The impact of therapy-related acute myeloid leukemia (AML) on outcome in 2853 adult patients with newly diagnosed AML

Therapy-related acute myeloid leukemia (t-AML) is a recognized clinical syndrome occurring as a complication after cytotoxic and/or radiation therapy.^1-3  At present, approximately 10% of all AMLs arise after a patient's exposure to chemotherapy and/or radiation for a primary malignancy or autoimmune disease.^4,5  Patients with t-AML are considered to have an inferior outcome compared with de novo AML.^2,3,5  The latency period between diagnosis of the primary disease and occurrence of t-AML ranges between several months to several years and may depend on the cumulative dose, dose intensity, and type of preceding chemotherapy and/or radiation therapy.^2,3 

With respect to cytogenetics, t-AML more frequently have abnormal cytogenetics; in particular, they have an increased prevalence of adverse-risk karyotypes.^5-7  At present, few data exist regarding the frequency of gene mutations.⁷  Chromosomal aberrations in t-AML are thought to be the consequence of mutational events induced by previous therapy.⁸  Hematopoietic progenitor cells, that survive with acquired mutations caused by non- or misrepair, are at risk for leukemic transformation and finally result in overt AML. Some patients may have an inherited susceptibility for the development of t-AML.^8-10 

Depending on the chemotherapeutic agent and/or radiation, 2 subtypes of t-AML can be distinguished. The most common subtype, occurring after exposure to alkylating agents and/or radiation with a latency period of 5-10 years, is frequently accompanied by unbalanced cytogenetic abnormalities, such as loss of all or parts of chromosomes 5 and/or 7.^5,11-14  The second less common subtype, arising after treatment with agents targeting topoisomerase II, has shorter latency period of 1-5 years and frequently exhibits balanced chromosomal rearrangements involving MLL, RUNX1, and PML-RARA.^5,7,15,16  However, because in recent years most patients have received treatment with both alkylating agents and drugs that target topoisomerase II for previous malignancy, a discrimination according to the type of previous therapy is often not feasible. Therefore, in the current World Health Organization (WHO) classification therapy-related myeloid neoplasms are no longer subcategorized.

Because pretreatment cytogenetic and molecular aberrations are the most powerful prognostic parameters for clinical outcome in de novo AML,^17,18  the question arises whether the diagnosis of t-AML itself indicates a poor prognosis or whether the inferior outcome results from the association with an adverse genetic risk profile. So far, only a few authors have evaluated the characteristics and clinical impact of t-AML, in particular in the context of other clinical and biologic prognostic markers. The objective of our study was to evaluate the characteristics and clinical impact of t-AML in a large cohort of adult AML patients treated within prospective multicenter treatment trials.

Patients were enrolled on 6 prospective multicenter treatment trials of the German-Austrian AML Study Group (AMLSG) between 1993 and September 2008. All patients received age-adapted intensive induction and consolidation therapy as previously described (AML HD93¹⁹ ; APL95²⁰ ; AML HD98A²¹ ; AML HD98B²² ; AMLSG 06-04, NCT00151255; AMLSG 07-04; NCT00151242). Treatments were significantly less intensive in trials for patients older than 60 years of age. In all trials patients with t-AML were eligible if they had completed therapy for the previous malignancy, had no active disease, and were considered by their physician to be at low risk of relapse. The studies were approved by the institutional review boards of all the participating centers. All patients gave informed consent for treatment and for cytogenetic and molecular genetic analyses according to the Declaration of Helsinki. The diagnosis of AML was based on French-American-British Cooperative Group criteria²³  for the trials AML HD93, AML HD98A, and AML HD98B and after 2004 on WHO 2001 criteria²⁴  for the trials AMLSG 07-04 and AMLSG 06-04.

All leukemia samples were studied centrally in the reference laboratories of the AMLSG at the University of Ulm and at Hannover Medical School. Chromosome banding was performed by the use of standard techniques, and karyotypes were described according to the International System for Human Cytogenetic Nomenclature.²⁵ 

Leukemia samples were analyzed for mutations in the FLT3 (FLT3 internal tandem duplication [ITD], n = 2355; FLT3 tyrosine kinase domain mutations at codons D835 and I836, n = 2145), NPM1 (n = 2300), MLL (partial tandem duplication, n = 1804), as well as CEBPA (n = 1091; analysis restricted to cytogenetically normal AML), as previously described.²⁶ 

The definition of complete remission (CR) or therapeutic failures followed recommended criteria.²⁷  Overall survival (OS), relapse-free survival (RFS), cumulative incidence of relapse (CIR), and cumulative incidence of death in CR (CID) were defined as recommended.²⁷  Cytogenetic categorization into favorable-, intermediate-, and adverse-risk group followed recommended criteria.²⁸  Pairwise comparisons between patient characteristics (covariates) were performed by Mann-Whitney or Kruskal Wallis test for continuous variables and by Fisher exact test for categorical variables.

