AI-driven risk stratification for disease progression after allo-SCT in AML.

e18562 Background: Disease progression is the leading cause of allogeneic stem cell transplantation (allo-SCT) failure in acute myeloid leukemia (AML). Outcomes are influenced by patient, disease, donor, and transplant-related factors. This study utilized machine learning to identify risk factors for disease progression after allo-SCT in AML. Methods: We retrospectively reviewed adults (>18 years) who underwent their first allo-SCT for AML between January 2017 and June 2024. Patients without pretransplant molecular mutation or post-transplant chimerism data were excluded. Elastic Net machine learning was applied to identify risk factors for disease progression. Results: A total of 64 patients (median age: 55 years, range: 23–72; 54.7% male) were included. Fifteen patients (23%) had adverse-risk disease per ELN2022. The majority were in complete remission (CR) at the time of transplant (97%), with 44 (69%) in first complete remission (CR1). Donor types included HLA-matched unrelated (n=50), HLA-identical sibling (n=7), haploidentical (n=3), and HLA-mismatched unrelated (n=3). 34 (53.1%) patients received reduced-intensity conditioning (RIC), and 18 (28%) received post-transplant cyclophosphamide (PTCy) as part of GVHD prophylaxis. The median follow-up for surviving patients was 21.9 months (range: 1.0–101.2). Full donor chimerism at day 100 was achieved by 54 patients. During follow-up, 16 (25%) patients experienced disease progression, with a median time to progression of 21.9 months. Elastic Net analysis identified mixed chimerism at day 100 (HR 9.7), absence of PTCy (HR 1.5), HCT-CI >2 (HR 1.35), secondary AML (HR 2.04), DNMT3A mutation (HR 1.23), FLT3-ITD mutation (HR 1.47), and adverse ELN2022 risk (HR 2.47) as significant predictors of disease progression. Favorable factors included NPM1 mutation (HR 0.79) and the development of low-grade acute GVHD (HR 0.86). Conclusions: AI models improve risk stratification for disease progression after allo-SCT by analyzing multiple variables simultaneously. This approach enables personalized post-transplant strategies to enhance outcomes. Further validation in larger cohorts is warranted. Patient characteristics. Variable N (%age) Median Age At Transplant 55 (23-72) Gender Male 36 (56.2%) Female 28 (43.8%) Donor Type MUD 48 (79.9%) MRD 6 (10.1%) Haplo 3 (5%) MMRD 3 (5%) CMV Data (R vs D) Positive vs Positive, n (%) 21 (32.8%) Negative vs Negative, n (%) 12 (18.8%) Positive vs Negative, n (%) 30 (46.9%) Negative vs Positive, n (%) 1 (1.6%) Donor-recipient gender match Match 46 (76.6%) Mismatch 14 (23.4%) Full Chimerism at D100 54 (90%) Conditioning regimen MAC 30 (46.9%) RIC/Non-MAC 34 (53.1%) GVHD Prophylaxis PTCy 18 (30%) Others 42 (70%)