Identification and Characterization of Oxidative Stress-Related Prognostic Biomarkers in Acute Myeloid Leukemia Using Integrated Bioinformatics Analysis

Oxidative stress contributes to acute myeloid leukemia (AML) biology and treatment resistance, but its prognostic relevance remains incompletely defined. We analyzed 553 patients from GSE37642, treating paired GPL96/GPL97 measurements from 417 patients as the discovery cohort and GPL570 (n = 136) as a platform-defined internal–external validation cohort. Human genes annotated to cellular response to oxidative stress (GO:0034599) were retrieved from the Enrich GO Biological Process 2023 library; 110 genes were measurable in both cohort definitions. Discovery-only preprocessing was frozen before validation. Seventeen genes with nominal univariate Cox p < 0.05 entered cross-validated LASSO–Cox modeling, which retained a 13-gene signature (ATP13A2, AKT1, DAPK1, HSPA1A, LRRK2, CPEB2, CAMKK2, CCS, PRKAA1, RBM11, TP53, GPX1, and TPM1). Using the fixed discovery-derived coefficients and median cutoff (0.00807), the score separated high- and low-risk patients in discovery (log-rank p = 3.51 × 10−12; continuous-score HR = 3.24, 95% CI 2.46–4.29; C-index = 0.627, 95% CI 0.591–0.661) and remained associated with survival in platform-defined validation (log-rank p = 0.019; HR = 1.90, 95% CI 1.20–3.00; C-index = 0.580, 95% CI 0.519–0.641). Validation AUCs were 0.590, 0.621, and 0.655 at 1, 3, and 5 years. A fully nested five-fold diagnostic yielded a mean held-out C-index of 0.571. Adding the risk score to age improved the nested-model fit in validation (likelihood-ratio p = 0.028), but the improvement in discrimination was small (Delta C-index = 0.019) and its paired out-of-bag bootstrap 95% CI included zero (−0.070 to 0.056); corresponding Delta AUC intervals also included zero. Schoenfeld diagnostics did not indicate proportional-hazards violations. These findings support a biologically interpretable oxidative-stress-associated prognostic signal with modest cross-platform discrimination, while emphasizing the need for independent contemporary validation before clinical use.