Validation of the 2022 European Leukemianet Genetic Risk Stratification of Acute Myeloid Leukemia

Background In 2022, an update of the widely adopted European LeukemiaNet (ELN) risk stratification system for AML has been published (Döhner et al. Blood 2022). So far, this new classification has not been validated in large and homogeneously treated patient cohorts. We set out to test the prognostic relevance of the ELN-2022 classification in intensively treated AML patients, and to compare it to ELN-2017.

Methods We studied 1,138 newly diagnosed AML patients (median age 57 years; range, 18-86 years) who received cytarabine-based induction chemotherapy in two subsequent multicenter phase III trials of the German AML Cooperative Group (AMLCG-1999, clinicaltrials.gov identifier NCT00266136, n = 857; and AMLCG-2008, NCT01382147, n = 259) between 1999 and 2012. Twenty subjects were excluded due to missing cytogenetics or FLT3-ITD ratio. None of the patients received FLT3 inhibitors or gemtuzumab ozogamicin during first-line treatment. Median follow-up for of survivors was 98 months (m). Findings were validated in a cohort of 1109 mostly younger AML patients treated on trials of the AML-SG study group.

Results Out of 1,118 patients classified according to ELN-2022, 363 (32%) were classified as favorable, 301 (27%) as intermediate, and 454 (41%) as adverse. For those ≥ 60a of age (n=518) the distribution was 20%, 23% and 57%. Compared to ELN-2017 15% of patients (n=171) were classified into a different risk category (3% into a more favorable and 12% into a less favorable category; Fig. 1).

According to ELN-2022, median OS was 9.5y (95% CI, 4.8-12.3y), 1.6y (95% CI, 1.8-2.0y) and 0.8y (95% CI, 0.7-1.0y) for favorable, intermediate and adverse risk groups. For the ELN-2017 categories, corresponding mean OS was 8.2 y (95% CI, 4.6-11.9y), 1.7y (95% CI, 1.2-2.0y) and 0.8y (95% CI: 0.7-0.9y). 5y OS by risk group was 55%, 34% and 15% for ELN-2022 and 54%, 31% and 12% for ELN 2017. Complete response rates by risk group were 73%, 65% and 45% for ELN-2022 and 72%, 66% and 41% for ELN-2017

As in ELN-2017, ELN-2022 adverse risk significantly associated with age, male sex, sAML and WBC count at diagnosis (p<0.0001 for all). These associations also held in patients aged <60 years (n = 518) or ≥60 years (n = 600). We did not find a significant association between the ELN-2022 adverse risk group and tAML (p=0.21)

Overall, time-dependent ROC-curves (Heagerty, Lumley and Pepe, 2000) show slightly better prognostic discrimination for ELN-2017 compared to ELN-2022 for OS at one year (AUC 0.66 vs. 0.65) and five years (AUC 0.72 vs. 0.71). Harrel's C-index confirms this finding, with values of 0.65 for ELN-2022 and 0.66 for ELN-2017 when controlling for age, sex and sAML.

Several aspects may contribute to the slight reduction of prognostic accuracy we observed: The 61 patients reclassified from favorable to intermediate-risk had significantly higher 5y OS rates than other intermediate patients (30% v. 48%). In contrast, patients reclassified from adverse to intermediate-risk (n = 21) had a worse 5y OS rate than other intermediate patients (36% v. 10%). Interestingly, their outcome was fairly similar to those reclassified from intermediate to adverse-risk (n=70), whose 5y OS appeared better than that of other adverse-risk patients (median OS 1.4 v 0.8y).

Sixty-five patients reclassified from intermediate to adverse-risk were reassigned based on the inclusion of additional myelodysplasia-related mutations (BCOR, EZH2, SF3B1, SRSF2, STAG2, U2AF1, ZRSR2) as adverse-risk markers. Therefore, we analyzed patients re-classified from favorable (n=10) or intermediate (n=65) to adverse-risk due to presence of one or more of these mutations. This cohort (n =75) had significantly better 5y OS (28 % v 13%) than patients with mutations in ASXL1, RUNX1 or TP53, or other adverse-risk genotypes, and seems to behave more like the intermediate cohort (Fig. 2). Median OS was significantly better as well (1.7 v 0.7y). This finding was confirmed in a validation cohort of 1109 patients. Here, we found a median OS of 1.6 v 1.1y).

Conclusions We conclude that the ELN-2022 risk classification identifies a larger group of adverse-risk patients, at the cost of slightly reduced prognostic accuracy compared to ELN-2017. This may in part be driven by the outcomes of patients re-assigned to the adverse-risk cohort based on the presence of myelodysplasia-associated gene mutations, which in our cohort had better outcomes than previously recognized adverse-risk subgroups.

Krug:AbbVie: Honoraria; , Sanofi: Honoraria; , Leo Pharma,: Honoraria; BMS: Honoraria. Metzeler:Daiichi Sankyo: Honoraria; Celgene/BMS: Consultancy, Honoraria, Research Funding; Novartis: Consultancy; Astellas: Honoraria; Pfizer: Consultancy; Jazz Pharmaceuticals: Consultancy; AbbVie: Honoraria.