A multivariable log-normalized linear regression model was used to identify factors that influenced the duration of the latency period, including the covariates type of primary malignancy (solid vs hematologic), sex, type of chemotherapeutic agent (alkylating agent, antimetabolite, antitubulin, intercalating agent, or topoisomerase II inhibitor; Table 1),^29-33  and radiation.

The Kaplan-Meier method was used to estimate the distribution of RFS and OS.³⁴  Confidence interval estimation for the survival curves was determined by the cumulative hazard function with the use of the Greenwood formula for the standard error estimation. A Cox model was used to identify prognostic variables.³⁵  CIR and CID and their standard errors (SEs) were analyzed according to the method described by Gray³⁶  and included only patients attaining CR, with time calculated from the date of CR to the occurrence of an event (relapse or death). Prognostic Cox regression models were used for the end points of relapse and death in CR as well as for OS. All models included a variable accounting for different treatment intensities in younger (age 18-60 years) versus older patients (age > 60 years). In all multivariable models no variable selection was performed, and full models were presented. We estimated missing data for covariates by using 50 multiple imputations in chained equations incorporating predictive mean matching.³⁷  There was no difference in clinical outcome (CR, CIR, CID, and OS) between patients with missing data and those with complete datasets (P = .13, P = .12, P = .17, and P = .51, respectively). All statistical analyses were performed with the statistical software environment R Version 2.10.1, by use of the R packages rms Version 2.1-0, and cmprsk Version 2.2-0.³⁸ 

In total, 3177 adult AML patients (median age, 54.5 years) were enrolled on 6 prospective treatment trials. In 3026 (95.2%) patients information on type of AML was available: 200 t-AML, 2653 de novo AML, and 173 AML with a history of a myelodysplastic syndrome (MDS) or myeloproliferative neoplasm at least 3 months before diagnosis of AML.³⁹  The subgroup of de novo AML patients included 77 patients with a history of a neoplasm but without previous chemotherapy or radiation. Patients with missing data on type of AML and patients with a history of MDS/myeloproliferative neoplasm without chemotherapy were excluded from this study.

Compared with patients with de novo AML, patients with t-AML were older (57.8 vs 53.2 years; P < .0001), and women were more frequently affected than men (P < .0001), mainly because of the high frequency of t-AML after treatment of breast cancer. t-AML was associated with lower median white blood counts (P = .003), lower platelet counts (P = .02), greater hemoglobin levels (P = .04), and lower percentages of blasts in peripheral blood (P = .002) and bone marrow (P = .03; Table 2).

Compared with de novo AML, t-AML more frequently had abnormal karyotypes (75% vs 51%, P < .0001; Table 3, Figure 1). The distribution among cytogenetic risk categories differed significantly between t-AML and de novo AML. Whereas there was no difference in the frequency of favorable-risk abnormalities, t-AML was underrepresented in intermediate-risk cases but more frequently exhibited adverse-risk karyotypes. t-AML more frequently had -5 or 5q− (P = .005), -7 (P = .008), 7q− (P = .001), t(9;11) (P = .0001), t(v;11)(v;q23) (P = .07, in trend), abnl(17p) (P < .0001), complex karyotypes (P < .0001), and monosomal karyotypes (P < .0001; defined according to Breems et al⁴⁰ ); in contrast, t-AML less frequently had normal karyotype (P < .0001), and in trend (P = .06) trisomy 8 within a noncomplex karyotype (Table 3). Figure 2 shows the comparison of t-AML and de novo AML patients exhibiting at least one cytogenetic abnormality, with the exclusion of the WHO category “AML with recurrent genetic abnormalities.”¹  Trisomy 8 in t-AML was frequently associated with a complex karyotype, whereas in de novo AML, trisomy 8 was the most frequent abnormality within a noncomplex karyotype. Regarding molecular aberrations, both NPM1 mutations and FLT3-ITD were significantly less frequent in t-AML (P < .0001 and P = .0005, respectively; Table 3).

