By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
This systematic review and meta-analysis compared the efficacy of different anthracyclines and anthracycline dosing schedules for induction therapy in acute myeloid leukaemia in children and adults younger than 60 years of age. Twenty-nine randomized controlled trials were eligible for inclusion in the review. Idarubicin (IDA), in comparison to daunorubicin (DNR), reduced remission failure rates (risk ratio (RR) 0·81; 95% confidence interval (CI), 0·66–0·99; P =0·04), but did not alter rates of early death or overall mortality. Superiority of IDA for remission induction was limited to studies with a DNR/IDA dose ratio <5 (ratio <5: RR 0·65; 95% CI, 0·51–0·81; P <0·001; ratio ≥5: RR 1·03; 95% CI, 0·91–1·16; P =0·63). Higher-dose DNR, compared to lower-dose DNR, was associated with reduced rates for remission failure (RR 0·75; 95% CI, 0·60–0·94; P =0·003) and overall mortality (RR 0·83; 95% CI, 0·75–0·93; P <0·001), but not for early death. Comparisons of several other anthracycline derivates did not reveal significant differences in outcomes. Survival estimates in adults suggest that both high-dose DNR (90 mg/m2 daily × 3 or 50 mg/m2 daily × 5) and IDA (12 mg/m2 daily × 3) can achieve 5-year survival rates of between 40 and 50 percent.
Acute myeloid leukaemia (AML) is a heterogeneous group of haematopoietic malignancies characterized by the proliferation of abnormal leukaemic blast cells of the myeloid lineage and impaired production of normal blood cells (Bonnet & Dick, 1997). The current standard treatment for induction of complete remission (CR) in previously untreated children and adults with AML is a combination of an anthracycline and cytarabine with or without addition of a third agent, such as etoposide (Pui, 2006; National Cancer Institute, 2012). The widely used intravenous combination of daunorubicin (DNR; at a dose of 45 mg/m2) given daily for 3 d, and cytarabine (100 mg/m2) given daily for 7 d, results in CR in 50–75% of adult patients (Wahlin et al, 1991; Vogler et al, 1992; Wiernik et al, 1992). Remission rates in children are somewhat higher as compared to adults (Creutzig et al, 2001).
In the late 1980s, idarubicin (IDA) was introduced and 3 randomized controlled trials (RCTs) reported higher CR rates for IDA as compared to DNR (Berman et al, 1991; Vogler et al, 1992; Wiernik et al, 1992). An advantage in long-term survival was only observed in one study (Berman et al, 1997). The anthraquinone derivate mitoxantrone (MTZ) has also been extensively used with cytarabine as part of effective induction regimens (Arlin et al, 1990; Wahlin et al, 1991; Pavlovsky et al, 1994). Several other anthracyclines, such as doxorubicin, aclarubicin (ACR), or rubidazone, have also been tested in AML induction therapy strategies, either as part of RCTs or in single arm studies (Yates et al, 1982; Morrison et al, 1992; Nagura et al, 1994). However, none of those drugs was found to be consistently superior when compared to other anthracyclines. More recently, it was recognized that increasing the anthracycline dose could improve CR rates and other outcomes (Fernandez et al, 2009; Lowenberg et al, 2009; Lee et al, 2011).
Taken together, anthracyclines have been a cornerstone in AML therapy for more than three decades. Although numerous RCTs have been conducted, there is no widely accepted consensus for the optimal formulation, dose or schedule of this important drug class for remission induction strategies in patients with AML. In 1998, an excellent systematic collaborative overview of randomized trials compared different anthracyclines as induction therapy for AML (Wheatley, 1998). The authors concluded at that time that induction regimens based on IDA achieved better remission rates and a better overall survival than those based on DNR. However, in the meantime, several additional RCTs have been published, which may limit the findings of this review. A more recent meta-analysis evaluated the efficacy of different AML induction regimens but specifically focussed on the elderly population (>60 years of age) (Ziogas et al, 2011).
Thus, no study has quantitatively synthesized the current evidence regarding the optimal formulation or dosage of anthracyclines during induction therapy for AML in children and adult patients younger than 60 years of age. To arrive at comprehensive estimates of efficacy from all randomized trials conducted to date, we undertook a systematic literature review and meta-analysis of all RCTs comparing: (i) Different anthracyclines during induction therapy in AML, and/or (ii) Different dosing or schedules for the same anthracycline during induction therapy in AML.
We followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement for reporting our results (Moher et al, 2009).
