Allogeneic Stem Cell Transplantation for Acute Myeloid Leukemia in First Complete Remission:  Systematic Review and Meta-analysis of Prospective Clinical Trials FREE

Achieving a cure for acute myeloid leukemia (AML), even for younger adult patients with de novo AML, remains a challenge. While more than 70% of such patients will enter a first complete remission (CR1) after induction chemotherapy, a substantial number experience disease relapse.¹ Allogeneic stem cell transplantation (SCT) after myeloablative conditioning is a curative treatment option for younger patients with AML in CR1. However, concerns regarding allogeneic SCT–related toxicity, and questions regarding its benefit, limit its use for patients who have attained an initial remission. Alternative therapies include intensive consolidation chemotherapy or autologous SCT. The current consensus, reflected in treatment guidelines of the National Comprehensive Cancer Network (V1.2009: available at http://www.nccn.org), is based on cytogenetic stratification into good-, intermediate-, and poor-risk AML. Patients with good-risk AML in CR1 are recommended to undergo consolidation chemotherapy, with autologous SCT considered an acceptable alternative. Patients with poor-risk AML in CR1 are recommended to undergo allogeneic SCT. There is no preferred therapy for patients with intermediate-risk AML in CR1; allogeneic SCT, consolidation chemotherapy, and autologous SCT are considered of equivalent benefit.

Multiple prospective trials have been undertaken to clarify the role of allogeneic SCT for AML in CR1. In the context of allogeneic SCT trial design, treatment assignment has typically been based on donor availability: patients with HLA-matched siblings are assigned to undergo allogeneic SCT (donor group), and those without matched siblings (or without siblings) are assigned to nonallogeneic SCT therapy (no-donor group). Although not randomized comparisons, these studies have nevertheless been widely accepted as providing good-quality evidence of treatment effect, because no evidence of major bias arising from differences in biological and socioeconomic factors has been identified.

Various prospective clinical trials, retrospective studies, and systematic reviews have helped determine the current treatment consensus for AML in CR1. Retrospective analyses are prone to errors of bias and confounding and may therefore provide inaccurate estimates of effect. Prospective biological assignment trials offer a means of reducing such errors. However, their results have thus far not provided definitive evidence to support treatment recommendations. While some individual trials have documented superior relapse-free survival (RFS), none has documented an overall survival benefit for allogeneic SCT across all cytogenetic risk groups. Within cytogenetic risk groups, the evidence regarding allogeneic SCT impact is also limited, as discussed in this article.

To arrive at comprehensive estimates of RFS and overall survival benefit from the totality of the clinical trial data available, we undertook a systematic literature review and meta-analysis of all prospective biological assignment clinical trials of allogeneic SCT vs consolidation chemotherapy, autologous SCT, or both for AML in CR1, on an intention-to-treat (ITT) donor vs no-donor basis.

We undertook searches of the MEDLINE (PubMed), Embase, and Cochrane Registry of Controlled Trials databases (updated March 2009), combining the search terms allogeneic; acut* and leukem*/leukaem*/leucem*/leucaem*/aml; myelo* or nonlympho*. MEDLINE and Embase searches were restricted to human studies. Studies identified underwent title and abstract review (J.K., C.C.), and clearly nonrelevant articles were discarded. Text review of the remainder was performed to assess their suitability. The bibliographies of retained articles were examined to identify additional studies. The abstracts of relevant scientific meetings were similarly examined to ensure complete review of the available data. International expert input was obtained to identify additional relevant trials, including those in non–English-speaking countries. Recent reviews and meta-analyses were also accessed to identify additional studies that met inclusion criteria.^2- 7

Studies included were prospective trials of adults (wholly or predominantly) with AML in CR1 that assigned participants to undergo allogeneic SCT vs a comparator of consolidation chemotherapy, autologous SCT, or both. Eligible trials reported hazard ratios (HRs) and 95% confidence intervals (CIs) for overall survival benefit, RFS benefit, or both on an ITT donor vs no-donor basis (or provided data to estimate HR by the method of Parmar et al⁸). When multiple articles reported on the study, the most updated data were analyzed. Unadjusted HRs were preferred in the analysis, because adjusted HRs, reported in a minority of studies, were considered likely to adjust for different covariates from study to study, potentially impeding analysis across studies. Furthermore, prospective biological treatment assignment was considered likely to equalize covariates over the large number of patients analyzed. Adjusted HRs, if reported, were used in sensitivity analyses.

