A total of 536 patients with primary AML were included in this study between September 2013 and February 2021 from the First Affiliated Hospital of Xi'an Jiaotong University. The patients were diagnosed with AML according to the 4th or 5th version of the WHO guidance (4, 5) and followed up until July 2022. General clinical characteristics [age, sex, the proportion of blasts in bone marrow (BM) samples, karyotype, white blood cell (WBC) counts, red blood cell (RBC) counts, hemoglobin, blood platelet counts (PLT), activated partial thromboplastin time (APTT), prothrombin time, thrombin time, fibrinogen, fibrinogen degradation product, D-dimer, and survival time], structural variations [5/5q deletion (-5/-5q),−7/-7q, inv (involvement) (16)(p13q22), t (translocation) (16;16)(p13;q22), t(8;21)(q22;q22), t(9;22)(q34.1;q11.2), inv(3)(p21p26), and t(3;3)(q21;q26)], and fusion genes (CBFB-MYH11, BCR-ABL1, KMT2A-PARTNER, TLS-ERG, PML-RARA, RUNX1-RUNXT1, NUP98-PARTNER, DEK-NUP214, FIPIL1-PDGFRα, AML1-ETO, NPM1-RARα, PLZF-RARα, and SET-NUP214) were collected and shown as Table 1. After manual evaluation of the karyotype, fusion, and mutation, patients were assigned to respective 4th/5th WHO and ELN 2017/2022 risk groups. This study was approved by the First Affiliated Hospital of Xi'an Jiaotong University. Informed consent forms were signed by each patient.

Genomic DNA (gDNA) was extracted from the formalin-fixed paraffin-embedded (FFPE) or fresh BM samples, followed by the targeted sequencing of 38 genes (ASXL1, BCOR, BCORL1, CBL, CEBPA, CSF3R, DNMT3A, ETV6, EV11, MECOM, EZH2, FLT3, FLT3-ITD, FLT3-TKD, GATA2, HOX11, IDH1, IDH2, JAK2, KIT, KMT2A, KMT2A-PTD, KRAS, MPL, MYC, NF1, NPM1, NRAS, NTRK3, RUNX1, SF3B1, SH2B3, SRSF2, TET2, TP53, TPMT, U2AF1, and ZRSR2) on the Novaseq (Illumina, USA) sequencing platform. The original sequencing was aligned with the human reference genome GRCh37. Single nucleotide variations (SNVs) and insertion and deletion (Indels) were screened by Shanghai Rightongene Biotechnology Co., Ltd. (Shanghai, China) based on the filtering conditions: (1) SNVs or Indels with a mutation allele frequency (MAF) of ≥0.001 in databases of 1,000 genomes project, 1,000 genome East Asian, ExAC all, or ExAC East Asian were removed; (2) SNVs or Indels with a variant allele frequency (VAF) of ≥ 1% were retained; (3) dbSNP (v147) sites present in COSMIC database were retained; and (4) SNPs or Indels including stopgain, stoploss, frameshift, non-frameshift, and splicing sites were retained.

To establish the training and test cohorts, 536 samples were randomly divided into two groups, the training/validation set (70%) and the test set (30%), respectively. The training set was subjected to 10-fold cross-validation to account for variability and provide risk estimates. The mutated genes (ASXL1, BCOR, BCORL1, CBL, CEBPA, CSF3R, DNMT3A, ETV6, EV11, MECOM, EZH2, FLT3, FLT3-ITD, FLT3-TKD, GATA2, HOX11, IDH1, IDH2, JAK2, KIT, KMT2A, KMT2A-PTD, KRAS, MPL, MYC, NF1, NPM1, NRAS, NTRK3, RUNX1, SF3B1, SH2B3, SRSF2, TET2, TP53, TPMT, U2AF1, and ZRSR2), rearrangements (AML1-ETO, BCR-ABL1, CBFB-MYH11, DEK-NUP214, KMT2A rearrangement, MECOM rearrangement, NPM1-RARα, NUP98 rearrangement, PLZF-RARα, RUNX1-RUNXT1, and SET-NUP214), and clinicopathologic features (age, sex, proportion of BM blasts, hemoglobin, WBC, and PLT) were included in the models. Conventional Cox regression was used to train the models for assessing survival with the selected variables by Lasso using the “Glmnet” package. The optimal cutoff values for risk score were determined using the X-tile software (Version 3.6.1, Yale University, USA) (8); thereafter, the patients were divided into favorable, intermediate, and adverse groups.

The statistical analysis was performed using GraphPad Prism software v.6 (GraphPad, San Diego, CA, United States). Kaplan–Meier curves with log-rank tests were used to analyze the OS and PFS of AML patients in different groups. A P-value of < 0.05 was considered significant.

The 5th WHO classification made some changes in the diagnostic criteria of AML, including (1) persons with KMT2A, MECOM, and NUP98 rearrangements and NPM1 mutation were diagnosed with AML regardless of the percentage of blasts; (2) the definition of AML with CEBPA mutation was changed to include both biallelic (biCEBPA) and single mutation of CEBPA located in the basic leucine zipper (bZIP) region (smbZIP-CEBPA); (3) the previous classification of AML with mutated RUNX1 was abolished; (4) the classification of AML with myelodysplasia-related changes (AML-MRC) has been changed to AML-MR, and the mutations in ASXL1, BCOR, EZH2, SF3B1, SRSF2, STAG2, U2AF1, and ZRSR2 genes were newly defined as the defining cytogenetic abnormalities; (5) the classification of AML with other defined genetic alterations was newly added; and (6) the classification of AML, not otherwise specified (NOS) was replaced with AML, defined by differentiation.

