, and no previous SCT predicted for improved OS.
Development of a Prognostic Score for AML Patients in First Relapse
Our aim was to develop a prognostic score suitable for broad clinical use. We applied multivariate Cox regression analysis with stepwise backward selection. Initially, all seven factors studied (ie, age, RFI, cytogenetics, previous SCT, FAB classification, WBC, number of cycles required to reach first CR) were included in the model. Factors that showed no or only limited statistically significant association (P > .01) with OS adjusted for the remaining factors in the model, were deleted from the model. The final model included RFI, age, previous SCT, and cytogenetics (Table 3). On the basis of the variables and associated regression coefficients of this final model, a prognostic score was derived with the formula: 0.016 x (age in years) – 0.068 x (RFI in months) – 0.50 x [t(8;21), no = 0, yes = 1] – 1.24 x [t(16;16) or inv(16), no = 0, yes = 1] + 0.43 x (previous SCT, no = 0, yes = 1).
The stability of the model was verified by 500 bootstrap replications of the complete backward selection process. In nearly all bootstrap replications (95%) the four factors were reproducibly selected, whereas non-M5 FAB classification was included in the model in 46% of the patients. This validated the choice of the four parameters. The factors WBC and number of cycles required to reach first CR were selected in only 13% and 1% of the patients, respectively (Table 3). For each bootstrap replication a prognostic score was calculated based on the variables included in the model fitted for that replication and the associated regression coefficients. Correlations between these scores and the prognostic score for the final model were all high, with a mean correlation of 0.96 (range, 0.78 to 1.00). This confirms that the amount of overfitting in the final model is limited.
Simplified Prognostic Score for AML Patients in First Relapse
the model presented in the previous section, a simplified prognostic score for adult patients with AML in first relapse was derived that might be applied in clinical practice. For this simplified score, we fitted the Cox regression model in which continuous factors age and RFI were replaced by indicator variables for different ranges. Subdivision of age in this simplified score in three intervals was based on calculation of hazard ratios for each 5-year age interval, and grouping together of intervals with lowest, highest, and intermediate hazard ratios. The same was done for the subdivision of RFI with subdivision in 6-month intervals. Estimated coefficients for this model are shown in Table 4. Division of these coefficients by 0.25 and rounding to the nearest integer led to the score points in Table 4.
The variables with the highest impact on the score are RFI and cytogenetics at diagnosis (both 0 to 5 points), followed by age at relapse (0 to 2 points), and previous SCT (2 points). This score has a theoretical range between 0 and 14, for which a low score corresponds with relatively favorable prognosis (ie, high OS). In contrast, a high score correlates with poor prognosis (ie, low OS). For all 667 patients in the study, the prognostic score was calculated with a range of 1 to 14. Patients with the same score were classified together and the mean OS was determined for each score outcome. We observed a gradual decline in OS with increasing score. Subsequently, three groups were distinguished on the basis of OS probabilities. Patients with scores from 1 to 6 were associated with favorable outcome (1-year OS, 70%; 5-year OS, 46%). A second group of patients with scores from 7 to 9 had comparatively less favorable outcome (1-year OS, 49%; 5-year OS, 18%). For patients with scores from 10 to 14, an adverse prognosis was derived (1-year OS, 16%; 5-year OS, 4%; Fig 2). Composition of these three risk groups reflects the contribution of different factors to the score (Table 5). Favorable group A (score, 1 to 6; n = 57; 9% of all patients) has the lowest mean age, the longest mean RFI, the lowest proportion of patients with previous SCT, and the most patients with favorable cytogenetics. Patients with intermediate prognosis (group B; score, 7 to 9; n = 165; 25% of all patients) differ from those in group A, particularly with regard to higher age and greater proportion of patients treated with SCT before first relapse. The majority of patients in the poor prognostic group C (score, 10 to 14; n = 445; 67% of all patients) had a notably short RFI and were (on average) of older age. In addition, the poor prognostic group contained few patients with favorable cytogenetics, whereas 29% of the patients had been treated with SCT before first relapse.
To compare predictions on the basis of the fitted model in Table 3 with those based on the simplified model, a subdivision was made from the range of the full score in three groups of the same size as the simplified score classes. Ninety-one percent of the patients were classified in the same group and the survival curves of the three groups from the fitted model and the simplified model were almost identical, indicating that almost no discriminatory information was lost. When the three prognostic groups were compared for patients in the AML4/4a and AML29 trial separately, nearly identical prognostic groups, both in size and OS, could be identified. A test for interaction between trial and prognostic group with respect to OS was not statistically significant.