The data were abstracted in a standardized format by 2 independent reviewers (J.K., C.C.). The data collected for each study included study name, study first author, publication year, period of enrollment, total number allocated to undergo therapy, number assigned to the donor and no-donor groups on an ITT basis, median patient age (years), median duration of follow-up (months), number of events (death, relapse) in each group, and study end points of overall survival benefit, RFS benefit, or both. We used overall survival and RFS (also reported as disease-free survival) as per the individual studies. Data on treatment-related mortality (also reported as nonrelapse mortality) were collected. We also collected data on therapy, ie, induction therapy regimen, interim therapy regimen (if any), stem cell source (bone marrow or peripheral blood), allogeneic SCT conditioning, autologous SCT conditioning, and consolidation chemotherapy regimen. Discrepancies in data extraction were to be resolved by consensus, referring back to the original article, and by contacting the study authors if necessary. When missing data were encountered, the primary authors were contacted to complete the data analysis.

We assessed for quality based on the requirement for prospective treatment assignment, the reporting of outcomes on an ITT basis, the study size, the number of participating centers, the adequacy of induction chemotherapy (treatment regimen; percentage of patients entering CR1), and the proportion of patients allocated to undergo allogeneic SCT who underwent therapy as assigned (Table 1, Table 2, and Table 3). Given the unambiguous end points (overall survival, RFS) and study treatments, we did not anticipate any impact of lack of blinding on outcomes. We did not explicitly score the methodologic quality of the included trials, because the value of doing so is controversial. Ad hoc scores may lack demonstrated validity, and results may not be associated with quality.^27- 30 Instead, we performed subgroup and sensitivity analyses and undertook tests of interaction, as is widely recommended.^29- 31

Data analysis was performed using STATA version 7 (StataCorp, College Station, Texas). The threshold of significance was P ≤ .05. Begg funnel plots and P values were used to investigate publication bias.³² Heterogeneity was assessed by a Q statistic (P value) and I² index.^33- 34 A forest plot with combined HRs (with 95% CIs) for RFS and overall survival benefit of allogeneic SCT (donor) vs nonallogeneic SCT (no donor) was constructed using fixed-effects meta-analysis, because we felt it was appropriate to maintain a direct relationship between sample size and relative weights, particularly for the smaller studies. In sensitivity analyses, random-effects meta-analysis of DerSimonian and Laird was also undertaken.³⁵

We explored our findings further by additional sensitivity analyses. To assess the potential impact of missing overall survival data from studies reporting only RFS outcomes vs those reporting overall survival as well as RFS end points, we looked for systematic differences in RFS outcomes between the 2 groups. We also evaluated the impact of including additional trials that stratified treatment options by cytogenetic risk (ie, restricting allogeneic SCT option to intermediate-risk AML, poor-risk AML, or both) to the initial analysis. In subgroup analyses, we assessed overall survival and RFS benefit for the cytogenetic risk subgroups: poor-, intermediate-, and good-risk AML. Tests of interaction across the subgroups were performed to assess whether benefit of allogeneic SCT varied significantly between the cytogenetic risk categories. Random-effects meta-analysis was also undertaken to assess the robustness of all survival end points. Adjusted HRs were used in additional sensitivity analyses.

