Therapies for acute myeloid leukemia in patients ineligible for standard induction chemotherapy: a systematic review

Acute myeloid leukemia (AML) represents a group of clonal hematopoietic stem cell disorders in which both failure to differentiate and over-proliferation in the stem cell compartment result in accumulation of non-functional cells, termed myeloblasts [1]. It is an aggressive hematopoietic malignancy and, if left untreated, patients die of infection or bleeding, usually within weeks. However, some older adults may have a slower progressive clinical course [2]. In adults, AML is the most common type of acute leukemia and has the shortest 5-year overall survival (OS) rate at 24%, with the mortality risk increasing with age [3].

In adults and children, AML is the second most common type of leukemia, making up 31% of total adult leukemia cases, and primarily affects the elderly (>65 years of age), with the average age at diagnosis of 68 years [1,4]. In 2022, it was estimated that AML will be diagnosed in 20,050 people of all ages in the USA [4]. As of 2017, the incidence of AML in Europe (France, Germany, Italy, Spain and the UK) was 3.5 per 100,000 individuals, with the reported prevalence in these five European countries ranging from approximately 2300 to 5000 cases [5].

AML treatment depends on disease subtype, patient age and overall patient health. Subject to patient fitness and eligibility, first-line treatment of AML in newly diagnosed patients involves induction and consolidation chemotherapy [6,7]. However, fewer than 50% of patients are candidates for first-line induction chemotherapy (IC) [8,9] due to older age, comorbidities, or Eastern Cooperative Oncology Group (ECOG) performance status [6,7], and only 10–20% of patients older than 80 years receive first-line IC [10].

Various treatment-decision algorithms have been developed to assess fitness for standard IC in newly diagnosed patients with AML [11–14]. Due to the complexity of this disease and poor outcomes in older patients, using one or more of these algorithms to identify patients who are not fit for standard IC can help facilitate discussions between clinicians and patients, thereby optimising personalised medicine.

For newly diagnosed patients with AML ineligible for first-line IC, the hypomethylating agents (HMAs) azacitidine and decitabine are the mainstay of therapy [6,7]. However, new data have emerged regarding the important role that cytogenetics and molecular abnormalities play in the risk stratification and management of patients with AML [15,16], and novel therapies targeting genetic mutations have resulted in improved survival rates in this difficult-to-treat population. Several mutations have previously been shown to affect overall survival (OS) in AML, including those in the NPM1 gene, CCAAT/CEBPA gene, FMS-related tyrosine kinase 3 gene (FLT3) [17], and, more recently, the IDH gene [6,18]. Mutations in IDH (IDH1 and IDH2) are among the most common genetic mutations in AML, detected in approximately 20% of patients [19]. IDH1 mutations in AML are present in 6–9% of patients, with a higher frequency of 8–18% among patients with normal karyotype AML (NK-AML) [18,20–23]. Multiple studies have examined the impact of IDH1/IDH2 mutations on prognosis, but the current data do not yet warrant their designation to one of the European LeukemiaNet (ELN) prognosis groups [15]. The IDH mutation does, however, serve as a potential target for new AML drug development, and the emerging use of IDH inhibitors may impact the prognostic outcome of patients with IDH1/IDH2-mutated (mIDH1/mIDH2) AML [15]. Screening for genetic mutations has become a standard part of the initial clinical workup in AML [6,7,15].

Treatment of AML is heterogeneous worldwide due to variable drug approval status, reimbursement practices, treatment guidelines, and definitions of ineligibility for IC. In the USA, the National Comprehensive Cancer Network (NCCN) guidelines recommend adding venetoclax (B-cell lymphoma [BCL]-2 inhibitor) to HMAs as the preferred therapy for patients who are ≥60 years old and ineligible for IC [6]. For patients with mIDH1 AML who are ≥60 years old and who are not candidates for IC or decline IC, the preferred regimens include HMA plus venetoclax, ivosidenib plus azacitidine, and ivosidenib monotherapy [6]. For patients with mIDH2 AML, NCCN guidelines recommend HMA plus venetoclax or enasidenib monotherapy, and for those with the FLT3 mutation, they recommend HMA plus venetoclax [6].

In Europe, the preferred regimens for this population according to the 2022 ELN AML recommendations include HMA plus venetoclax, low-dose cytarabine (LDAC) plus venetoclax, ivosidenib plus azacitidine for patients with mIDH1, and best supportive care (BSC) [15]. In Europe, ivosidenib received an orphan drug designation by the EMA for the treatment of AML in 2016 [24], and a marketing authorisation application was submitted to the EMA in March 2022 for this drug to be used in combination with azacitidine in patients with mIDH1 AML [15,25].

With the rapidly expanding landscape of therapies for AML, understanding the efficacy and safety outcomes of available therapies can assist in decision-making when managing this patient population. The objective of this systematic literature review (SLR) was to identify and qualitatively assess the clinical efficacy and safety evidence for current and emerging treatments for patients with newly diagnosed AML who are ineligible for first-line IC.

Materials & methods

Study design

This SLR was conducted in accordance with the recommendations of the Cochrane Handbook for Systematic Reviews and Interventions [26], the general principles of the Centre for Reviews and Dissemination (University of York) guidance [27] for undertaking reviews in healthcare, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [28], and the methods for systematic reviews as specified by the National Institute for Health and Care Excellence [29].

Literature search

A systematic literature search was conducted using patient/population, intervention, comparison and outcomes (PICOS) terms to identify English-language publications of clinical trials (any phase) that were published and indexed in the Embase (1974 to 28 October 2021), MEDLINE (1946 to 28 October 2021) and Cochrane Central Register of Controlled Trials (CENTRAL) (1991 to 28 October 2021) databases. The search targeted publications that evaluated therapeutic interventions in adults (≥18 years old) newly diagnosed with AML/secondary AML (sAML) and ineligible for standard IC. The search strategy combined free text words and indexing terms (e.g. medical subject headings [MeSH] terms for MEDLINE and EMTREE terms for Embase) relevant to AML and study designs with Boolean operators (e.g. ‘and’, ‘or’, ‘not’), using Embase-, MEDLINE-, and CENTRAL-specific search syntax. The complete literature search strategy and terms are described in the supplemental material (Supplementary Tables 1–5).

