Researchers developed a model for predicting survival for patients with acute myeloid leukemia (AML) and presented details of the model in a new report in the journal Frontiers in Oncology.
The model is based on machine learning (ML) with a random forest algorithm that included gene expression and overall survival data from patients with AML. The model, which the researchers named Stellae-123 (ST-123) based on inclusion of 123 variables, was constructed using datasets from the Gene Expression Omnibus. The model training set came from the GSE37642 database and it contained data obtained from 562 patients in the phase 3 AMLCG-1999 study. The validation set came from the GSE68833 database, with data from 137 patients in The Cancer Genome Atlas (TCGA) dataset.
The researchers found that the variables that were of greatest importance in the model were age and expression of the genes KGM5B and LAPTM4B. The researchers explained these genes had previously been associated with carcinogenesis. LAPTM4B had also previously been associated with prognosis in myelodysplastic syndrome and some solid tumors.
Continue Reading
The model demonstrated high concordance in both the training and validation analyses (concordance indexes, 0.7228 and 0.6988, respectively). The researchers reported that the model continued to demonstrate similar concordance at a 5-year follow-up.
They noted the model was especially accurate at prognosis for those patients who showed the greatest risk of mortality. Model performance was not as strong for predictions related to the period of time in the days soon after diagnosis.
The model additionally showed potential at aiding in prognosis with high-risk mutations. This was analyzed in a subset of patients in the validation cohort who had high-risk mutations, such as a TP53 mutation and/or deletion in 16 patients, a RUNX1 mutation in 14 patients, or an ASXL1 mutation in 6 patients. When analyzing patients with these high-risk mutations, the model continued to show high concordance indexes for survival predictions.
The researchers considered the level of concordance with ST-123 to be higher than that for the European Leukemia Net risk classifier, and they determined it was especially useful for stratification of high-risk patients.
“In conclusion, our results indicate that survival of patients with AML can be predicted by applying ML tools to transcriptomic data, and that such predictions are particularly precise among patients with high-risk mutations,” the researchers concluded in their report.
Reference
Mosquera Orgueira A, Peleteiro Raíndo A, Cid López M, et al. Personalized survival prediction of patients with acute myeloblastic leukemia using gene expression profiling. Front Oncol. 2021;11:657191. doi:10.3389/fonc.2021.657191