This work was performed in accordance with the Quality of Reporting of Meta-analyses (QUOROM) guidelines for meta-analysis of randomized clinical trials.²⁹

The initial online databases and abstract search identified 1712 articles (Figure 1). After screening titles and abstracts, 1660 nonrelevant articles were excluded. The remaining 52 articles were retrieved for further review. They were reviewed independently in a structured format, and 23 articles were discarded because they did not prospectively compare myeloablative allogeneic SCT vs nonallogeneic SCT options for adult patients with AML in CR1 on an ITT basis, were not assigned treatment trials, reported noncomparable patient cohorts, or represented repeat publications of the same trial. An additional relevant trial was identified by expert input. Recent review articles and meta-analyses were also retrieved.^2- 4 These did not yield additional relevant trials.

The search identified 30 potentially relevant trials that evaluated allogeneic SCT vs nonallogeneic SCT therapies (consolidation chemotherapy, autologous SCT, or both) for AML in CR1. Six trials did not report outcomes based on treatment assigned, and their non-ITT data were not further evaluated. Twenty-four trials provided prospective data on overall survival outcomes, RFS outcomes, or both that were extractable on an ITT donor vs no-donor basis.^9- 26 These studies were included in the analysis (Tables 1, 2, and 3). Eighteen trials reported RFS outcomes across all AML cytogenetic risk categories; 15 reported overall survival outcomes across all AML cytogenetic risk categories. Six trials restricted the allogeneic SCT option to intermediate-risk AML in CR1, poor-risk AML in CR1, or both; their cytogenetic risk–stratified overall survival/RFS outcomes are included in sensitivity and subgroup analyses. No major discrepancies were noted between reviewers regarding trial inclusion or data extraction. The few minor discrepancies in extraction of study clinical characteristics were resolved by consensus after referring to the study reports.

Factors considered significant in terms of overall study quality included (1) Adequate sequence of allocation (the method of generation of sequence of random assignment). Biological treatment assignment was the typical method of “randomization” across studies of AML in CR1. (2) Allocation concealment (concealment of treatment assignment from those involved in care until after assignment occurred). This was considered nonrelevant for these prospective studies, because treatment assignment was automatic based on presence or absence of an HLA-matched donor, and HLA typing necessarily had to be reported early, to make logistical arrangements for allogeneic SCT vs nonallogeneic SCT therapies. (3) Blinding (of patients, physicians, data collectors). Blinding was considered less relevant to these trials, given the unambiguous treatments assigned and the “hard” end points (overall survival, RFS) measured. (4) Dropouts/follow-ups. Dropouts were considered likely to bias against overall survival benefit, because the allogeneic SCT groups were anticipated to have lower rates of patients undergoing therapy as assigned. Follow-up data were documented. (5) Selective reporting (including publication bias). We made extensive efforts to minimize publication bias with a broad literature search, seeking expert input (including from non–English-speaking countries), abstract review, and search of recent review articles. Data from unpublished trials (AMLHD98A, JALSG AML97) are included in this analysis. Statistical assessment for possible publication bias was performed. (6) Other (study-specific). We have highlighted study-specific issues (Tables 1, 2, and 3), and restricted the primary RFS and overall survival analysis to studies involving “all comers” (ie, offering treatment assignment to allogeneic SCT vs nonallogeneic SCT therapies for all AML cytogenetic risk groups).

Overall, the studies in the analysis were considered of good quality, typically prospective multicenter trials that reported outcomes on a donor vs no-donor basis analyzed as ITT; were performed at the national level in the United States, Europe, and Japan; and were published in respected peer-reviewed journals. They enrolled patients between 1982-2006. Numbers of patients in the allogeneic SCT and nonallogeneic SCT groups ranged from 58 to 1305. Some studies combined individual patient data across multiple trials and reported aggregate survival end points. Eligible patients typically comprised adults with newly diagnosed AML who were younger than 40 to 60 years, with adequate organ function and absence of significant concomitant disease (Table 3). Two trials included a minority population of pediatric patients (16% and 21%, respectively).^14,21 Allogeneic SCT treatment adherence was reasonable for most studies, with only 1 trial reporting less than 60% compliance. One study (Eastern Cooperative Oncology Group [ECOG] EST3483), reported solely in summary form with missing data on several parameters, was considered marginal.¹⁰ In a sensitivity analysis, removal of this study did not affect combined estimates of allogeneic SCT benefit. A subset of the studies reported survival outcomes stratified by AML cytogenetic risk. The cytogenetic criteria used in different studies (eg, Southwest Oncology Group [SWOG]/ECOG, Medical Research Council [MRC], European Organization for Research and Treatment of Cancer [EORTC]/Gruppo Italiano Malattie Ematologiche Maligne dell’Adulto [GIMEMA] cytogenetic risk classification) are substantially similar but differ in some minor respects.³⁶ Studies using nonstandard risk stratification were also analyzed by cytogenetic risk (SWOG/ECOG; MRC).^23- 24,26

