A Five-Gene Signature for the Prediction of Event-Free Survival of Both Pediatric and Adult Acute Myeloid Leukemia.

Background

While adult and pediatric acute myeloid leukemia (AML) exhibit genetic distinctions, investigating their common gene expression program is critical for understanding the fundamental biological mechanisms that drive diverse cellular responses. However, existing gene signatures, predominantly tailored for overall survival (OS), may not adequately forecast event-free survival (EFS). EFS represents the time patients survive without disease recurrence, progression, or further treatment, a crucial metric for evaluating drug efficacy and assessing the clinical benefits of treatment.

Methods

We performed an integrated analysis of adult TCGA and pediatric TARGET expression datasets to pinpoint genes and pathways associated with EFS in both adult and pediatric AML. Additionally, we constructed a predictive model using one dataset and validated it on the other to unveil a novel gene signature.

Results

A five-gene signature comprising F2RL3, IL2RA, MYH15, SIX3, and SOBP was identified for EFS for both adult and pediatric AML. The test Area Under the ROC Curves (AUCs) for the 2-year and 5-year cutoffs of adult TCGA data were 0.851 (95% CI (Confidence Interval): 0.778-0.923) and 0.848 (95% CI: 0.729-0.968), respectively. Similarly, the test AUCs for the 2-year and 5-year cutoffs of Pediatric TARGET data were 0.725 (95% CI: 0.640-0.811) and 0.695 (95% CI: 0.59-0.80), respectively. When patients were stratified into three equal-sized prognostic subtypes based on the five-gene test risk scores, the P-values of tertile partitions for TARGET and TCGA data were 2.32e-6 and 5.12e-14, respectively, indicating superior performance compared to cytogenetic risk stratification within the same data (TARGET: P = 0.0019; TCGA: P = 0.0086). Despite being identified for EFS, the five-gene signature successfully stratified patients into distinct OS groups across two additional independent datasets.

Conclusions

This five-gene signature demonstrates robust performance in both EFS and OS risk prediction and might be clinically significant.