The ER Stress-related Gene Prognostic Signature for Predicting Chemosensitivity and Prognosis in AML.

Abstract

Introduction: Acute myeloid leukemia is characterized by high heterogeneity, and the current European Leukemia Net (ELN) risk stratification system is not universally applicable to all AML patients, requiring approximately three weeks for testing.

Aim: This study aimed to develop an applicable prognostic tool capable of addressing the limitations of current methods. We selected AML patients from the clinic and TCGA database to explore the role of ER stress in response to chemotherapy.

Methods: Patients from the TCGA database were employed as the training cohort, and two GEO datasets were used as external validation cohorts. Univariate/multivariate COX and LASSO regression were exemplified to establish the prognostic model. Kaplan-Meier and timedependent ROC were used to assess and compare the efficiency of the model with ELN stratification and other models. In the training cohort, we selected 5 ER stress-related genes to predict chemosensitivity and establish the ERS-5 prognostic model.

Results: The model successfully predicted the overall survival of patients (p < 0.0001, HR = 4.86 (2.79-8.44); AUC = 0.83). It was verified in validation cohorts and could further stratify the risk of various AML subgroups. It also enhanced the ability of ELN to predict the response of patients with AML to main chemotherapeutic drugs. Finally, an "ERS-5" risk score was constructed by the nomogram based on the ERS-5 model and age.

Conclusion: Consequently, in this study, the ERS-5 model was constructed, which allowed more rapid (about 3 hours) and accurate risk stratification and complemented the ability of ELN to assess chemosensitivity.