The studies typically assigned patients on the basis of the availability of a HLA-matched sibling donor for treatment allocation to the allogeneic SCT group. The comparator group typically comprised autologous SCT, consolidation chemotherapy, or both. If both nonallogeneic SCT alternatives were offered, randomization between the nonallogeneic SCT groups was often performed later, introducing potential bias because higher-risk patients experiencing early relapse may not be randomized between their nonallogeneic SCT therapies. To address this potential bias, a donor vs no-donor comparison was undertaken, based solely on initial assignment to the allogeneic SCT or nonallogeneic SCT groups.

We could not assess the quality or completeness of sibling HLA testing or exclude patients listed as having no siblings from this analysis, understanding that the inclusion of such patients may introduce a bias into the treatment comparison.³⁷ However, the trials that expressly permitted allogeneic SCT for patients lacking HLA-matched sibling donors did not report significant differences between sibling and nonsibling donor outcomes.²⁵ This suggests a lack of systematic bias between outcomes in patients with AML who underwent transplantation and who had a HLA-matched sibling vs those who did not have a HLA-matched sibling.

The clinical trials also varied with respect to trial design and therapeutic interventions, with differences in induction chemotherapy (and the possibility of reinduction), interim chemotherapy, and consolidation chemotherapy (where applicable). Autologous SCT commonly involved myeloablative conditioning (usually identical to that used for allogeneic SCT) and autologous bone marrow infusion (some studies used peripheral blood stem cells). Allogeneic SCT comprised myeloablative conditioning (with or without radiation) followed by infusion of allogeneic donor bone marrow or peripheral blood stem cells (unmanipulated or variably T-cell depleted), with graft-vs-host disease prophylaxis often comprising cyclosporine and methotrexate. Importantly however, despite the variability in patient eligibility, trial design, and study interventions, interstudy heterogeneity for the overall survival or RFS end points was not significant.

We subsequently undertook detailed quantitative assessments of the relevant studies. Additional sensitivity and subgroup analysis were also undertaken and are described in detail.

We constructed Begg funnel plots to evaluate for publication bias. For RFS benefit, the plots tended to maintain a symmetric distribution, both for the primary analysis of 18 trials reporting RFS outcomes in patients with AML across all cytogenetic risk groups (P = .50) and for the analysis that included 6 additional trials with allogeneic SCT restricted to intermediate-risk patients with AML, poor-risk patients with AML, or both (P > .99). For overall survival benefit, the plots also tended to maintain a symmetric distribution, for the primary analysis of 15 trials reporting overall survival outcomes for all patients at cytogenetic risk for AML (P = .28) as well as for the analysis that included 6 additional trials with allogeneic SCT restricted to intermediate-risk patients with AML, poor-risk patients with AML, or both (P = .62).

Eighteen clinical trials reported end points of overall RFS across all cytogenetic risk groups. The summary hazard estimate for overall RFS benefit also varied between studies, ranging from 0.50 (donor group better) to 1.56 (no-donor group better). Interstudy heterogeneity was nonsignificant (P = .45), with I² = 0. In a fixed-effects forest plot, the combined HR for overall RFS benefit with allogeneic SCT was 0.80 (95% CI, 0.74-0.86) (Figure 2). The combined estimate indicates statistically significant reduction in hazard of death or AML relapse with allogeneic SCT in CR1, across all cytogenetic risk groups (P < .01).