Treatment After Relapse
We classified the various therapeutic approaches applied after first relapse as no further intensive treatment, intensive chemotherapy, intensive therapy followed by autologous SCT, intensive therapy followed by allogeneic SCT, donor lymphocyte infusion following allogeneic SCT in first- or second-line treatment, anti-CD33 antibody treatment, or treatment unknown. Eighty-one percent of all patients received some form of treatment. Of these, 46% achieved a second CR (Table 6). Patients treated with autologous or allogeneic SCT in first CR had a lower second CR rate (33%), compared with patients not treated with SCT in first CR (49%).
Although the majority of patients in any of the prognostic groups received salvage treatment, the percentage of second CR in the favorable prognostic group A was higher (85% of the patients that received treatment) than those in intermediate and poor prognostic groups B and C (60% and 34% of the patients that received treatment, respectively). Because patients who developed first relapse were treated off protocol at the discretion of the local medical team, analysis of probability of OS in relation to treatment strategy was obviously hampered by unavoidable selection bias. Therefore, data relating to salvage treatment have mainly descriptive value. For instance, treatment with SCT after relapse was offered to 220 patients. Most of these transplantations (60%) were given as consolidation treatment in second CR, but a considerable proportion of transplantations (40%) were given while the patient was still in first relapse and reinduction treatment with chemotherapy had not resulted in second CR. As a result, 47 AML patients in first relapse who received a transplantation never reached a second CR.
To reduce selection bias, we restricted analysis of outcome in relation to treatment to patients who actually reached a second CR (Table 7). The 1-year OS of patients from prognostic groups A and B was comparable for the three main treatment modalities (ie, chemotherapy, autologous SCT, or allogeneic SCT; 1-year OS, 64% to 100%). Patients in the poor prognostic group C had a lower probability of second CR, and OS of these patients with second CR was lower compared with the more favorable prognostic groups A and B. Although in all prognostic groups the best long-term survival was observed in patients who could be treated with allogeneic SCT, the possibility of selection bias prevents any definite conclusions.
DISCUSSION
In this study, we present a prognostic score for patients (age 15 to 60 years) with AML in first relapse based on multivariate analysis of 667 patients using four clinically relevant parameters (ie, RFI, cytogenetics, age, and previous SCT). Three prognostic subsets were defined. Patients from favorable group A are more likely to attain second CR and show an almost 50% OS rate at 5 years. Intermediate prognostic group B contains patients with less favorable prognosis. Although a significant number of patients reach second CR, the proportion of patients with long-term OS is less (ie, probability of 18% OS at 5 years). Patients with a poor risk index (group C) have a distinctly dismal prognosis. When we analyzed only patients who reached second CR, differences in OS between favorable and poor prognostic groups were still apparent. This difference was present in all three salvage treatment modalities. These data together are consistent with the powerful prognostic impact of the predictive score independent of subsequent therapy. They are most likely determined by intrinsic disease (eg, cytogenetics) and host (eg, age) characteristics.
To verify the presented prognostic score with four parameters, analysis with 500 bootstrap replications was performed. In 95% of the bootstrap replications, the four factors were reproducibly selected. Subsequently, a high correlation of 0.96 was shown between the calculated prognostic scores from the bootstraps and the presented prognostic score, confirming that overfitting was limited in this model. An approach of splitting the data set in a training set and a validation set was not performed because data splitting has the significant disadvantage of reducing sample size for both model development and model testing. This would have created a less stable model, as has been discussed by Harrell.23 The prognostic score merits validation in future studies involving independent data sets.
Four statistically significant parameters generated a prognostic score for AML patients in first relapse that is easy to apply in clinical practice. Thus, integration of four prognostic factors, which were previously individually applied to patients with relapsed AML, yields one simple score. Although the presented prognostic system contains four different covariates that are partially dependent on each other, the analysis suggests that each individual factor has statistical significance (Table 3). Addition of other possible relevant parameters (ie, FAB classification, WBC, number of cycles of induction chemotherapy to reach first CR) did not enhance the prognostic index.
Previously, RFI has been considered as the major factor determining prognosis after relapse,9,10,12 and risk stratification methods for AML in first relapse were based on length of RFI only.16,17 When we compared the prognostic index presented here with a method solely based on length of RFI, the present prognostic index appears to identify patients with relatively favorable prognosis more accurately. For this comparison, we distinguished the population of 667 AML patients in first relapse of this study into three groups according to length of RFI, as previously proposed (group 1, RFI of > 24 months; group 2, RFI from 13 to 24 months; group 3, RFI of 12 months).17 Group 1, with the longest RFI, contained only patients from our favorable and intermediate prognostic groups A and B (Table 8). However, intermediate group 2 (114 patients) contained 39 patients from poor prognostic group C with limited prognosis. Conversely, group 3 with the shortest RFI (505 patients) included 99 patients with favorable and intermediate prognosis according to our analysis. This comparison confirms that the factors age, cytogenetics, and previous SCT add relevant prognostic value to the factor RFI.
