Outcome of Intensively Treated Elderly AML Patients Reported to the Harmony Alliance Compares Well to Outcome of Control Patients of the Prospective Randomized HOVON 103 Study in Elderly AML

Introduction: The evaluation of novel drugs in hemato-oncology is hampered by relatively large sample sizes needed for sufficiently powered, randomized controlled trials (RCT). External control data are increasingly applied in prospective phase II and III studies, and might be derived from Real-World Data (RWD) sources, such as national cancer registries. The HARMONY Alliance recently reported on a big data platform, consisting of data that were generated by cooperative European acute myeloid leukemia (AML) study groups ( Lang et al., Trials, 2020). HARMONY or cancer registry data might supplement prospectively collected trial control data. Here, we investigated whether and to what extent data included in the HARMONY AML database and RWD from the Netherlands Cancer Registry (NCR) compare to the control arm of the recently reported prospective randomized HOVON-103 (H103) study in elderly AML ( Ossenkoppele et al., Leukemia, 2020; Janssen et al., Leukemia, 2022).

Methods: The H103 trial consecutively randomized newly diagnosed AML patients aged >65 years between standard induction chemotherapy with or without lenalidomide, tosedostat, or selinexor. External control patients concerned newly diagnosed (HARMONY: 2010-2018; NCR: 2014-2018) and intensively treated (intensified cytarabine/anthracycline) elderly AML patients aged >60 years. H103 were matched 1:1 using nearest neighbor propensity scores with NCR and HARMONY patients for age, European Leukemia Net (ELN) 2022 risk classification, and leukocyte count at diagnosis, while HARMONY patients were additionally matched for WHO performance score. Patients with missing values for propensity score factors were excluded from that analysis. The primary endpoint was overall survival (OS), calculated by the Kaplan-Meier method.

Results: 67% of HARMONY patients (n=510) concerned trial patients. Median age was lower for HARMONY patients compared with H103-controls (n=279) (67 vs. 69 years, P<0.01), with also a preponderance of male patients in the H103 cohort (54% vs. 62%, P=0.03). Age and male sex did not differ between H103-controls and NCR patients (n=320) (69 vs. 69 years, P=0.46; 62% vs. 62%, P=1.0) (Table 1). WHO performance score of 0 or 1 was reported in 49% and 41% of H103-controls which was less frequent in HARMONY and NCR data (8% and 31%, P<0.01; 26% and 18%, P<0.01, respectively). Unknown WHO status was more prevalent in the external cohorts than in H103, at 43%, 54%, and 1% for the HARMONY, NCR, and H103 cohorts, respectively. ELN 2022 risk (favorable vs. non-favorable) was reported in 24% vs. 76% of H103 patients, which was similar in HARMONY patients (26% vs. 74%; P=0.73). ELN 2022 risk differed, however, for NCR patients (17% vs 83%; P=0.03). The median leukocyte count at diagnosis was lower for H103-controls than for HARMONY and NCR patients (3.65 vs. 17.1, P<0.01; 3.65 vs. 13.1, P<0.01, respectively). Propensity scores matched 278 HARMONY patients and 278 NCR patients to H103-controls. The 2-year OS (estimate±SE) was 40±3% for H103-controls vs. 35±3% for HARMONY patients (P=0.22), and vs. 30±3% for NCR patients (P=0.052), respectively (Figure 1).

Conclusion: Characteristics of H103-controls and HARMONY patients appeared largely comparable, while NCR patients were more frequently ELN 2022 non-favorable risk. Following matching, OS was similar between H103-controls and HARMONY patients, while NCR patients showed inferior OS. The inclusion of a large proportion of trial patients in the HARMONY cohort, adhering to strict inclusion and exclusion criteria and fewer comorbidities compared to RWD patients, might explain these observations. These results suggest that matched data of the HARMONY Alliance might supplement prospectively collected control data in studies evaluating intensive therapy in elderly AML.