We further evaluated the studies with sensitivity and subgroup analyses (Figure 2). In a sensitivity analysis, we repeated the initial RFS analysis, including 6 additional trials that restricted allogeneic SCT option for intermediate-risk patients with AML, poor-risk patients with AML, or both. The combined HR was 0.78 (95% CI, 0.73-0.84), also indicating a significant RFS benefit with allogeneic SCT (P < .01). We also assessed for systematic differences in effect estimates between the 15 trials that reported on RFS as well asoverall survival end points (group 1) vs the 3 trials that reported only on RFS end points (group 2). The combined HR for the 15 trials in group 1 was 0.80 (95% CI, 0.74-0.87) and for the 3 trials in group 2 was 0.79 (95% CI, 0.61-1.02). The near-identical summary effect estimate (HR ≈ 0.80) of group 1 and 2 indicates a lack of systematic difference in survival outcomes between the groups. A test of interaction between the 2 groups was not significant, as anticipated. We subsequently evaluated RFS outcomes by cytogenetic risk category of good-, intermediate-, and poor-risk AML. Sixteen trials reported RFS outcomes by cytogenetic risk. Good-risk AML had a combined HR of 1.06 (95% CI, 0.80-1.42) across 10 trials, indicating a lack of RFS benefit (P = .68). Intermediate-risk AML had a combined HR of 0.76 (95% CI, 0.68-0.85) across 14 trials, indicating significant RFS benefit with allogeneic SCT in CR1 (P < .01). Poor-risk AML had a combined HR of 0.69 (95% CI, 0.57-0.84) across 14 trials, indicating significant RFS benefit (P < .01). Tests of interaction between the 3 cytogenetic risk groups were statistically significant (P < .05), notably between good-risk vs poor-risk (P = .02) and vs intermediate-risk AML (P = .03) but not between poor- vs intermediate-risk AML (P = .40).

Two studies reported adjusted HRs for RFS end points. Use of adjusted HRs did not change our findings regarding RFS benefit with allogeneic SCT (Figure 2). Similarly, use of random-effects meta-analysis did not alter any conclusions regarding RFS benefit. In the 18 trials that reported end points of overall RFS across all cytogenetic risk groups, the combined HR for random-effects RFS benefit with allogeneic SCT was 0.80 (95% CI, 0.74-0.86). This overall estimate indicates statistically significant reduction in hazard of death or AML relapse with allogeneic SCT in CR1, across all cytogenetic risk groups (P < .01). Including 6 additional trials that restricted allogeneic SCT options for intermediate-risk patients with AML, poor-risk patients with AML, or both, resulted in a combined random-effects HR of 0.78 (95% CI, 0.71-0.85), also indicating a significant RFS benefit with allogeneic SCT (P < .01). We also evaluated RFS outcomes by cytogenetic risk category of good-, intermediate-, and poor-risk AML. Good-risk AML had a combined random-effects HR of 1.06 (95% CI, 0.80-1.42) across 10 trials, indicating a lack of RFS benefit (P = .68). Intermediate-risk AML had a combined random-effects HR of 0.76 (95% CI, 0.64-0.92) across 14 trials, indicating significant RFS benefit with allogeneic SCT in CR1 (P < .01). Poor-risk AML had a combined random-effects HR of 0.67 (95% CI, 0.52-0.85) across 14 trials, indicating significant RFS benefit with allogeneic SCT in CR1 (P < .01).

Fifteen trials reported overall survival end points across all cytogenetic risk groups and were included in the primary analysis. The summary hazard estimate for overall survival benefit varied between studies, ranging from 0.81 (donor group better) to 1.91 (no-donor group better). Interstudy heterogeneity was nonsignificant (P = .27), with I² = 18.5%. In a fixed-effects forest plot, the combined HR for overall survival benefit with allogeneic SCT was 0.90 (95% CI, 0.82-0.97) (Figure 3). The combined estimate indicates statistically significant reduction in hazard of death with allogeneic SCT across all cytogenetic-risk groups for AML in CR1 (P < .01).