Interventions of interest included chemotherapy (decitabine, azacitidine, FLUGA [fludarabine, cytarabine and filgrastim], sapacitabine, ATRA [all-trans retinoic acid], LDAC, GRASPA [L-asparaginase encapsulated in red blood cells, eryaspase] and guadecitabine), immunotherapy (lenalidomide, durvalumab, talacotuzumab, lintuzumab and gemtuzumab ozogamicin), targeted therapy (venetoclax, enasidenib, gilteritinib, ivosidenib, volasertib, glasdegib, barasertib [AZD1152] and tipifarnib), and BSC.

Reference lists from recently published SLRs and meta-analyses in AML were also evaluated to ensure that all relevant evidence was captured. In addition, abstracts published between 1 January 2016 and 31 December 2021 from the following congresses were searched separately to identify relevant studies that are registered as completed but not yet published: European Society for Medical Oncology, American Society of Clinical Oncology, International Society for Pharmacoeconomics and Outcomes Research, European Haematology Association and American Society of Haematology.

Literature reviews & study selection

All abstracts identified in the literature searches were reviewed independently by two reviewers, and a sample of the abstracts was quality checked by a third independent reviewer. The reviewers then assessed the full texts of the included publications against the same eligibility criteria applied during the abstract review.

Trials were excluded if the study participants included patients who were eligible for first-line IC. Studies were excluded if they were performed in animals, patients <18 years of age, or patients with acute promyelocytic leukemia, or if they described interventions other than pharmacological interventions or BSC. Additionally, studies were excluded if they did not provide data on specific outcomes of interest (e.g. OS, event-free survival [EFS], relapse-free survival [RFS], disease-free survival [DFS], complete response [CR], partial response [PR] and adverse events [AEs]) or were not clinical trials (e.g. case studies, editorials, pilot studies and pharmacokinetic studies).

Data extraction & interpretation

Outcomes of interest were extracted from full-text studies by using a standardised data extraction template in MS Excel^®. They were organised for HMAs, LDAC, BSC, and ‘other’ (if they could not be categorised in the former three groups) as the four comparative treatment arms compared with chemotherapy, immunotherapy and targeted therapy. Comparative statistical analyses were not performed. If the publication did not specify the data for OS, EFS and PFS rates, these data were extracted from the Kaplan–Meier curves in the publications using WebPlotDigitizer [30]. At the time of writing, the most recent publication was used to report data of studies for which updated results were published after the dates of the systematic literature search.

Results

Selection of publications

A total of 4503 references were identified from electronic databases (MEDLINE: 828; Embase: 2629; CENTRAL: 1046). Following the removal of duplicate records, screening against the eligibility criteria, abstract title screening, and full-text review, 184 publications were included. Manual screening of relevant congresses and clinical trial registries as well as reference checks of other published reviews resulted in the identification of 49 additional records that met the inclusion criteria. Of all included articles, 26 unique studies [31–56] reported in 60 publications [31–90] were prioritised for data extraction. The prioritisation was based on study design, prioritising randomised controlled trials (RCTs) and studies with a sample size of 20 participants or greater. Figure 1 depicts the PRISMA flow diagram of the study selection process.

Key characteristics of the 26 included RCTs [31–56] are described in Supplementary Table 6. Of the 26 RCTs, nine (35%) were phase II/IIb trials [32,36,37,39,40,46,47,52,56], 12 (46%) were phase III trials [33,38,41–45,49,51,53–55], four (15%) were phase II/III trials [31,34,35,50], and one (4%) did not report study phase [48].

The total studied population size included in this SLR was approximately 7000 patients with AML who were ineligible for standard IC. The baseline characteristics of these patients are presented in Supplementary Table 7. Most of the studies included patients ≥75 years of age. The median age was reported in 23 studies and ranged from 64 [48] to 80 years [47]. In all studies except one [52], more men were included than women. The proportion of men ranged from 45 [52] to 80% [47]. Five studies [37,41,43,58,91] reported the race/ethnicity of enrolled patients; the majority of patients were White (range: 16 [91] to 100% [37]), followed by Asian (range: 2 [43] to 26% [91]), Black (range: 0.44 [43] to 2.2% [58]) and other/not reported (range: 2 [43] to 62% [91]).

Survival outcomes

Of the 26 identified clinical trials, all [31–56] reported data on OS, 12 trials [38–42,46,49–51,53–55] reported data on EFS, and six trials [34,35,40–42,53] reported data on RFS (Supplementary Table 8). Although EFS is a frequently reported end point in AML trials, not all trials used a uniform definition of EFS. Most commonly, EFS was defined as time from randomisation to disease progression, treatment failure, relapse, or death [38,39,46,49,50,53,55]. The definition of treatment failure varied among studies, with the VIALE-A trial defining it as failure to achieve CR or <5% bone marrow blasts after at least six cycles of treatment [38], and the AGILE trial defining it as lack of CR by week 24 [49]. The BI 1230.4 study only included disease progression or death in the EFS definition [40], the ASTRAL-1 trial reported EFS results as part of a post hoc analysis [68], and two trials (POLO-AML-2 and NCT02752035) did not define EFS in the publications [41,54].

Figure 2 illustrates the OS results for the nine trials [33,37,38,40,46,53,78,82,91] that demonstrated statistically significant differences in the OS outcome between study arms, and Figure 3 illustrates EFS results for five [38,40,53,82,91] of these nine trials that also reported EFS results.

Hypomethylating agents

In ten of the 26 studies, HMAs (azacitidine or decitabine) were compared with other interventions, including other chemotherapies (FLUGA [53], sapacitabine [44], valproate plus ATRA [46]), immunotherapies (lenalidomide [47], durvalumab [56], talacotuzumab [50]), and targeted therapies (venetoclax [38], enasidenib [39], gilteritinib [54], ivosidenib [49]).