Data sources and searches
We performed an electronic search of OVID MEDLINE (from 1946 to January 2012), EMBASE (from 1980 to January 2012), and The Cochrane Central Register of Controlled Trials (CENTRAL; until the fourth quarter of 2011) (detailed search strategy provided in Supplemental Data A). The search was restricted to articles published in English language. We also searched relevant references and conference proceedings from 2010 to 2012 using the Web of Science.
We included all RCTs that included the following comparisons during induction therapy in AML: (i) Different anthracyclines, and/or (ii) Different dosing or schedules for the same anthracycline, in which the baseline chemotherapy regimen was the same. Mitoxantrone, an anthracenedione derivate that is structurally related to the anthracycline derivates, was also considered in this study. There was no restriction by study site/country, quality of the study, or follow-up period. Further, there was no restriction by dose, frequency, method of drug administration, length of induction therapy, or concurrent chemotherapy. Studies designed to evaluate treatment effect specifically in the elderly (age ≥60 years) were excluded (trials with a median age ≥60 years were included as long as stratified outcome data for patients <60 years could be extracted). We also excluded studies designed to evaluate treatment effect specifically in patients with myelodysplastic syndrome (MDS; threshold for exclusion ≥50% of patients had MDS).
Two authors inspected the abstract of each reference identified by the search and applied the inclusion criteria. For possibly relevant articles, the full article was obtained and reviewed by the same two authors independently. Final inclusion of studies was determined by agreement of both reviewers; disagreements were resolved by consensus or involvement of a third author if consensus could not be achieved. Agreement between reviewers was evaluated by using a kappa statistic (Landis & Koch, 1977).
Data extraction and quality assessment
Two authors independently extracted data from included trials. Data extraction was performed using a standardized data collection form. When missing data for the primary outcome measure were encountered, the corresponding authors were contacted to retrieve these data.
The primary outcome measure was failure to achieve CR by end of induction (time point defined by each study). Secondary outcome measures were: (i) cumulative event rate (the first event was defined as death from any cause, leukaemia progression/relapse or non-response, or second malignancy), (ii) cumulative relapse rate, (iii) death rate during induction (time point defined by each study), and (iv) cumulative death rate. If possible, event rate, relapse rate, and cumulative death rate were determined at 5 years from diagnosis or treatment initiation. If only other time points were reported, these were used as long as the time frame was a minimum of 3 years. If no clear time point was reported for these outcomes, a minimum median follow-up of 2 years was required for data extraction. Data were extracted as proportions if results were only reported as probability of remission rate, events, relapse or mortality, and if the number of events was not explicitly provided.
To assess methodological quality and risk of bias, included articles were independently examined by two authors for: (i) sequence generation, (ii) allocation concealment, (iii) blinding, (iv) incomplete outcome data, and (v) intention-to-treat (ITT) analysis (definitions/criteria were derived from Higgins & Green, 2008).
Data synthesis and analysis
Outcomes with at least two eligible studies were synthesized; synthesis was performed using Review Manager (version 5·0). We performed per protocol (PP) analysis and determined risk ratios (RR) with 95% confidence intervals (CI) for dichotomous data (Mantel–Haenszel method). P-values<0·05 were considered statistically significant.
We considered sub-group analyses to investigate the effects of age, French-American-British (FAB) subgroup, and (cytogenetic) risk group. Subgroup analysis required a minimum of two studies in each group. We also performed post-hoc sub-group analyses for the primary outcome measure (remission failure) to investigate the effect of the DNR/IDA dose ratio (<5 vs. ≥5). Finally, we performed best-worst case analyses to account for the issue of missing data.
Because we anticipated heterogeneity between studies, a random-effects model was used for all analyses. Statistical heterogeneity was initially inspected graphically (forest plot) and assessed by calculating tests of heterogeneity using the Cochran Q test (Chi-square test). We quantified the degree of heterogeneity using the I2 statistic (Borenstein et al, 2009). We considered investigating publication bias using a funnel plot, in which the standard error of the effect estimate of each study was plotted against the estimate (Sutton et al, 2000). An asymmetric plot suggests possible publication bias (Higgins & Green, 2008).
Figure 1 illustrates the flow diagram of trial identification and selection. A total of 7136 titles and abstracts were assessed, and 80 full articles were retrieved. Of these, 29 satisfied the eligibility criteria and were included in the review. The reviewers had excellent agreement on articles for inclusion, with a kappa statistic of 0·95 (95% CI 0·87–1·00). The reasons for excluding 51 articles are shown in Fig 1. However, subgroup information (children) from one duplicate study excluded from the review was retrieved for analysis of the impact of year of publication and anthracycline dosing on overall survival rates (see below) (Gibson et al, 2011).