We further evaluated the studies with sensitivity and subgroup analyses (Figure 3). In a sensitivity analysis, we repeated the initial analysis with 6 additional trials that provided overall survival allogeneic SCT outcomes restricted to intermediate-risk patients with AML, poor-risk patients with AML, or both. The combined HR was 0.87 (95% CI, 0.80-0.94), also indicating a significant overall survival benefit with allogeneic SCT (P < .01). We also evaluated overall survival outcomes by cytogenetic risk category of good-, intermediate-, and poor-risk AML. Sixteen trials reported overall survival outcomes stratified by cytogenetic risk. Good-risk AML had a combined HR of 1.07 (95% CI, 0.83-1.38) across 10 trials, indicating a lack of significant overall survival benefit (P = .59). Intermediate-risk AML had a combined HR of 0.83 (95% CI, 0.74-0.93) across 14 trials, indicating significant overall survival benefit with allogeneic SCT (P < .01). Poor-risk AML had a combined HR of 0.73 (95% CI, 0.59-0.90) across 14 trials, indicating significant overall survival benefit with allogeneic SCT (P < .01). Tests of interaction between the 3 cytogenetic risk groups were borderline statistically significant (P = .07), primarily between good- vs poor-risk (P = .02) and likely vs intermediate-risk AML (P = .07) but not between poor- vs intermediate-risk AML (P = .30).

Three studies reported adjusted HRs for overall survival end points. Use of adjusted HRs did not change our findings regarding allogeneic SCT overall survival benefit (Figure 3). Similarly, use of random-effects meta-analysis did not alter conclusions regarding overall survival benefit. In the 15 trials that reported overall survival end points across all cytogenetic risk groups, the combined random-effects overall survival benefit with allogeneic SCT was 0.90 (95% CI, 0.82-1.00). This indicates statistically significant reduction in hazard of death with allogeneic SCT for AML in CR1 across all cytogenetic risk groups (P = .04). Including 6 additional trials that restricted allogeneic SCT options for intermediate-risk patients with AML, poor-risk patients with AML, or both, the combined random-effects HR was 0.87 (95% CI, 0.78-0.98), also indicating significant allogeneic SCT overall survival benefit (P = .02). We also evaluated overall survival outcomes by cytogenetic risk category of good-, intermediate-, and poor-risk AML. Good-risk AML had a combined random-effects HR of 1.06 (95% CI, 0.64-1.76) across 10 trials, indicating a lack of overall survival benefit (P = .81). Intermediate-risk AML had a combined random-effects HR of 0.84 (95% CI, 0.71-0.99) across 14 trials, indicating significant overall survival benefit with allogeneic SCT for AML in CR1 (P = .03). Poor-risk AML had a combined random-effects HR of 0.60 (95% CI, 0.40-0.90) across 14 trials, also indicating significant overall survival benefit with allogeneic SCT for AML in CR1 (P = .01).

Despite multiple prospective studies over the past 2 decades, the role of allogeneic SCT for adult patients with AML in CR1 remains ill-defined. A meta-analysis by Yanada et al⁵ of 5 prospective trials indicated an overall survival benefit with allogeneic SCT (P = .04), and meta-regression suggested that the overall survival benefit may be restricted to poor-risk AML (P = .12). In addition to the limited number of trials and the use of indirect evidence (meta-regression) to indicate possible cytogenetic subgroup benefit, double counting of allogeneic SCT data from individual studies that reported allogeneic SCT vs autologous SCT and vs consolidation chemotherapy outcomes separately remains an unaddressed source of bias. An ITT donor vs no-donor analysis offers a better means to address such concerns. Cornelissen et al²² combined donor vs no-donor data from 4 cooperative groups (Bordeaux Grenoble Marseilles Toulouse [BGMT], Haemato-Oncology Co-operative Group [HOVON]/Swiss Group for Clinical Research [SAKK], MRC, and EORTC) in a meta-analysis to demonstrate a statistically significant survival benefit for allogeneic SCT; in cytogenetic subgroup analyses, an overall survival benefit was documented for intermediate-risk but not poor-risk AML. The limited number of trials assessed (eg, omission of the EORTC/GIMEMA-AML8A study) has likely precluded general acceptance of their study findings.

