Can transcriptome-based risk modeling improve survival stratification across cytogenetic groups in acute myeloid leukemia?

Abstract

6533 Background: Cytogenetic risk stratification is central to prognostication in acute myeloid leukemia (AML), yet survival outcomes vary considerably within each group. We hypothesized that transcriptomic features could further stratify survival within these cytogenetic categories, independent of clinical factors. Methods: We analyzed gene expression microarray data for 172 AML patients from the TCGA-LAML cohort (dbGaP phs000178). Multivariable Cox proportional hazards modeling incorporating age, gender, and CALGB cytogenetic risk was used to identify genes independently associated with overall survival. LASSO regression was applied to derive a transcriptomic gene signature and compute a patient-level transcriptomic risk score. Kaplan–Meier survival analysis, log-rank testing, and Cox modelling were performed in the overall cohort and within each cytogenetic subgroup. Model performance was assessed using concordance index (C-index) and likelihood ratio testing. Differential expression and Gene Ontology Biological Process (GOBP) enrichment were performed between high and low transcriptomic risk groups. Results: The transcriptomic risk score significantly stratified overall survival and, importantly, retained prognostic significance within favorable, intermediate, and poor cytogenetic risk groups (Table). The concordance index improved from 0.69 with clinical variables (age, gender, cytogenetics) alone to 0.82 after adding the transcriptomic risk score. Likelihood ratio testing confirmed a highly significant improvement in model fit with the inclusion of the transcriptomic score (χ² = 98.7, p < 2.2×10⁻¹⁶). GOBP enrichment of DEGs between high and low transcriptomic risk groups revealed upregulation of CD4-positive T-cell activation (STAT6, IL12RB1, ZBTB7B), negative regulation of immune response (LGALS9, PTPN6, FGL2), regulation of leukocyte and lymphocyte differentiation (ZC3H12A, VNN1, HCLS1), and response to bacterial molecules and lipopolysaccharide (NLRP3, MYD88, CASP1) confirming dysregulated immune signaling as a hallmark of adverse prognosis. Conclusions: A transcriptomic risk score provides powerful, independent prognostic information beyond cytogenetics and clinical factors in AML. These findings support the integration of transcriptomic biomarkers into existing risk models to refine AML prognostication. Survival association of transcriptomic risk score across CALGB groups in AML patients. CALGB Group HR* (95% CI) p value** All AML patients (N =172) 3.74 (2.48–5.63) 2.81×10⁻¹⁰ Favourable (N= 34) 14.53 (1.87–112.83) 1.05×10⁻² Intermediate (N = 97) 6.49 (3.43–12.26) 8.43×10⁻⁹ Poor (N = 38) 2.68 (1.26–5.69) 1.01×10⁻² *HR: High vs Low genomic risk. **Cox p from multivariable Cox model.