Of the three studies [44,46,53] that compared HMAs with other chemotherapies, two [46,53] showed that treatment with HMAs significantly improved survival; one trial showed that combining a HMA with sapacitabine did not result in a significant improvement in survival [44]. In the PETHEMA-FLUGAZA trial [53], azacitidine treatment resulted in a median OS (mOS) of 9.8 months compared with 4.1 months with FLUGA (p = 0.005). Median EFS (mEFS) was also significantly longer with azacitidine compared with FLUGA (4.9 months vs 3.0 months; p = 0.001), but the increase in median RFS (mRFS) duration with azacitidine compared with FLUGA did not reach statistical significance (12.1 months vs 7.9 months; p = 0.43). In a subgroup analysis of patients with genomic variations, the mOS did not differ between the treatment groups among patients with mIDH1 (Supplementary Table 8) and, in the IDH2 subpopulation, the OS benefit with azacitidine was similar to the overall population (Supplementary Table 8). Although the number of observations in this subanalysis was limited (n = 34 patients with mIDH1 AML; n = 36 patients with mIDH2 AML), the trial reported a numerically lower mOS in the azacytidine-treated group with mIDH1 AML (7.0 months; 95% CI: 0–20.4) compared with mIDH2 AML (12.0 months; 95% CI: 2.3–21.7) [61].

The DECIDER trial [46] reported a mOS with decitabine of 4.8 months, compared with 6.1 months with decitabine plus valproate (hazard ratio [HR] = 0.85, 95% CI: 0.57–1.28), 8.4 months with decitabine plus ATRA (HR = 0.58, 95% CI: 0.37–0.91), and 7.7 months with decitabine plus ATRA plus valproate (HR = 0.62; 95% CI: 0.4–0.95). An exploratory analysis [70] showed that adding ATRA to decitabine monotherapy in a subgroup of patients with 20–30% bone marrow blasts resulted in a significant OS benefit (ATRA: 12.5 months vs no ATRA: 7.6 months; HR = 0.47, 95% CI: 0.24–0.94; p = 0.032).

In the SEAMLESS trial [44], no improvement in mOS was seen when comparing the combination of decitabine and sapacitabine with decitabine monotherapy (Supplementary Table 8).

Three studies [47,50,56] that evaluated outcomes when immunotherapies (lenalidomide, durvalumab and talacotuzumab) were added to HMAs did not show survival benefits compared with HMA monotherapy (Supplementary Table 8).

Of the ten studies with HMAs as the comparator therapy, four [38,39,54,91] evaluated the addition of targeted therapy to an HMA. In the VIALE-A trial, the combination of venetoclax plus azacitidine significantly increased the mOS by approximately 5 months compared with azacitidine plus placebo (14.7 months vs 9.6 months; HR = 0.66; 95% CI: 0.52–0.85; p < 0.001); similarly, the mEFS increased by almost 3 months (9.8 months vs 7.0 months; HR = 0.63; 95% CI: 0.50–0.80; p < 0.001). Compared with azacitidine plus placebo, OS and EFS rates improved consistently over 12, 18 and 24 months in patients who received venetoclax plus azacitidine (Supplementary Table 8). In a subgroup of 89 patients with IDH1 or IDH2 mutations at baseline, OS rates at 12 months were also significantly improved in the venetoclax plus azacitidine group compared with the azacitidine plus placebo group (66.8% vs 35.7%; HR = 0.35; 95% CI: 0.20–0.60; p < 0.001). In a subgroup of 34 patients with mIDH1 AML, the hazard of death was lower in the venetoclax plus azacitidine group compared with the azacitidine plus placebo group (HR = 0.28; 95% CI: 0.12–0.65).

The AGILE trial [49] investigated the combination of ivosidenib plus azacitidine in previously untreated patients with mIDH1 AML ineligible for induction chemotherapy. A 16-month improvement in mOS was seen with ivosidenib plus azacitidine compared with azacitidine plus placebo (24.0 months vs 7.9 months; HR = 0.44; 95% CI: 0.27–0.73; p = 0.001); a 19-month improvement was observed in mEFS (22.9 months vs 4.1 months; HR = 0.39; 95% CI: 0.24–0.64; p < 0.001). Similarly, OS and EFS rates continually improved over 24 months in patients who received ivosidenib plus azacitidine compared with those who received azacitidine plus placebo (Supplementary Table 8).

In contrast to the VIALE-A and AGILE trials, mOS and EFS durations were not significantly different on adding enasidenib to azacitidine in a population with mIDH2 AML [39] or adding gilteritinib to azacitidine in patients with FLT3-mutated (mFLT3) AML compared with azacitidine alone (Supplementary Table 8).

Low-dose cytarabine

In ten of the 26 studies, LDAC monotherapy (or in combination with placebo) was compared with LDAC in combination with chemotherapy [34,35], immunotherapy [36,52], targeted therapy [31,37,40,41,55] or ‘other’ therapy (GRASPA) [32].

No significant differences were seen in two-year OS or RFS when LDAC monotherapy was compared with chemotherapy consisting of clofarabine [35] or sapacitabine [34] (Supplementary Table 8).

The five studies that evaluated adding a targeted therapy to LDAC reported significant improvements in survival compared with LDAC monotherapy [31,37,40,41,55] (Supplementary Table 8). Two trials [40,41] investigated the combination of volasertib with LDAC in comparison with LDAC monotherapy (with or without placebo) but reported dissimilar results. In the BI 1230.4 trial [40], the mOS and EFS durations were significantly longer with volasertib plus LDAC than with LDAC monotherapy (OS: 8.0 months vs 5.2 months; HR = 0.63; 95% CI: 0.4–1.0; p = 0.047; EFS: 5.6 months vs 2.3 months; HR = 0.57; 95% CI: 0.35–0.92; p = 0.021). In contrast, the POLO-AML-2 trial [41] showed no significant difference in mOS or EFS between volasertib plus LDAC and LDAC plus placebo (OS: 5.6 months vs 6.5 months; HR = 0.97; 95% CI: 0.8–1.2; p = 0.76; EFS: 3.3 months vs 2.8 months; HR = 0.96; 95% CI: 0.8–1.2; p = 0.67). In both trials, the RFS duration with volasertib plus LDAC was longer than with LDAC alone, but p-values were not reported for this outcome [40,41].