The presented data are in agreement with previously reported smaller studies that have shown a positive relation between younger age and longer duration of RFI before first relapse with OS.9-15 In addition, this study contains 62 AML patients who experienced relapse and had favorable cytogenetics [ie, t(16;16), inv(16), or t(8;21)] at diagnosis. The presence of favorable cytogenetics at diagnosis continued to express favorable prognostic value relating to survival after relapse. This is also confirmed by Kern et al,24 who show that karyotype instability between diagnosis and relapse does not influence the prognostic effect of favorable cytogenetics. Interestingly, AML patients with t(16;16) or inv(16) have better prognosis compared with AML patients with t(8;21). The analysis reveals a difference in prognosis between patients who experienced relapse with or without previous SCT. Reasons for the poor outcome of patients who have had previous SCT may relate to the cumulative toxicity of prior cytotoxic therapy and lack of additional therapeutic options. Alternatively, it is also possible that leukemia that relapses after high-dose therapy and SCT is a priori of a more aggressive type.
The prognostic index of adult patients (age 15 to 60 years) with AML in first relapse provides insight into the considerable variation in prognosis of patients with AML at the point of first relapse, and leads to a practical framework for therapeutic decision making. Patients with relatively good prognosis regarding survival are patients with a prognostic score below 10, which assigns them to favorable- and intermediate-risk groups A and B. These are the best candidates for additional intensive salvage treatment. In contrast, patients from poor-risk group C have highly unfavorable perspectives at first relapse. Many of the latter could be considered for experimental or palliative approaches. In fact, available conventional, experimental, and palliative treatment options can be considered in the perspective of a quantitative prognostic estimate.
Appendix
The following centers and investigators participated in the study: the Netherlands: VU University Medical Center, Amsterdam (P.C. Huijgens, G.J. Ossenkoppele); University Medical Center, Utrecht (L.F. Verdonck, A.W. Dekker); University Hospital, Groningen (E. Vellenga, S.M.J.G. Daenen); Erasmus University Medical Center and Daniel den Hoed Cancer Center, Rotterdam (B. Lwenberg, G.E. de Greef, P. Sonneveld, W.L.J. van Putten, D.A. Breems); Academic Medical Center, Amsterdam (J. Van der Lelie); University Hospital, Maastricht (H.C. Schouten); Leijenburg Hospital, The Hague (P.W. Wijermans); Sophia Hospital, Zwolle (M. van Marwijk Kooy); Hospital Eemland, Amersfoort (S. Wittebol); Medical Center Twente, Enschede (M.R. Schaafsma); and Antoni van Leeuwenhoek Hospital, Amsterdam (J.W. Baars); Belgium: Hospital Gasthuisberg, Leuven (M.A. Boogaerts, G. Verhoef); Cliniques Universitaires Saint-Luc, Brussels (A. Ferrant); Cliniques Universitaires de Mont-Godinne, Yvoir (A. Bosly); and Hpital de Jolimont, Haine-St. Paul (A. Delannoy); Switzerland: University Hospital, Zürich (E. Jacky, J. Gmür); University Hospital, Bern (M.F. Fey, A. Tobler); University Hospital, Lausanne (T. Kovacsovics); University Hospital, Basel (A. Gratwohl, A. Tichelli); Kantonsspital, Aarau (M. Wernli); Hospital St. Giovanni, Bellinzona (G. Marini, L. Leconcini); Hpital Cantonal Universitaire, Geneva (B. Chapuis); Kantonsspital, St. Gallen (U. Hess); University Hospital, Neuchtel (D. Piguet); and Kantonsspital, Winterthur (T. Kroner); Germany: Johannes Gutenberg University Hospital, Mainz (M. Theobald, J. Beck); and Nordwest Hospital, Frankfurt am Main (A. Knuth). Cytogenetic review: A. Hagemeijer (Leuven, Belgium), S.L. Bhola (Leiden, the Netherlands), and M. Jotterand-Bellomo (Lausanne, Switzerland). Central data management: M. van Os, A.J.M. Meurisse, C. van Hooije (HOVON Cooperative Group Data Center, Rotterdam, the Netherlands).