A total of 485 and 487 of the 536 patients were classified according to the 4th and 5th WHO classifications, respectively, as the mutation location of CEBPA or the VAF of FLT3-ITD is not available. Six subgroups, AML with RUNX1::RUNX1T1 (n = 70), AML with CBFB::MYH11 (n = 29), AML with DEK::NUP214 (n = 5), AML with BCR::ABL1 (n = 4), AML with GATA2::MECOM (n = 2), and AML with NPM1 (n = 89), remained unchanged according to the 5th WHO, although the subgroup of AML with GATA2::MECOM was changed to AML with MECOM rearrangement. The group that changed the most was the AML, NOS (not otherwise specified) subgroup (n = 212); in detail, 8 patients were divided into the AML with KMT2A-rearrangement subgroup, 25 patients were divided into the AML-MR subgroup, and one case was divided into the AML with NUP98-rearrangement subgroup based on the 5th WHO classification (Figure 1A, Table 2).

Diagnostic criteria were largely unchanged in the new proposal, except for the following: (1) persons with FLT3-ITD were classified as intermediate regardless of the VAF; (2) the previously defined favorable risk of biCEBPA was changed to bZIP in-frame mutated CEBPA; (3) KAT6A::CREBBP and mutated ASXL1, BCOR, EZH2, RUNX1, SF3B1, SRSF2, STAG2, U2AF1, or ZRSR2 were added as adverse events; and (4) VAF ≥ 10% was defined as an additional condition for the adverse event of TP53 mutation.

A total of 483 and 498 patients were submitted to risk stratification based on the 2017 and 2022 ELN guidance, respectively. The stratification for most patients (442/501) according to the 2022 ELN guidance was consistent with the 2017 ELN. In contrast, 16 patients in the favorable group based on the 2017 ELN guidance were regrouped into the intermediate group based on the 2022 ELN guidance due to the low VAF value of FLT3-ITD and the lack of the bZIP in-frame mutated CEBPA. In addition, 13 patients in the intermediate group based on the 2017 ELN guidance were regrouped into the adverse group based on the 2022 ELN guidance due to the mutations of BCOR, SRSF2, U2AF1, and ZRSR2. Seventeen unclassified patients (the detailed information of CEBPA and FLT3-ITD unknown) based on the 2017 ELN guidance were regrouped into the intermediate group (Figure 1B, Table 3).

Moreover, we evaluated the PFS and OS of subgroups based on the WHO classification and ELN guidance. Both PFS and OS rates were higher in AML patients with RUNX1::RUNX1T1, CBFB::MYH11, and biCEBPA, while the PFS and OS rates were lower for patients with MLLT3::KMT2A and GATA2::MECOM according to the 4th WHO classification (Figure 2A). Similarly, the PFS and OS rates for patients with RUNX1::RUNX1T1, CBFB::MYH11, and biCEBPA were higher, and the prognosis for patients of the AML-MR and AML defined by differentiation subgroups was worse according to the 5th WHO classification (Figure 2B). From the ELN subgroups, both PFS and OS curves of patients of the intermediate and adverse groups were not very distinguishable according to the 2017 (Figure 3A) and 2022 ELN guidance (Figure 3B), although the OS and PFS curves for the favorable group and other groups could be distinguished well. Also, we assessed whether the therapeutic means affected the prognosis of AML patients. Hematopoietic stem cell transplantation (HCT) significantly improved the PFS and OS of patients in favorable, intermediate, and adverse groups as compared with the non-HCT patients. However, the PFS and OS curves for the non-HCT patients of the intermediate and adverse groups remained not very distinguishable (Figure 3C). These results affirmed the clinical value of both WHO and ELN, but a better prognosis model should be established in Chinese cohorts.

To establish a better prognosis model for Chinese patients with newly diagnosed AML, mutation, rearrangements, and clinicopathologic features were considered. For the PFS model, risk score was first calculated according to the following formula, risk score = 0.324134^*sex (male = 1, female = 2) + 0.025315^*age- 0.88687^*CBFB:: MYH11-0.801312^*NPM1+0.89387^*RUNX1-0.757403^*RUNX1::RUNX1T1+1.244844^*SH2B3+1.120662^*TP53, in which positive was defined as “1” and negative was defined as “0” in terms of gene mutation and fusion. Then, the patients were divided into three groups, favorable (risk score ≤ 1.80), intermediate (risk score < 1.8 < 2.37), and adverse (risk score ≥ 2.37). As shown in Figure 4A, the PFS time of the favorable group was obviously longer than the intermediate and adverse groups, as well as the intermediate group vs. the adverse group in both training and test sets. Furthermore, the nomogram demonstrated the contributions of the selected factors to the 1-, 3-, and 5-year PFS probability (Figure 4B).

For the OS model, the risk score was first calculated using the formula = 0.319677^*sex (male = 1, female = 2) + 0.027509^*age- 0.883355^*CBFB::MYH11-0.84456^*NPM1+0.870257^*RUNX1-0.73479^*RUNX1::RUNX1T1+1.240181^*SH2B3+1.106195^*TP53. After calculating the risk score of each patient, the patients were divided into three groups, favorable (≤ 1.94), intermediate (< 1.94 < 2.33), and adverse (≥2.33). The PFS time of the favorable group was obviously higher as compared with the intermediate and adverse groups, as well as the intermediate group vs. the adverse group in both the training set and test set (Figure 4C). In addition, the nomogram demonstrated the contributions of the selected factors to the 1, 3, and 5-year OS probability (Figure 4D). Collectively, we constructed a prognosis model for Chinese AML patients through the integration of the mutations, fusions, and clinical information.