In the VIALE-C trial [55,82], the final analysis [82] showed that venetoclax plus LDAC resulted in significantly longer mOS and mEFS durations compared with LDAC plus placebo (OS: 8.4 months vs 4.1 months; HR = 0.7; 95% CI: 0.50–0.98; p = 0.04; EFS: 4.9 months vs 2.1 months; p = 0.002). A subgroup analysis [55] of 33 patients with an mIDH1/2 demonstrated a numerical improvement in mOS with venetoclax plus LDAC compared with LDAC plus placebo (10.8 vs 9.0), but the sample size was small, and the differences were not statistically significant.

In the BRIGHT AML 1003 trial [37], glasdegib plus LDAC resulted in significantly improved mOS versus LDAC alone in the primary analysis (8.3 months vs 4.3 months; HR = 0.46; 95% CI: 0.35–0.62; p = 0.0002); these benefits were maintained in the final post hoc analysis [67]. In a small subgroup of patients with mIDH1de novo AML (n = 8), the mOS duration with glasdegib plus LDAC (n = 6) was 7.8 months (95% CI: 3.3–27.1 months) compared with 1.6 months (95% CI: 1.3–1.9 months) with LDAC alone (n = 2). In two patients with mIDH1 sAML, the mOS duration with glasdegib plus LDAC was 13.8 months (95% CI: 3.1–24.4 months); there were no patients with mIDH1 sAML assigned to the LDAC alone group. In eight patients with mIDH2de novo AML, the mOS duration was 5.0 months (95% CI: 1.1–23.1 months) with glasdegib plus LDAC; there were no patients with mIDH2de novo AML assigned to the LDAC alone group. In three patients with mIDH2 sAML, the mOS duration was 4.6 months (95% CI: 2.2–7.1 months) compared with 5.8 months (95% CI: 4.5–6.5 months) with LDAC alone (n = 4).

The SPARK-AML1 study demonstrated that the addition of targeted therapy – barasertib – to LDAC numerically improved the mOS compared with LDAC monotherapy, but the sample size was small (n = 77) and statistical significance was not reached (8.2 months vs 4.5 months; HR = 0.88; 95% CI: 0.49–1.58; p = 0.663).

Combining LDAC with immunotherapy did not improve survival outcomes in the trials evaluating lenalidomide [36] or lintuzumab [52] (Supplementary Table 8).

LDAC was also evaluated with GRASPA (erythrocyte encapsulated l-asparaginase), for the treatment of AML in the ENFORCE 1 trial [32]; the OS duration was statistically similar compared with LDAC monotherapy (Supplementary Table 8).

Best supportive care

Limited data were available comparing BSC with other treatment strategies. In the two studies [33,43] where BSC was a comparator group, BSC included administration of blood product transfusions, antimicrobials and other therapies for the symptomatic treatment of AML and its complications, according to institutional policies; hydroxyurea was permitted in both studies. In the AML-19 trial [33], a longer mOS duration was seen in the gemtuzumab ozogamicin group compared with the BSC group (4.9 months vs 3.6 months; HR = 0.69; 95% CI: 0.53–0.90; p = 0.005). In the FIGHT-AML-301 trial [43], targeted therapy – tipifarnib – resulted in similar mOS duration as in the BSC group (Supplementary Table 8).

Other interventions

During data extraction, the interventions in four trials [42,45,48,51] could not be categorised in one of the above groups and were, therefore, categorised as ‘other’. All four studies compared HMAs, LDAC, or BSC in one of the treatment arms. Three [42,45,51] compared HMA monotherapy with treatment choice (TC) and one trial [48] compared LDAC with BSC.

In the DACO-016 trial, TC included BSC or LDAC, and although statistical differences in OS duration were not detected in the intention-to-treat (ITT) analysis [45], the increase in OS duration with decitabine therapy compared with TC therapy reached statistical significance in the ad hoc mature analysis (7.7 months vs 5.0 months; HR = 0.82; 95% CI: 0.68–0.99; p = 0.03) [78]. OS rates were also significantly higher in the decitabine arm compared with the TC arm at 18 months (p = 0.027) and 24 months (p = 0.019) in the post hoc analysis [77].

In the AZA-AML-001 trial [42], 488 patients with AML and >30% BM blasts were randomised to the azacitidine arm or the ‘combined conventional care regimen (CCR)’, which included BSC, LDAC, or standard IC. Numerically, but not statistically significantly, higher mOS and mEFS durations were seen in the azacitidine arm compared with the combined CCR arm (Supplementary Table 8). However, the mRFS duration was slightly shorter with azacitidine (9.3 months vs 10.5 months; p = 0.6). Subgroup analyses for each of the treatments in the CCR group showed that azacitidine monotherapy resulted in significantly longer mOS compared with BSC, but not compared with LDAC or IC (Supplementary Table 8) [42]. Subgroup analysis [64] of patients with mIDH2 AML showed slight improvements in OS rates at 18 and 24 months in the azacitidine arm; however, improvements in OS were not seen in patients with mIDH1 AML, with numerically lower OS rates at 12 months with azacitidine in the mIDH1 AML group (49.2%) compared with the mIDH2 AML group (60.5%). Sample sizes were very small in these subgroups with mIDH AML that received azacitidine (n = 6 patients with mIDH1 AML; n = 21 with mIDH2 AML) (Supplementary Table 8).

In the ASTRAL-1 trial [51], mOS duration and EFS rates were similar in the guadecitabine and TC (azacitidine, decitabine, or LDAC) arms [51,68], and mOS duration was similar in a subgroup analysis [85] of the different treatments within the TC arm (Supplementary Table 8).

A small study compared LDAC with BSC in 60 patients with AML and showed that the mOS duration was 8.4 months in the BSC group but was not reached in the LDAC group (Supplementary Table 8).

Transfusion independence outcomes

Of the 26 identified clinical trials, seven [37–39,42,45,49,55] reported transfusion independence (TI) rates after intervention, either in the main publication [38,39,42,55] or in an associated paper [65,79,81,84,91]. These seven trials used variable definitions for TI. All seven studies defined TI as no transfusions for 56 consecutive days, but two (VIALE-A [38] and BRIGHT AML 1003 [79]) evaluated TI in all randomised patients, two trials (AG221-AML-005 [39], AZA-AML-001 [42,91]) reported TI only among patients who were transfusion dependent at baseline, and three trials (AGILE [91], VIALE-C [55,82,84], and DACO-016 [65]) reported TI for all randomised patients as well as patients who were transfusion dependent at baseline. Of these seven trials, five [38,65,79,81,91] reported statistically significant differences in OS between study arms, and the TI results for these five trials are illustrated in Figure 4.