Characteristics of included studies are presented in Table 1. DNR versus IDA was reported in eight trials (Eridani et al, 1989; Berman et al, 1991; Vogler et al, 1992; Wiernik et al, 1992; Masaoka et al, 1996; Creutzig et al, 2001; Mandelli et al, 2009; Ohtake et al, 2011), DNR versus MTZ in 6 trials (Arlin et al, 1990; Wahlin et al, 1991; Pavlovsky et al, 1994; Nikanfar et al, 2007; Mandelli et al, 2009; Burnett et al, 2010), IDA versus MTZ in 3 trials (Beksac et al, 1998; De Moerloose et al (2011), Mandelli et al, 2009) and DNR versus ACR was reported in two trials (Nagura et al, 1994; de Nully Brown et al, 1997). Another 10 anthracycline comparisons were evaluated in single studies as outlined in Table 1 (Paul et al, 1981, 1991; Yates et al, 1982; Morrison et al, 1992; Harousseau et al, 1996; Pignon et al, 1996; Takemoto & Ogawa, 1998; Intragumtornchai et al, 1999; Creutzig et al, 2010). High-dose versus standard-dose DNR was investigated in three studies (Yates et al, 1982; Fernandez et al, 2009; Lee et al, 2011) and one study compared continuous administration with bolus administration of MTZ (Koc et al, 2004). The majority of studies considered a second course of induction therapy for patients who did not achieve remission after the first course (schedule and dose intensity were altered in the second course for some studies).
Risk of bias assessment was conducted for all 29 studies. Considering all trials, allocation generation and concealment information were available for 5 and 4 trials, respectively. Blinding status was not reported in any study. Withdrawal information could be retrieved from 23, and ITT analysis was reported for 12 trials (Supplemental Data B).
IDA versus DNR
When weighted data from eight studies were synthesized (3382 patients), remission failure rate for IDA was superior to DNR (RR 0·81; 95% CI, 0·66–0·99; P =0·04). There was substantial heterogeneity between these studies (I2 = 56%; P =0·03), indicating differences in the effect of IDA relative to DNR between included trials. Stratified analysis (DNR/IDA dose ratios <5 vs. ≥5) was performed post-hoc to explore this heterogeneity. Superiority of IDA was only observed in studies with a higher relative IDA dose as defined by a DNR/IDA dose ratio <5 (ratio <5: RR 0·65; 95% CI, 0·51–0·81; P <0·001; in contrast to ratio ≥5: RR 1·03; 95% CI, 0·91–1·16; P =0·63) (Fig 2 and Table 2). The difference between these two subgroups was statistically significant (P < 0·001).
Table 2. Synthesized risk ratios for study comparisons
Outcome or Subgroup
RR [95% CI]
DNR, Daunorubicin; IDA, Idarubicin; MTZ, Mitoxantrone; ACR, Aclarubicin; HD indicates higher DNR dose (90 or 45 mg/m2) and SD indicates standard dose (45 or 30 mg/m2); RR, risk ratio; CI, confidence interval.
Numbers written in bold refer to statistically significant differences.
Associated forest plots are provided in Supplemental Data C.
IDA versus DNR
0·81 [0·66, 0·99]
DNR/IDA dose ratio <5
0·65 [0·51, 0·81]
DNR/IDA dose ratio ≥5
1·03 [0·91, 1·16]
Death during induction
1·23 [0·89, 1·69]
0·90 [0·80, 1·02]
DNR versus MTZ
1·06 [0·95, 1·18]
Death during induction
0·92 [0·75, 1·13]
1·04 [0·99, 1·10]
IDA versus MTZ
1·10 [0·95, 1·27]
Death during induction
1·07 [0·82, 1·38]
0·91 [0·72, 1·16]
0·98 [0·92, 1·05]
DNR versus ACR
1·06 [0·57, 1·97]
1·03 [0·95, 1·13]
HD- versus SD DNR
0·75 [0·60, 0·94]
Death during induction
0·99 [0·76, 1·30]
0·83 [0·75, 0·93]
Overall mortality by cytogenetic groups
0·85 [0·77, 0·95]
0·83 [0·57, 1·22]
0·76 [0·65, 0·90]
0·93 [0·80, 1·08]
There was no significant difference between IDA and DNR in regard to death during induction (6 trials; 3205 patients: RR 1·23; 95% CI, 0·89–1·69; P =0·21) and overall mortality (4 trials; 2973 patients: RR 0·90; 95% CI, 0·80–1·02; P =0·11) (Table 2). However, Fig 3 illustrates that there qualitatively appeared to be a larger benefit for IDA in terms of a reduction in overall mortality, as the DNR/IDA dose ratio decreased.
