Abstract
Most patients with acute myeloid leukemia (AML) may obtain remission upon induction chemotherapy, but relapse is frequent and associated with poor survival. Previous prognostic models for outcomes after relapse lacked analysis of comprehensive molecular data. A validated prognostic model integrating clinical, cytogenetic, and molecular variables may support treatment decisions. We studied 943 patients with AML who relapsed after intensive induction treatment in a development cohort (HOVON-SAKK). A random survival forest algorithm was used to evaluate the association of clinical parameters, cytogenetic abnormalities, and molecular variables at diagnosis with overall survival (OS). Relapsing patients (n = 377) who were enrolled in the NCRI-AML18 trial were used for validation. In the development cohort, the median age at relapse was 58 years, and patients were classified as 2022 European LeukemiaNet favorable (22%), intermediate (31%), and adverse risk (48%). One-third underwent allogeneic transplantation in the first complete remission. Variable selection yielded 9 variables associated with 1-year OS, including relapse-free interval, age, white blood cell count, mutated TP53, FLT3 internal tandem duplication, core-binding factor abnormalities, t(v;11q23)/KMT2A rearrangement, and complex/monosomal karyotype, which were assigned points according to their estimated hazard ratios. Three prognostic groups were defined with distinct 1-year OS in both development (favorable, 51% ± 3%; intermediate, 29% ± 3%; and poor, 14% ± 2%, respectively) and validation cohorts (51% ± 4%, 26% ± 5%, and 14% ± 3%, respectively). Validation confirmed the improved accuracy in predicting outcomes for patients with AML in first relapse. The revised AML relapse model improved on previous prognostic models for outcomes after first relapse. It provides stratification that might support tailoring second line treatment.