The median latency period between diagnosis of primary malignancy and the occurrence of t-AML was 4.04 years (range, 0.33-44.14 years). One hundred forty-two (71%) patients with t-AML had a previous solid cancer (Table 4). Breast cancer was the most common neoplasm (n = 74; 52%), followed by thyroid (n = 12; 8%; all patients had received radioiodine therapy), gastrointestinal (n = 10 patients; 7%), prostate (n = 9; 6%), and testicular cancer (n = 9, 6%). Twenty-eight of 142 (20%) patients had various other neoplasms. Fifty-two (27.5%) patients had a primary hematologic malignancy, with non-Hodgkin lymphoma (n = 25; 46%) and Hodgkin lymphoma (n = 20; 36%) being the most common ones. A total of 10 of the 55 (18%) patients had various other hematologic malignancies (Table 4). Three patients had undergone cytotoxic therapy for the treatment of an autoimmune disease, 2 with multiple sclerosis and 1 with a rheumatologic disorder (Table 4).

In 180 (90%) of the 200 t-AML patients, the treatment records were complete with respect to the treatment modality of primary disease; detailed information on type of chemotherapeutic agents and dosages were available in 148 (74%) of the 200 patients. Sixty-nine patients had previous chemotherapy only, 56 radiation only, and 55 patients had both chemotherapy and radiation. Only 7 patients received single-agent chemotherapy.

Chemotherapeutic agents were classified by mechanism of action (Table 1). In a multivariable log-normalized linear regression model, younger age at diagnosis of primary malignancy (P = .006) as well as administration of intercalating agents (P = .01) and topoisomerase II inhibitors (P = .009) were associated with a shorter latency period between diagnosis of primary malignancy and the occurrence of t-AML. In addition, we were interested in the association of latency period and subsequent cytogenetic abnormality categorized into (1) t(15;17); (2) t(8;21); (3) inv(16) or t(16;16); (4) t(9;11); (5) t(v;11)(v;q23); (6) -7; (7) 7q−; (8) -5 or 5q−; (9) abnl(17p); (10) complex karyotype ≥ 3 abnormalities, and (11) monosomal karyotype. The log-normalized linear regression model revealed that t(9;11) (P = .0006, median, 1.9 years) was associated with a shorter and -5 or 5q− (P = .009; median, 9.3 years) with a prolonged latency period.

To identify an association between different chemotherapeutic agents and induction of specific cytogenetic abnormalities, we performed a multinomial regression analysis in which the outcome variable was categorized as follows: (1) t(15;17); (2) t(8;21); inv(16) or t(16;16); (3) t(9;11) or t(v;11)(v;q23); (4) NPM1; (5) normal karyotype excluding NPM1; (6) -5 or 5q−; -7; 7q−; abnl(17p); and (7) all other abnormalities. This model revealed that treatment with intercalating agents was significantly associated with the induction of cytogenetic abnormalities (P = .01), particularly t(9;11) or t(v;11)(v;q23) (P = .02), and the group all other abnormalities (P = .05).

Seventy-seven (3%) of the 2653 de novo AML patients had a history of previous neoplasm who did not receive chemotherapy and/or radiation. These patients were significantly older compared with all other patients with de novo AML (60 vs 53 years; P < .0001). All patients had previous solid cancer, commonly prostate cancer (n = 18, 23%), breast cancer (n = 8, 10%), gastrointestinal cancer (n = 8, 10%), as well as bladder cancer, renal cell carcinoma, and malignant melanoma (n = 7 each, 9%). The latency period to the occurrence of AML was 5.0 years (range, 1 day to 43.9 years). The cytogenetic profile of these cases showed in trend a greater frequency of adverse-risk abnormalities (19/67 [28%] vs 441/2314 [19%]; P = .06), in particular -5 or 5q− abnormalities (10/67 [15%] vs 177/2314 [8%], P = .04). OS of these patients was comparable with that of de novo AML patients (hazard ratio 1.04; P = .87).

Response to induction therapy for t-AML and de novo AML was as follows: CR 63% and 67% (P = .21), refractory disease 25% and 22% (P = .38), and early/hypoplastic death 12% and 9% (P = .20), respectively. In univariate as well as in multivariable analysis, type of AML did not impact the achievement of CR (P = .13, P = .62, respectively).