DNR versus MTZ, IDA versus MTZ, and DNR versus ACR
Rates for remission failure (6 trials: DNR versus MTZ; 3 trials: IDA versus MTZ; 2 trials: DNR versus ACR), death during induction (5 trials: DNR versus MTZ: 2 trials: IDA versus MTZ), relapse (2 trials: DNR versus ACR), and overall mortality (2 trials: DNR versus MTZ; 3 trials: IDA versus MTZ; 2 trials: DNR versus ACR) did not differ between the groups for any outcome measure (Table 2).
High-dose versus standard-dose DNR
A total of three studies provided information to compare different DNR dose intensity regiments. Both remission failure rate (1404 patients; RR 0·75; 95% CI, 0·60–0·94; P =0·003) and overall mortality rate (1040 patients; RR 0·83; 95% CI, 0·75–0·93; P <0·001) were lower in the group of patients treated with relatively higher doses of DNR (cumulative dose per cycle: 135–270 mg/m2 vs. 90–135 mg/m2). There was no difference in death during induction between high-dose- and standard dose groups (1021 patients; RR 0·99; 95% CI, 0·76–1·30; P =0·97) (Table 2).
Results of the stratified analysis according to cytogenetic risk groups are shown in Fig 4. No significant subgroup differences were detected between low-, standard-, and high-risk groups (I2 = 36·5%; P = 0·21).
Figure 5 illustrates the impact of year of publication and anthracycline dosing (Fig 5A: DNR; Fig 5B: IDA) on absolute overall survival rates. All trials reporting overall survival data for DNR and IDA (irrespective of the randomized comparator) were included in this graph, including two recently published paediatric trials that were not eligible for meta-analysis (study 1: paediatric data of a larger trial included in meta-analysis; study 2: comparison of liposomal DNR versus IDA) (Creutzig et al, 2010; Gibson et al, 2011). First, overall survival rates improved substantially during the last two decades, achieving 40–50% in some trials. Second, survival estimates in adults suggest that both high-dose DNR (90 mg/m2 daily × 3 or 50 mg/m2 daily × 5) and IDA (12 mg/m2 daily × 3) are associated with similar 5-year survival. Third, data extracted from paediatric trials indicated relatively favourable survival rates in children as compared to adults with a lower dose of DNR (50–60 mg/m2 daily × 3) and a comparable dose of IDA (12 mg/m2 daily × 3).
There were not enough data available to perform sub-group analyses to investigate the effects of age, FAB subtype, and cytogenetic risk group (other than for DNR dose intensity; see above) for any outcome measure. Best-worst case analyses (remission failure rate) did not significantly impact on the results (data not shown).
Given that all analyses included less than 10 studies each, tests for funnel plot asymmetry were not performed (Higgins & Green, 2008).
We made several important observations in this meta-analysis. First, we found consistent data indicating that dose intensification of DNR is more efficacious for induction therapy in AML compared with lower dose DNR. Second, IDA is better than DNR when the DNR dose is relatively low, but the two anthracycline formulations have similar efficacy when the DNR dose is higher, as defined by a DNR/IDA dose ratio ≥5. Our meta-analysis is important because there has been uncertainty regarding the optimal formulation and dose of anthracyclines during induction therapy for newly diagnosed AML in children and adults younger than 60 years.
Eight trials compared IDA and DNR. We anticipated similar effects on remission failure rates of the included trials before initiating data synthesis. Accordingly, we pooled the weighted estimates of each study. However, initial assessment revealed substantial heterogeneity between the studies. Our analysis suggests that this heterogeneity is largely related to different DNR/IDA dose ratios applied in the different trials, with greater efficacy of IDA with lower relative DNR dosing, indicating that the equivalent anti-leukaemic activity for DNR/IDA is equal or higher than a 5:1 dose ratio (e.g. 60 mg/m2 DNR and 12 mg/m2 IDA, respectively). There is a paucity of literature to support a definite isotoxic/-efficacious dose conversion between anthracyclines; however, a DNR/IDA conversion factor ≥5 is consistent with the Children's Oncology Group ‘Long-Term-Follow-Up Guidelines’ for paediatric cancer survivors (Children's Oncology Group's, 2008). Moreover, our observation may explain why studies conducted in the late 1980s and 1990s reported superior outcome data for IDA (Wheatley, 1998). In those trials, the DNR/IDA dose ratio was exclusively lower than 5:1. While our study was able to provide further information in regard to relative DNR and IDA doses, the impact of absolute doses is less certain. We attempted to address the impact of absolute DNR and IDR doses graphically, although it is important to state that other factors, such as patient selection or concomitant chemotherapy, may differ between the studies. Thus, conclusions about the effect of absolute doses of anthracyclines must be drawn very carefully.