In three of the seven trials that evaluated TI, HMA (azacitidine) was compared with targeted therapy [38,39,91]. The VIALE-A study [38], which included all patients in the TI definition, reported similar baseline transfusion-dependence rates in the venetoclax plus azacitidine (red blood cells [RBC]: 50.0%; platelets [PLT]: 24.0%) and azacitidine plus placebo (RBC: 52.0%; PLT: 22.0%) groups. After treatment, RBC and PLT TI rates in the venetoclax plus azacitidine groups were significantly higher than the rates in the azacitidine plus placebo group RBC: 59.8% (95% CI: 53.9–65.5) versus 35.2% (95% CI: 27.4–43.5); p < 0.001; PLT: 68.5% (95% CI: 62.8–73.9) versus 49.7% (95% CI: 41.3–58.1; p < 0.001; Figure 4).

The AGILE trial [91] reported similar baseline RBC and/or PLT transfusion-dependence rates of 54% in the ivosidenib plus azacitidine and placebo plus azacitidine groups in patients with mIDH1 AML. Among this group of transfusion-dependent patients, ivosidenib plus azacitidine resulted in a significantly higher rate of RBC and/or PLT TI compared with placebo plus azacitidine (46.2% vs 17.5%; p = 0.006). An evaluation of all patients (regardless of baseline transfusion status) showed a numerically greater proportion of patients in the ivosidenib plus azacitidine group experiencing RBC and/or PLT TI compared with the placebo plus azacitidine group (62.5% vs 51.4%; p = 0.21) (Figure 4).

The AG221-AML-005 trial [39] showed mixed results in patients with mIDH2 AML. A post hoc analysis showed that, among those who were transfusion dependent at baseline, a higher proportion of patients experienced RBC TI in the enasidenib plus azacitidine group than in the azacitidine monotherapy group (59% vs 41%; p-value not reported). In contrast, the proportion of patients that converted to PLT TI was lower in the enasidenib plus azacitidine group than the azacitidine monotherapy group (47% vs 63%; p-value not reported).

Two of the seven studies evaluating TI compared LDAC to targeted therapy [55,79,82,84]. Both studies demonstrated higher TI rates in patients treated with LDAC in combination with targeted therapy compared with LDAC with or without placebo. In the VIALE-C study, the primary analysis showed that the RBC and PLT TI rates were significantly higher in patients treated with venetoclax plus LDAC compared with those treated with LDAC plus placebo [RBC: 41% (95% CI: 32–49) versus 18% (95% CI: 10–29); p = 0.001; PLT: 48% (95% CI: 39–56) versus 32% (95% CI: 22–45); p = 0.04)] [55], with similar results seen in the 6-month follow-up analysis [82]. The 6-month analysis also reported TI rates in the subgroup of patients who were transfusion dependent at baseline, showing numerically higher RBC and PLT TI rates in the venetoclax plus LDAC group compared with the LDAC plus placebo group [RBC: 40.4% (95% CI: 30.9–50.5) versus 16.7% (95% CI: 7.9–29.3); PLT: 28.8% (95% CI: 17.1–43.1) versus 12.5% (95% CI: 2.7–32.4)]; statistical analysis was not performed for this subgroup (Figure 4). In a cohort of 27 Japanese patients in the VIALE-C trial, post-baseline RBC and PLT TI rates were similar in both arms, likely due to the small sample size [84].

In the post hoc analysis of the BRIGHT AML 1003 trial [79], higher rates of RBC and PLT TI were seen among patients treated with glasdegib plus LDAC compared with those treated with LDAC alone (RBC: 36.0% vs 8.0%; PLT: 47.0% vs 14.0%; p-values not reported) (Figure 4).

In the DACO-016 study, RBC and PLT TI rates were significantly higher in patients randomised to decitabine compared with those randomised to TC (RBC: 30.0% vs 19.0%; p = 0.0058; PLT: 48.0% vs 35.0%; p = 0.0028)] [65]. In the subgroup of patients with transfusion dependence at baseline, post-baseline RBC and PLT TI rates were also significantly higher in the decitabine group compared with the TC group (RBC: 26.0% vs 13.0%; p = 0.0026; PLT: 31.0% vs 13.0%; p = 0.0069; Figure 4).

In the AZA-AML-001 trial, a numerically higher proportion of patients in the azacitidine arm achieved RBC transfusion independence compared with CCR (38.5% vs 27.6%) as well as PLT TI (40.6% vs 29.3%) [42].

Safety

Of the 26 trials included in the SLR, six [34–36,48,54,56] did not report rates of all-cause or treatment-emergent/treatment-related AEs (TEAEs/TRAEs; Supplementary Table 9). Figure 5 illustrates the incidence of grade ≥3 key AEs (febrile neutropenia, pneumonia and QT interval prolongation) reported in the nine clinical trials [33,37,38,40,45,46,53,55,91] with statistically significant differences in mOS between study arms.

HMAs

Safety end points were reported in three trials [44,46,53] that compared an HMA with chemotherapy. In all three trials, the most common haematologic toxicities were neutropenia, febrile neutropenia and thrombocytopenia. Pneumonia was the most common non-haematologic toxicity reported in these studies (Supplementary Table 9).

Of the three studies [47,50,56] that compared HMA monotherapy with immunotherapies, two [47,50] reported no relevant difference in safety outcomes between the arms (Supplementary Table 9), and one study [56] did not report rates of AEs.

Of the four studies that compared HMA plus targeted therapy with HMA monotherapy (or with placebo), two trials reported higher rates of AEs in the targeted therapy arm compared with the HMA arm [38,39]. The other two trials reported similar safety outcomes in both arms [49,54]. In the VIALE-A trial [38], the most frequently reported grade ≥3 haematologic AEs in the azacitidine plus venetoclax and azacitidine plus placebo groups were thrombocytopenia, neutropenia, and febrile neutropenia, and the most common grade ≥3 non-haematologic AEs were hypokalaemia and pneumonia (Supplementary Table 9).

