Construction And Validation Of Relapse Prediction Model For Patients With Non-genetic High-risk Acute Myeloid Leukemia

Background:Acute myeloid leukemia(AML)is a class of malignant clonal disorders of hematopoietic stem and progenitor cell origin.Although up to 85%of patients younger than 60 years achieve complete remission after standard induction therapy,only 35%40%can be fully cured.Whereas patients older than 60 years have a worse prognosis,with a cure rate of only 5%-15%.The European LeukemiaNet(ELN)genetic risk classification is the currently accepted standard method for prognostic stratification of patients with acute myeloid leukemia and includes only selected gene mutations and cytogenetic abnormalities,without considering gene expression condition.Most adult patients with acute myeloid leukemia fall into the intermediate risk group according to the ELN risk group classification.In the low-risk group,40%of patients with nuclear binding factor acute myeloid leukemia,50%of patients with NPM1 mutations,and 44%of patients with CEBPA biallelic mutations relapse in long-term follow-up.Because AML is highly heterogeneous,the actual risk profile of many patients is not optimally reflected in current genetic risk classification scheme.The high recurrence risk in patients with non-genetic high-risk AML suggests the need for more refined risk stratification strategy.Objective:Using bioinformatics methods,we analyzed the GSE146173 transcriptome sequencing dataset of Gene Expression Omnibus(GEO)database,explored the recurrence mechanism of non-genetic high-risk AML patients,screened key genes to construct a relapse prediction model for AML,and verified it with BEATAML1.0COHORT dataset.Methods:1.The expression matrix and clinical information of the GSE146173 datasets were downloaded from the GEO database.Gene expression data were analyzed using Wilcoxon rank sum test on R software to gain differentially expressed genes between relapse and non-relapse groups of non-genetic high-risk AML patients.The occurrence mechanism of recurrence in patients was explored by performing enrichment analysis of differential genes.2.Recurrence related genes were firstly screened by log rank test,univariate Cox regression,Lasso regression,and then the key genes were choosed by multivariate Cox regression to construct a relapse prediction model.Receiver Operating Characteristic Curve,calibration curve,and Kaplan-Meier survival analysis were used to evaluate the model,and the BEATAML1.0-COHORT dataset was used to externally validate the model.Results:1.Through the analysis,117 differentially expressed genes(15 upregulated genes and 102 downregulated genes)were found between the AML relapse and non-relapse groups.Enrichment analysis revealed that these genes were mainly associated with signaling receptor ligand binding,IL-17 signaling pathway,TNF signaling pathway and JAK-STAT signaling pathway.2.A total of 16 genes were selected by log rank test and univariate Cox regression,then 14 genes by Lasso regression and finally 9 genes(CXCL5,IGHV343,TPT1P10,SPATA12,ACTL10,ENPP7P1,E2F3P1,HNRNPA1P12,RN7SKP287)by multivariate Cox regression to construct a relapse prediction model.3.The AUC value of the ROC curve was 0.91.The calibration curve showed good agreement of the model.Survival analysis showed that the model could better predict the prognosis of patients.Univariate and multivariate analyses indicated that the model was an independent predictor of relapse in AML patients.Conclusion:1.This study developed a relapse prediction model for patients with non-genetic high-risk AML,which had good predictive ability,and the nomogram was drawn for convenient clinical use.2.In this study,we found that the differentially expressed genes between relapse and non-relapse groups were mainly enriched in signaling receptor ligand binding and signaling pathways,suggesting that they may be potential relapse mechanisms in patients with non-genetic high-risk AML.