Thus, current recommendations from the National Comprehensive Cancer Network and from the American Society of Blood and Marrow Transplantation, both based on literature review and expert consensus, stratify treatment by cytogenetic risk and recommend allogeneic SCT for patients younger than 55 years with poor-risk AML in CR1; recommend against allogeneic SCT for patients with good-risk AML in CR1; and find insufficient evidence to recommend allogeneic SCT for patients with intermediate-risk AML in CR1.⁴ The direct evidence supporting these recommendations remains limited, as previously discussed. In part, this may be because all clinical trial data have not been systematically assessed. Quantitatively integrating data from all available trials will likely enhance understanding of the role of allogeneic SCT for AML in CR1. The robustness of any conclusions can be systematically assessed in secondary analyses.

To comprehensively assess the usefulness of up-front allogeneic SCT for AML in CR1, we therefore undertook a systematic literature review and meta-analysis of published data from clinical trials allocating allogeneic SCT vs nonallogeneic SCT options (autologous SCT, consolidation chemotherapy, or both) for such patients. We focused on an ITT analysis based on donor availability to capture information from all patients with AML evaluated for up-front allogeneic SCT with a donor search as part of a prospective trial. Prior meta-analyses have shown that survival after autologous SCT is equivalent to that with consolidation chemotherapy for patients with AML in CR1, supporting the decision to combine the nonallogeneic SCT treatment options in a single no-donor category.^6- 7,38

The systematic literature search identified 24 relevant trials comparing allogeneic SCT vs nonallogeneic SCT treatment for AML in CR1, none of which individually reported an allogeneic SCT overall survival benefit across all cytogenetic risk groups (Table 1), possibly owing to limited sample size (power calculations were not routinely described in the study reports). The trials are mature, enrolling patients between 1982 and 2006, and further long-term follow-up is unlikely to yield substantially different results. The trials varied with regard to patient eligibility, study trial design, cytogenetic risk classification, and specific interventions (Table 2). Importantly however, interstudy heterogeneity was not significant for overall survival or RFS end points, indicating that the impact of study differences was limited.

We considered the effect, if any, of differences in cytogenetic risk classification between studies. Such differences, if significant, may be anticipated to increase the between-studies heterogeneity for each cytogenetic risk group's end points, which was not observed. This is likely because the various cytogenetic risk classification schemes are fairly similar, though not identical.²² While we could not assess the impact of such differences directly, individual prospective studies that directly compared cytogenetic risk classifications (SWOG/ECOG, EORTC/GIMEMA, MRC) documented highly concordant effect estimates, independent of the classification schema used.^17- 18 It is therefore unlikely that variability between cytogenetic risk classifications significantly affected our analysis.

We also considered the role of treatment compliance. This likely disproportionately affects the allogeneic SCT (donor) group, because a significant fraction of patients with donors did not receive allogeneic SCT. Such crossover, analyzed on an ITT basis, is anticipated to reduce the observable survival benefit of allogeneic SCT. Typically, the studies reported an allogeneic SCT compliance rate of greater than 60%, which is considered reasonable for such prospective trials (one trial reported an allogeneic SCT compliance rate of 55%, and, in a sensitivity analysis, its removal did not affect the overall conclusions). In addition, the impact of salvage allogeneic SCT after AML relapse cannot be estimated but likely diminishes any observable overall survival benefit of up-front allogeneic SCT. Furthermore, the inclusion of older trials, some more than 2 decades old, likely also biases against allogeneic SCT, since advances in supportive care (eg, growth factors, improved anti-infective strategies, better prophylaxis/therapy of graft-vs-host disease) and transplantation methodology (eg, peripheral blood stem cells) are considered responsible for improvement in allogeneic SCT outcomes.