In the AG221-AML-005 trial [39], neutropenia, thrombocytopenia, and anaemia were the most common grade 3–4 TRAEs (Supplementary Table 9). There was no incidence of differentiation syndrome in the azacitidine monotherapy group, but 12 (18%) patients in the combination group experienced 14 such events. The median time to onset of differentiation syndrome was 28.5 days (inter-quartile range [IQR]: 17.0–34.5); all 14 events resolved within a median of 11.5 days (IQR: 7.0–19.0).

In the AGILE trial [91], the rates of AEs were similar between the ivosidenib plus azacitidine arm and the azacitidine plus placebo arm. Febrile neutropenia, neutropenia, anaemia, thrombocytopenia and pneumonia were the most common grade ≥3 TEAEs in both arms (Supplementary Table 9). The proportion of patients with infections was lower in the ivosidenib plus azacitidine group (any grade: 28%; grade ≥3: 21%) than the azacitidine plus placebo group (any grade: 49%; grade ≥3: 30%). The rates of febrile neutropenia were also lower in the ivosidenib plus azacitidine group (any grade: 28%; grade ≥3: 28%) than the azacitidine plus placebo group (any grade: 34%; grade ≥3: 34%). Consistent with these observations, an increase in the absolute neutrophil count (ANC) was noted from baseline over time in the ivosidenib plus azacitidine arm, particularly during the first treatment cycle. QT interval prolongation was reported more frequently with ivosidenib plus azacitidine (any grade: 20%; grade ≥3 10%) than with azacitidine plus placebo (any grade: 7%; grade ≥3: 3%). Differentiation syndrome was seen more frequently in the ivosidenib plus azacitidine group any grade: 14% (10 events); grade 3: 4% (3 events); grade 4: 0%) than the azacitidine plus placebo group [any grade: 8% (6 events); grade 3: 3% (2 events); grade 4: (1% (one event)], with a median onset of 19.5 days (range, 3.0–33.0).

In a study comparing gilteritinib plus azacitidine with azacitidine monotherapy in patients with mFLT3 AML [54], rates of AEs were similar between arms (Supplementary Table 9).

Low-dose cytarabine

Two trials compared LDAC monotherapy (or plus placebo) with LDAC plus immunotherapy [36,52]. The LI-1 trial found that the rate of AEs was higher with lenalidomide plus LDAC than with LDAC monotherapy, including thrombotic events [36] (Supplementary Table 9). In contrast, the trial comparing lintuzumab plus LDAC with LDAC plus placebo showed no difference in the rates of AEs, except for a higher frequency of infusion-related reactions with lintuzumab [52] (Supplementary Table 9).

All five trials that compared LDAC monotherapy (or with placebo) with LDAC plus targeted therapy reported higher rates of AEs in the combination therapy arms [31,37,40,41,55] (Supplementary Table 9). In the VIALE-C trial [55], the rates of AEs were higher in the venetoclax plus LDAC group compared with the LDAC plus placebo group. In the primary [55] and final [82] analyses, the most common haematologic grade ≥3 TEAEs in the venetoclax plus LDAC and LDAC plus placebo were febrile neutropenia, neutropenia, thrombocytopenia and anaemia (Supplementary Table 9).

In the BRIGHT AML 1003 trial [37] the rates of AEs were higher in the glasdegib plus LDAC arm than the LDAC monotherapy arm, with anaemia, thrombocytopenia and febrile neutropenia as the most common haematologic grade ≥3 TEAEs. Grade 3 QT_c prolongation (QTc ≥500 ms) was seen in 5/83 (6.0%%) patients treated with glasdegib plus LDAC and in 2/17 (11.8%) patients treated with LDAC monotherapy.

In the SPARK-AML1 trial [31], the incidence of grade ≥3 AEs was higher in the barasertib arm (83%) than the LDAC monotherapy arm (69%).

In the ENFORCE 1 trial, serious AEs were reported by 91.4% of patients receiving GRASPA plus LDAC and by 82.1% of patients receiving LDAC monotherapy [32] (Supplementary Table 9).

Best supportive care

In the two studies [33,43] where BSC was a comparator group, the AML-19 trial [33] reported no differences in the rate of AEs between the immunotherapy, gemtuzumab ozogamicin and BSC arms (Supplementary Table 9). In contrast, the FIGHT-AML-301 trial [43] showed higher rates of AEs with tipifarnib, particularly grade 4 neutropenia and thrombocytopenia as well as grade 3 or 4 diarrhoea, than with BSC (Supplementary Table 9).

Other interventions

Of the four trials [42,45,48,51] that compared other treatment strategies in patients with AML, three [42,45,51] reported safety outcomes. In the DACO-016 [45] and ASTRAL-1 [51] trials, the rate of AEs was higher in the decitabine (DACO-016) and guadecitabine (ASTRAL-1) arms compared with the TC arms (Supplementary Table 9). Conversely, comparable safety was seen in the azacitidine and CCR arms in the AZA-AML-001 trial (Supplementary Table 9) [42].

Risk of bias

The quality of the 26 included RCTs was assessed using the revised Cochrane Risk of Bias tool for RCTs (RoB 2.0) [92]. The risk of bias was assessed in five distinct domains, with each answer leading to judgements of ‘low risk of bias’, ‘some concerns’, or ‘high risk of bias’ (Figure 6). Concerning overall study risk of bias, 11 studies [33,38,41–43,45,46,49,50,52,55] showed a low risk, ten studies [34,35,37,39,40,44,47,48,51,53] showed some concerns, and five studies [31,32,36,54,56] showed a high risk of bias.

Discussion

Multiple treatment strategies have been investigated for patients with newly diagnosed AML who are ineligible for chemotherapy. Studies that evaluated the strategy of combining LDAC with novel therapies showed that adding immunotherapy to LDAC did not provide any added survival benefit [36,52]; adding targeted therapy, such as venetoclax or glasdegib, to LDAC improved survival rates and TI rates but resulted in higher rates of AEs [37,40,67,82].