The median follow-up for survival in the entire cohort was 4.12 years (95% confidence interval [95%-CI] 3.97-4.28); the estimated 4-year RFS and OS were 38.5% (95%-CI 36.3%-40.9%) and 37.1% (95%-CI 35.2%-39.1%), respectively.

Outcome of patients with t-AML was significantly inferior: the 4-year RFS rates were 24.5% (95%-CI 17.7%-33.9%) and 39.5% (95%-CI 37.2%-42.0%; age-stratified log-rank test P < .0001) and the 4-year OS rates were 25.5% (95%-CI 19.6%-33.1%) and 37.9% (95%-CI 36.0%-40.0%; age-stratified log-rank test P = .001) for t-AML and de novo AML patients, respectively (Figure 3). Both greater CIR (age-stratified test, P = .01) and CID (age-stratified test, P = .002) contributed to the inferior outcome of t-AML patients. In t-AML patients previous therapy (radiation, chemotherapy, or both treatment modalities) and latency period had no impact on outcome.

Allogeneic hematopoietic stem cell transplantation (HSCT) in first CR was performed in 487 of 2064 (24%) patients ≤ 60 years and in 30 of 789 (4%) patients > 60 years of age. There was a significant greater proportion of younger patients receiving an allogeneic HSCT in first CR 40/89 (45%) in t-AML compared with de novo AML 410/1445 (28%; P = .002). Because dose intensity in postremission therapy and proportion of patients receiving allogeneic HSCT differed markedly between trials for patients ≤ 60 years versus those > 60 years, we performed subset analyses in these 2 age cohorts. Of note, in patients ≤ 60 years there was no statistically significant difference in CIR between t-AML and de novo AML (4-year CIR, 45.1% vs 46.3%; P = .63), whereas a marked difference was found in CID (4-year CID, 22.9% vs 8.6%; P < .0001; Figure 4A). The significantly greater CID rates were present in this age cohort regardless of the type of postremission therapy but were pronounced in patients who received allogeneic HSCT (Figure 5A-B). The 4-year CID rates in younger patients with t-AML and de novo AML after intensive chemotherapy were 12.3% (SE 5.4%) and 5.3% (SE 0.7%) and after allogeneic HSCT 35.8% (SE 8.3%) and 17.2% (SE 2.0%), respectively. Again no difference was present in CIR (Figure 5C-D). The distribution of causes of deaths in CR was comparable between t-AML and de novo AML and deaths mainly were caused by infections. The 4-year OS rates in patients, receiving allogeneic HSCT or other intensive consolidation therapy in first CR were 42.6% (95%-CI 29.0%-62.7%) and 37.5% (95%-CI 25.2%-55.8%) for t-AML as well as 58.0% (95%-CI 53.0%-63.6%) and 56.6% (95%-CI 53.5%-59.9%) for de novo AML, respectively. In contrast, t-AML patients > 60 years showed a significantly greater CIR (P < .0001), whereas there was no difference in CID (P = .44; Figure 4B).

Multivariable cause-specific Cox regression analyses on relapse and death in CR showed again a significant adverse impact of t-AML in younger patients for death in CR (P < .0001) but not for relapse (P = .39), whereas the contrary was the case in older patients with a significant adverse impact of t-AML on relapse (P = .02), but not on death in CR (P = .22). The Cox regression model on OS revealed t-AML as a poor prognostic factor in younger intensively treated patients (P = .01), but not in older less intensively treated patients (P = .34; Table 5).

Furthermore, we were interested in the effect of t-AML in specific genetically defined subsets: t(15;17); t(8;21); inv(16) or t(16;16); t(9;11); NPM1. In multivariable models adjusted for WBC, in patients exhibiting an inv(16) or t(16;16) t-AML was a significant adverse prognostic factor for OS (hazard ratio 2.35; P = .04).

