Development and validation of a promising 5-gene prognostic model for pediatric acute myeloid leukemia

Mol Biomed. 2024 Jan 2;5(1):1. doi: 10.1186/s43556-023-00162-y.

Abstract

Risk classification in pediatric acute myeloid leukemia (P-AML) is crucial for personalizing treatments. Thus, we aimed to establish a risk-stratification tool for P-AML patients and eventually guide individual treatment. A total of 256 P-AML patients with accredited mRNA-seq data from the TARGET database were divided into training and internal validation datasets. A gene-expression-based prognostic score was constructed for overall survival (OS), by using univariate Cox analysis, LASSO regression analysis, Kaplan-Meier (K-M) survival, and multivariate Cox analysis. A P-AML-5G prognostic score bioinformatically derived from expression levels of 5 genes (ZNF775, RNFT1, CRNDE, COL23A1, and TTC38), clustered P-AML patients in training dataset into high-risk group (above optimal cut-off) with shorter OS, and low-risk group (below optimal cut-off) with longer OS (p < 0.0001). Meanwhile, similar results were obtained in internal validation dataset (p = 0.005), combination dataset (p < 0.001), two treatment sub-groups (p < 0.05), intermediate-risk group defined with the Children's Oncology Group (COG) (p < 0.05) and an external Japanese P-AML dataset (p = 0.005). The model was further validated in the COG study AAML1031(p = 0.001), and based on transcriptomic analysis of 943 pediatric patients and 70 normal bone marrow samples from this dataset, two genes in the model demonstrated significant differential expression between the groups [all log2(foldchange) > 3, p < 0.001]. Independent of other prognostic factors, the P-AML-5G groups presented the highest concordance-index values in training dataset, chemo-therapy only treatment subgroups of the training and internal validation datasets, and whole genome-sequencing subgroup of the combined dataset, outperforming two Children's Oncology Group (COG) risk stratification systems, 2022 European LeukemiaNet (ELN) risk classification tool and two leukemic stem cell expression-based models. The 5-gene prognostic model generated by a single assay can further refine the current COG risk stratification system that relies on numerous tests and may have the potential for the risk judgment and identification of the high-risk pediatric AML patients receiving chemo-therapy only treatment.

Keywords: C-index; Pediatric acute myeloid leukemia; Prognostic; Risk stratification.

Publication types

  • Research Support, Non-U.S. Gov't
  • Validation Study

MeSH terms

  • Adolescent
  • Biomarkers, Tumor / genetics
  • Child
  • Child, Preschool
  • Female
  • Gene Expression Profiling / methods
  • Gene Expression Regulation, Leukemic
  • Humans
  • Infant
  • Kaplan-Meier Estimate
  • Leukemia, Myeloid, Acute* / drug therapy
  • Leukemia, Myeloid, Acute* / genetics
  • Leukemia, Myeloid, Acute* / mortality
  • Male
  • Prognosis
  • Proportional Hazards Models

Substances

  • Biomarkers, Tumor