Although not meeting the inclusion criteria for this SLR, it should be noted that data from a phase II single-arm trial showed that combining LDAC with cladribine (a purine nucleoside and inhibitor of S-adenosylhomocysteine hydrolase), alternating with decitabine, resulted in a mOS of 13.8 months and DFS of 10.8 months in patients with AML who were ≥70 years old and unfit for standard IC [93]. This regimen, which was generally well tolerated, may offer another treatment option for this population.

Trials that compared novel therapies with BSC reported improved survival with immuno-chemotherapy (gemtuzumab ozogamicin) [33], but not targeted therapy (tipifarnib) [43]. Evaluation of other treatment strategies showed that OS improved significantly with decitabine compared with TC (supportive care or LDAC) [78], but no significant improvement was observed with azacitidine compared with BSC [42].

HMAs are the mainstay of therapy, with some additional strategies showing survival benefits, such as adding FLUGA or ATRA chemotherapy to HMAs, but the toxicity rates in this difficult-to-treat patient population are unacceptably high [46,53].

Of the newer treatments, adding immunotherapies to HMA therapy did not add any survival benefit [47,56]. Interestingly, allogeneic haematopoietic cell transplantation (alloHCT) is the most commonly used and best-established form of immunotherapy in AML, resulting in the longest OS duration in patients who are candidates for the procedure [6,7,15]. Unlike immunotherapy drugs, which are used in the setting of a very dysfunctional immune system in the presence of AML cells, alloHCT results in instalment of healthy immune and T cells from a donor, leading to a graft-versus-leukemia (GvL) response [94–96]. To improve the efficacy of immunotherapy drugs for patients with AML who are not candidates for alloHCT, novel cell-surface targets will need to be identified and evaluated for efficacy and toxicity for use in AML.

Two new targeted therapies, venetoclax and ivosidenib, have shown the most promising and consistent beneficial survival results when added to an HMA [38,91]. In the VIALE-A study [38], venetoclax plus azacitidine was studied in 431 patients with AML ineligible for IC, with a subgroup of 34 patients with the IDH1 mutation. Although a reduction in the risk of death was shown with venetoclax and azacitidine compared with azacitidine plus placebo in the subgroup with mIDH1 AML, the results should be interpreted with caution due to the small sample size, lack of stratification for mutations at baseline, and lack of statistical testing for interaction between the mutations and survival. In a pooled analysis [97] of the VIALE-A trial and a single-arm phase Ib study that evaluated venetoclax with an HMA (azacitidine or decitabine), the mOS in the 44 patients with mIDH1 AML was 15.2 months (95% CI: 7.0–not estimable) in the combination therapy group compared with 2.2 months (95% CI: 1.1–5.6) in the HMA monotherapy group from the VIALE-A study (HR = 0.19; 95% CI: 0.08–0.44). Due to the drawbacks of pooling data, such as differences in study design, lack of a control group in the phase I trial and, therefore, a limited number of patients in the azacitidine monotherapy group, and the small sample sizes in the IDH1/2 molecular subgroups, the results of this analysis should be interpreted with caution.

The low incidence of IDH1 mutations in patients with AML makes it challenging to conduct phase 3 studies in this population. The AGILE trial [91] evaluated ivosidenib plus azacitidine in 146 patients with mIDH1 AML and is the only phase III study in the AML population that exclusively enrolled patients with this mutation. The results from this trial showed significant increases in the duration of mOS and mEFS as well as improvements in the rates of OS and EFS at 24 months with ivosidenib plus azacitidine compared with azacitidine plus placebo. Compared with the VIALE-A trial, where the HR for OS in all randomised patients was 0.66 (95% CI: 0.52–0.85; p < 0.001) [38], the AGILE trial demonstrated a HR of 0.44 (95% CI: 0.27–0.73; p = 0.001) in the mIDH1 population [91], representing an increase in mOS by approximately 5 months when adding venetoclax to azacitidine and approximately 16 months when adding ivosidenib to azacitidine. Similarly, the HR for EFS was 0.63 (95% CI: 0.50–0.80; p < 0.001) in the VIALE-A trial [38] and 0.33 (95% CI: 0.16–0.69; p = 0.002) in the AGILE trial [91], representing an increase in the mEFS duration by approximately 3 months with venetoclax plus azacitidine and by approximately 19 months with ivosidenib plus azacitidine. This significant survival benefit with ivosidenib plus azacitidine may be partially due to the synergy seen in a preclinical trial with the two drugs [98]. The simultaneous administration of ivosidenib and azacitidine leads to inhibition of leukemia stem cells through suppression of mitogen-activated protein kinase and RB/E2F signalling, which are needed for cell proliferation and survival, resulting in a depletion of leukemia cells in mice by 7960-fold with this combination compared with azacitidine alone [98].

Interestingly, the phase II AG221-AML-005 trial [39], conducted specifically in the mIDH2 AML population, had a comparable control group of azacitidine plus placebo as the phase III AGILE trial [91] performed in the mIDH1 AML population and the pooled analysis that included the VIALE-A and a phase 1 trial with venetoclax plus azacitidine [97]; however, the median survival durations in the control group were notably longer in the AG221-AML-005 trial (mOS: 22.0 months, 95% CI: 14.6–not reached; mEFS: 15.9, 95% CI: 13.0–not reached) compared with that in the AGILE trial (mOS: 7.9 months, 95% CI: 4.1–11.3; mEFS: 4.1, 95% CI: 2.7–6.8) and the subpopulation of patients with mIDH2 AML in the pooled analysis of the VIALE-A and phase I trial with venetoclax and azacitidine (mOS: 13.0 months, 95% CI: 3.8–15.8). It is not clear why such differences in survival were seen in the control groups of these three trials, but it may be due to the difference in the studied populations. In contrast to the AGILE trial, which enrolled patients with mIDH1 AML, the AG221-AML-005 trial enrolled patients with mIDH2 AML, and some data suggest that mIDH2 is an independent favourable prognostic factor in patients with AML [99,100]. The small sample size of 33 patients with mIDH2 AML in the pooled analysis of the VIALE-A and phase 1 trials with venetoclax and azacitidine may have contributed to the shorter mOS compared with that seen in the AG221-AML-005 trial.