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Lwenberg B, Downing JR, Burnett A: Acute myeloid leukemia. N Engl J Med 341:1051-1062, 1999
Burnett AK: Acute myeloid leukemia: Treatment of adults under 60 years. Rev Clin Exp Hematol 6:26-45, 2002
Grimwade D, Walker H, Oliver F, et al: The importance of diagnostic cytogenetics on outcome in AML: Analysis of 1612 patients entered into the MRC AML 10 trial. Blood 92:2322-2333, 1998
Wheatley K, Burnett AK, Goldstone AH, et al: A simple, robust, validated and highly predictive index for the determination of risk-directed therapy in acute myeloid leukaemia derived form the MRC AML 10 trial. Br J Haematol 107:69-79, 1999
Slovak ML, Kopecky KJ, Cassileth PA, et al: Karyotypic analysis predicts outcome of preremission and postremission therapy in adult acute myeloid leukemia: A Southwest Oncology Group/Eastern Cooperative Oncology Group study. Blood 96:4075-4083, 2000
Byrd JC, Mrozek K, Dodge RK, et al: Pretreatment cytogenetic abnormalities are predictive of induction success, cumulative incidence of relapse, and overall survival in adult patients with de novo acute myeloid leukemia: Results from Cancer and Leukemia group B (CALBG 8461). Blood 100:4325-4336, 2002
Leopold LH, Willemze R: The treatment of acute myeloid leukemia in first relapse: A comprehensive review of the literature. Leuk Lymphoma 43:1715-1727, 2002
Lee S, Tallman MS, Oken MM, et al: Duration of second complete remission compared with first complete remission in patients with acute myeloid leukemia. Leukemia 14:1345-1348, 2000
Kantarjian HM, Keating MJ, Walters RS, et al: The characteristics and outcome of patients with late relapse acute myelogenous leukemia. J Clin Oncol 6:232-238, 1988
Mortimer J, Blinder MA, Schulman S, et al: Relapse of acute leukemia after marrow transplantation: Natural history and results of subsequent therapy. J Clin Oncol 7:50-57, 1989
Keating MJ, Kantarjian H, Smith TL, et al: Response to salvage therapy and survival after relapse in acute myelogenous leukemia. J Clin Oncol 7:1071-1080, 1989
Uhlman DL, Bloomfield CD, Hurd DD, et al: Prognostic factors at relapse for adults with acute myeloid leukemia. Am J Hematol 33:110-116, 1990
Angelov L, Brandwein JM, Baker MA, et al: Results of therapy for acute myeloid leukemia in first relapse. Leuk Lymphoma 6:15-24, 1991
Thalhammer F, Geissler K, Jager U, et al: Duration of second complete remission in patients with acute myeloid leukemia treated with chemotherapy: A retrospective single-center study. Ann Hematol 72:216-222, 1996
Kern W, Schoch C, Haferlach T, et al: Multivariate analysis of prognostic factors in patients with refractory and relapsed acute myeloid leukemia undergoing sequential high-dose cytosine arabinoside and mitoxantrone (S-HAM) salvage therapy: Relevance of cytogenetic abnormalities. Leukemia 14:226-231, 2000
Hiddemann W, Martin WR, Sauerland CM, et al: Definition of refractoriness against conventional chemotherapy in acute myeloid leukemia: A proposal based on the results of retreatment by thioguanine, cytosine arabinoside, and daunorubicin (TAD9) in 150 patients with relapse after standardized first line therapy. Leukemia 4:184-188, 1990
Estey E, Kornblau S, Pierce S, et al: A stratification system for evaluating and selecting therapies in patients with relapsed or primary refractory acute myelogenous leukemia. Blood 88:756, 1996
Lwenberg B, Boogaerts MA, Daenen SMGJ, et al: Value of different modalities of granulocyte-macrophage colony-stimulating factor applied during or after induction therapy of acute myeloid leukemia. J Clin Oncol 15:3496-3506, 1997
Lwenberg B, Van Putten W, Theobald M, et al: Effect of priming with granulocyte-colony-stimulating factor on the outcome of chemotherapy for acute myeloid leukemia. N Engl J Med 349:743-752, 2003
Mitelman F ICSN 1995: An international system for human cytogenetic nomenclature. Basel, Switzerland, Karger, 1995
Barlow RE, Bartholomew DJ, Bremner JM, et al: Statistical Inference Under Restrictions. New York, NY, Wiley, 1972
Efron B, Tibshirani R: An Introduction to the Bootstrap. New York, NY, Chapman and Hall, 1993
Harrell FE: Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression and Survival Analysis. New York, NY, Springer, 2001
Kern W, Haferlach T, Schnittger S, et al: Karyotype instability between diagnosis and relapse in 117 patients with acute myeloid leukemia: Implications for resistance against therapy. Leukemia 16:2084-2091, 2002(Dimitri A. Breems, Wim L.)