Therapy-related AML is increasing in prevalence with greater life expectancy and improved survival of patients treated with chemotherapy and/or radiation for previous malignancies and other disorders. However, there is a paucity of prospective treatment data because these patients have often been excluded from clinical trials. The frequency of t-AML in our large cohort was 7%, which is comparable with previously reported data.^4,5 

Consistent with previous studies,^5,6,13  patients with t-AML more frequently had abnormal karyotypes compared with de novo AML (75% vs 51%); in particular, there was a high prevalence of adverse-risk cytogenetics. Among specific abnormalities, t(9;11), -5 or 5q−, -7, 7q−, abnl(17p), complex karyotypes, and the recently described monosomal karyotype category⁴⁰  were significantly overrepresented among t-AML (Table 3; Figures 1–2). For a large proportion of our cases, data on the mutational status of the NPM1 and FLT3 genes were available. Of note, the frequency of both NPM1 mutations and FLT3-ITD was significantly lower in t-AML, indicating that t-AML leukemogenesis in most cases follows mechanisms different from those seen in de novo AML. However, when we focused on patients with cytogenetically normal AML, no difference in the incidence and distribution of mutated NPM1 and FLT3-ITD between t-AML and de novo AML was evident (Table 6), which is consistent with previous reports.^7,41 

The median latency period between diagnosis of primary malignancy and occurrence of t-AML was 4 years, which is in line with published data.^4-6  Beyond the known association between treatment with anthracyclines, as the major compound of intercalating agents (Table 1), or the application of topoisomerase II inhibitors and a short latency period for the development of t-AML, we were able to show that younger age at diagnosis of primary malignancy also was associated with a shorter latency period. With respect to molecular markers, we did not detect an association of previous radiation to FLT3 mutations, which is in contrast to data from Christiansen et al.⁴² 

To evaluate whether the inferior prognosis of t-AML was attributable to an unfavorable genetic profile, or whether the variable “t-AML” itself predicted an inferior outcome, we performed multivariable analyses on the clinical end points response to induction therapy, RFS and OS. In these analyses, t-AML proved to be an adverse prognostic factor for RFS and OS but not for response to induction therapy. Of note, t-AML patients ≤ 60 years had a greater CID regardless of the applied therapy, whereas CIR was not different compared with de novo AML patients, likely reflecting cumulative toxicity of primary and secondary cancer therapy.^43,44  The 4-year transplant-related mortality of 38.5% in our study compares even favorably to the 48% mortality rate at 5 years reported by the Center for International Bone Marrow Transplant Research.⁴⁴ 

Intriguingly, results after induction therapy were not different between t-AML and de novo AML patients, pointing to the fact that dosage and modality of treatment during postremission therapy had a marked impact on the cumulative toxicity of cancer therapy. Therefore, intensive induction therapy should not be withheld for t-AML patients, and dose-reduced regimes for allogeneic HSCT should be considered.

In contrast, t-AML patients > 60 years showed a significantly greater CIR and no difference in CID. One reason for the greater CIR in older AML patients might be the lower dosage of applied chemotherapy during postremission therapy compared with younger patients.

In summary, our results add to previous knowledge that t-AML proves to be an adverse prognostic factor for RFS and OS, independent of other clinical and biologic variables. The inferior outcome, especially in intensively treated younger adult patients, was mainly attributable to an increased risk of death in CR, possibly reflecting cumulative toxicity of cancer treatment.

We are grateful to all members of the German-Austrian AML Study Group (AMLSG) for providing leukemia specimens and clinical data.

This work was supported in part by grants 01GI9981 (Network of Competence Acute and Chronic Leukemias) and 01KG0605 (IPD-Meta-Analysis: A model-based hierarchical prognostic system for adult patients with acute myeloid leukemia [AML]) from the Bundesministerium für Bildung und Forschung (BMBF), Germany.

Contribution: S.K. collected data, designed research, analyzed and interpreted data, and wrote the paper; K.D., J.K., C.-H.K., H.A.H., G.H., M.v.L.T., S.W., A.K., K.G., M.R., D.N., and A.G. provided study materials or patients and collected data; B.S., G.G., D.S., and C.M. collected data; M.Z. analyzed and interpreted data; H.D. and R.F.S. provided study materials or patients, collected data, designed research, analyzed and interpreted data, and wrote the paper; and all authors approved the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

A list of AMLSG institutions and investigators participating in this study appears in the supplemental Appendix (available on the Blood Web site; see the Supplemental Materials link at the top of the online article).

Correspondence: Richard F. Schlenk, MD, Department of Internal Medicine III, University Hospital of Ulm, Albert-Einstein-Allee 23, 89081 Ulm, Germany; e-mail: richard.schlenk@niklinik-ulm.de.