One efficacy end point that has not been well characterised in AML studies is TI, despite evidence from other haematologic cancers (e.g. myelodysplastic syndromes) showing that transfusion dependence significantly increases the economic, clinical, and quality of life (QoL) burden in patients [101]. Study cross-comparisons and interpretation of these data are challenging due to the various assessment methods used across studies and some trials [42,79] not reporting baseline transfusion-dependence data.

The added risk of toxicities with combination therapy requires careful consideration, since adding an additional drug generally may result in increased toxicity, as was seen in clinical trials that evaluated adding certain targeted drugs to an HMA [38,39] and to LDAC [31,37,40,41,55], or adding immunotherapy to LDAC [36]. In the VIALE-A study, a higher rate of serious AEs was reported with the combination of venetoclax and azacitidine compared with azacitidine plus placebo, with febrile neutropenia and pneumonia being the most reported serious AEs [38]. Interestingly, the rates for these two AEs were lower in the ivosidenib plus azacitidine group than in the azacitidine plus placebo group in the AGILE trial, and the rate of infections was lower in the ivosidenib plus azacitidine group. These findings are likely due to the increase in the ANC and leucocytosis seen with ivosidenib. This increase in neutrophils has not been observed with the other targeted therapies studied in patients with AML but has been noted in some studies with ivosidenib [102,103]. A bone marrow analysis in a phase I trial of patients with mIDH1 AML showed that ivosidenib induced myeloid differentiation and trilineage haematopoietic recovery without an intervening period of bone marrow aplasia, indicating that functional neutrophils are involved in haematopoietic recovery, consistent with ivosidenib's mechanism of action of differentiation [102].

The incidence of differentiation syndrome reported with ivosidenib plus azacitidine in the AGILE study [91] was similar to that seen with enasidenib plus azacitidine in the AG221-AML-005 trial [39], with supportive care resulting in resolution of this adverse event. QTc interval prolongation has been noted with targeted therapies [104], and grade ≥3 QTc prolongation (QTc >500 ms) was seen in the AGILE trial [91] and the BRIGHT AML 1003 trial [37]. Torsades de pointes was not seen in these studies, but the risk of QTc prolongation should be considered when initiating targeted therapies, and electrolyte levels should be closely monitored throughout the duration of treatment.

As the landscape of AML therapies evolves, health-related quality of life (HRQoL) is a crucial consideration in the overall management of AML, since the goal of therapy is not just to prolong survival, but to optimise QoL and minimise the risk of cytopenia-related complications [15]. As HRQoL scores in adults with AML have been reported to be significantly lower than in the population norms and lower than those in many other cancers [105,106], this element becomes a great challenge and is of utmost importance in older patients with AML who are ineligible for standard IC [107].

Interpreting HRQoL outcomes is challenging due to the various instruments used to capture the physical, emotional, social and other domains. Additionally, the study population evaluable for HRQoL in AML studies is often smaller than the ITT population, diminishing over the study period due to the high mortality rate. Our SLR identified only four trials [38,55,63,91] that reported HRQoL data with targeted therapies. In the AG221-AML-005 trial, treatment with enasidenib plus azacitidine did not result in improvement of HRQoL outcomes [39]. In the VIALE-C study, better improvements from baseline in fatigue and global health status were seen in the venetoclax plus LDAC arm compared with the LDAC plus placebo arm, and there was a trend of longer time to deterioration in the venetoclax plus LDAC arm across each subscale [55]. In the VIALE-A study, there were no differences from baseline in the QoL measures between the venetoclax plus azacitidine group and the azacitidine plus placebo group [38]. In the AGILE trial, improvements from baseline across all European Organisation for Research and Treatment of Cancer QLQ-C30 subscales were noted in the ivosidenib plus azacitidine arm [91]. As targeted therapies lead to prolonged survival in patients with AML, it will be important to assess the long-term tolerance of these agents and to compare the HRQoL outcomes associated with these therapies.

The limitations associated with SLRs should be considered when interpreting these findings. Characteristic to all SLRs, some risks include missing relevant publications due to improper indexing within databases, restricting this SLR to English-language publications, or articles published after the searches were performed. Three major databases were searched for data extraction, along with relevant congress proceedings, to avoid missing relevant publications. It is challenging to make cross-study comparisons or generalised conclusions with SLRs due to different outcomes reported across studies. Variability in the baseline characteristics of enrolled patients can also make it challenging to compare outcomes across studies, although the key baseline characteristics that may have affected survival outcomes (age, cytogenics, and AML type) were similar across the nine trials [33,37,38,40,45,46,53,55,91] that demonstrated statistically significant differences in the OS outcome between study arms (Supplementary Figures 1–3). AEs are not uniformly reported in the trials included in our SLR. Nine trials [37,39,42–47,55] did not report rates of all-cause AEs, creating a potential gap and challenge in comparing AEs across studies. Additionally, two of the trials [39,43] reported rates of TRAEs instead of TEAEs. Lastly, five trials [31,32,36,54,56] showed a high risk of bias, and not all 26 included trials reported on certain outcomes of interest, such as TI and HRQoL outcomes.

Conclusion

Patients with AML who are ineligible for IC have historically had poor outcomes with the available treatment options. Novel therapies have improved clinical outcomes in this patient population, with targeted therapy combined with an HMA showing the most promising results. Adding ivosidenib or venetoclax to azacitidine or adding volasertib, venetoclax or glasdegib to LDAC significantly improved OS. In line with the most recent NCCN and ELN guidelines [6,15], ivosidenib in combination with azacitidine in mIDH1 AML and venetoclax in combination with azacitidine in the general AML population showed the most promising results. Some treatment-specific AEs highlighted in this SLR require attention during treatment, and management recommendations have been developed for these. Other important factors to consider for treatment decisions include patient's genetic characteristics, comorbidities, fitness, and social support. New treatment options have been and will continue to be incorporated in treatment guidelines for AML. The efficacy of the new treatment combinations for second-line treatment of patients ineligible for IC is largely unknown and has not been evaluated in this review. Thus, the sequential efficacy of treatments cannot be considered for the choice of the first-line treatment. However, targetable mutations may get lost at relapse in a variable proportion of patients, which may preclude the use of the targeting agent at relapse. Successful development of targeted agents in AML supports the approach of dissecting AML into well-described genetic subgroups and developing treatment strategies for each subgroup individually.