In addition to DNR versus IDA, numerous RCTs were conducted to compare anthracycline derivates. Neither, DNR (standard-dose) versus MTZ, IDA versus MTZ, nor DNR versus ACR revealed compelling differences in terms of superior outcome for one or the other drug. Nine other trials investigated other combinations, all published in the 1980s and 1990s. Their findings do not seem to have a major impact on current clinical practice. Given the evaluable evidence, it is difficult to summarize the role of anthracyclines other than DNR and IDA. Ongoing and future trials will need to determine whether MTZ or other anthracycline derivatives are superior to DNR and IDA.
Interestingly, the interim data of an ongoing paediatric trial (AML-BFM 2004) revealed a 5-year overall survival rate of 78% for children receiving liposomal DNR (L-DNR) without increased toxicity (Creutzig et al, 2010). Given the reduced toxicity of L-DNR and a trend towards better survival rates as compared to IDA, L-DNR will be further used in the forthcoming AML-BFM study. However, whether these findings can be confirmed in other paediatric trials or in the adult setting remains unknown.
Yates et al (1982) published the first study showing that DNR dose intensification resulted in better outcomes (45 mg/m2 vs. 30 mg/m2 daily × 3). Two decades later, two large RCTs evaluated the efficacy of further dose intensification of DNR (Fernandez et al, 2009; Lee et al, 2011). This dose intensification of DNR resulted in significant improvements in the rates of CR and overall survival. Subgroup analysis (cytogenetic risk groups) for overall survival was hampered by a limited number of studies. However, it remains unclear to whether the high-dose strategy can substantially improve overall survival among patients in the high-risk group. New treatment approaches are clearly needed for patients with poor-risk cytogenetic features.
While our review suggested that higher dose DNR is more efficacious than lower dose DNR, a critical limitation of the studies included in the review was a failure to report long-term outcomes. The major downside for higher dose DNR will be cardiac toxicities and second malignant neoplasms (SMN) (Hijiya et al, 2009; van der Pal et al, 2012). It is well recognized that the risk of cardiotoxicity depends on the cumulative anthracycline dose (van der Pal et al, 2012). Whether the shorter term benefits of higher dose DNR will be offset by future cardiac toxicities is unknown. It is also not known whether this relationship may be influenced by administration of cardioprotectants, such as dezrazoxane, and whether the risk to benefit ratio may differ between children and adults (van Dalen et al, 2011). Both studies evaluating DNR dose of 270 mg/m2 per cycle were limited to patients aged between 18 and 60 years. In our review, the highest cumulative DNR dose given during an induction cycle in a paediatric RCT was 180 mg/m2 (Creutzig et al, 2001). Thus, the optimal approach in children is unknown.
Our systematic review has three important limitations. First, meta-regression would have been an optimal approach to better examine potential confounders at the study level (i.e. concomitant chemotherapy or consolidation therapy), and to further analyse the dose-response relationship between DNR/IDA dosing and clinical outcomes. However, there were too few studies to consider such an approach. Second, a meta-analysis of individual patient data would have been a way to better adjust for patient level covariates, such as patient age, FAB subgroup or cytogenetic subgroup. However, as long as the anthracycline randomization stratified for these factors, individual patient level meta-analysis may not offer substantial advantages but would be greatly onerous in terms of costs and time. Finally, outcomes in AML are known to have improved over time related to improved supportive care (Pulte et al, 2009). It is possible that there is an interaction between anthracycline effect and efficacy of supportive care although empiric evidence to support such a relationship is lacking.
Future research should definitely consider an individual patient meta-analysis to better understand the impact of age and leukaemia characteristics on clinical outcomes in patients who receive an anthracycline during induction therapy. Such an analysis should be performed with mature study data that captures late events, such as cardiac toxicity and SMN. Finally, a decision-analytical model may be beneficial. Specifically, decision making for AML induction therapy is difficult because it involves trade-offs between benefits (CR and survival rates) and risks (toxicity and SMN). A state transition model (Markov model), for example, may be particularly useful, as such an approach can incorporate the effect of different anthracyclines or anthracycline dosing schedules, as well as the effect of cardioprotectants, such as dezrazoxane, to determine an overall optimal treatment strategy.
Lillian Sung is supported by a New Investigator Award from the Canadian Institutes of Health Research (Grant No. 87719). All authors made substantial contributions to the research design and acquisition, analysis and interpretation of the data, drafting and critically version the manuscript as well as approval of the final version of the manuscript being submitted.