Our primary finding is that the totality of the prospective trial data indicates statistically significant RFS and overall survival benefit with allogeneic SCT for adult AML in CR1. This conclusion is supported by a variety of sensitivity and subgroup analyses, as reported above. Additionally, our analyses indicate that allogeneic SCT benefit likely varies by AML cytogenetic risk. We document significant RFS and overall survival benefit for allogeneic SCT in intermediate- and poor-risk AML, and a lack of significant RFS or overall survival benefit for good-risk AML. With regard to comparative absolute survival, we might anticipate 5-year overall survival rates in the control (nonallogeneic SCT) group of approximately 45% and 20% for intermediate- and poor-risk AML, respectively. Applying the corresponding estimated HRs of 0.83 (95% CI, 0.74-0.93) and 0.73 (95% CI, 0.59-0.90) for patients assigned to undergo allogeneic SCT for AML in CR1 increases the projected overall survival rates to 52% (95% CI, 48%-55%) and 31% (95% CI, 23%-39%) for intermediate- and poor-risk AML, respectively.

There are limitations to our analysis. We are aware of relevant studies that have not yet been reported (eg, MRC AML 12/15, GOELAM2).^39- 41 As a meta-analysis of the published literature, we extracted summary statistics (HRs) from individual studies to determine combined estimates. Dependence on published articles limits the level of detail that can be captured regarding subgroups that may have greater or lesser benefit from allogeneic SCT. We could not assess outcomes for clinically relevant subgroups other than cytogenetic risk. For instance, patient age is a likely relevant factor, and some, though not all, studies have indicated improved allogeneic SCT outcomes in younger adults.^18,21- 22 The median patient age in most trials in this report is in the 30s, and while the age eligibility was up to 60 years in individual studies, it remains unclear if older eligible patients obtained an equivalent benefit.

With regard treatment toxicity, we have summarized available treatment-related mortality data for individual studies (Table 3). However, the variable and limited data reported precluded a more formal analysis and highlight the need for more systematic reporting of this important end point. We also note that while patients in this analysis predominantly had de novo AML, eligibility criteria in some studies permitted enrollment of patients with prior myelodysplastic syndrome or therapy-related AML. Lastly, the impact of comorbid conditions could not be assessed, because trial eligibility criteria disbarred entry for such patients. Nonetheless, for treatment outside of the research setting, it has significant impact on allogeneic SCT outcomes in AML.^42- 43

A meta-analysis of individual patient data from the relevant clinical trials is a way to obtain more complete estimates of RFS and overall survival benefit with allogeneic SCT as well as to assess the impact of additional factors such as patient age. A broad overview of transplantation for AML in CR1 using individual patient data is currently being conducted by the Acute Leukemia Stem Cell Transplant Trialists' Collaborative Group. Nonetheless, to our knowledge our quantitative analysis of data from 24 trials comprising 6007 prospectively assigned patients provides the most complete estimate of allogeneic SCT benefit available.

Cytogenetic and molecular risk profiling in AML is an evolving field and can further stratify outcomes within a known cytogenetic risk group. For instance, Schlenk et al²⁴ from the German Austrian AML Study Group reported that for patients with cytogenetically normal AML (who would be classified as intermediate-risk), allogeneic SCT was beneficial for those with FLT3 internal tandem duplication (FLT3-ITD) or, in the absence of FLT3-ITD, for those without mutations in NPM1 and CEBPA, whereas for the subgroup with mutations in NPM1 and without FLT3-ITD there was no apparent benefit to having a matched sibling. However, such novel genetic lesions, as well as whole-genome analyses, RNA and microRNA profiles that have the potential to further refine AML risk are not in routine clinical use.^44- 45

While enrollment in therapeutic trials is to be encouraged, our findings provide evidence to guide clinical decision making and future trial design. We find evidence to support AML treatment based on cytogenetic risk. We conclude that allogeneic SCT does not provide significant benefit for good-risk AML in CR1 and that allogeneic SCT offers significant RFS and overall survival benefits for intermediate- and poor-risk AML in CR1. However, within these general guidelines, there remains a need to further individualize the allogeneic SCT decision, based on factors like patient age, comorbidity, and the presence of additional molecular lesions.