Survival prediction optimization of acute myeloid leukaemia based on T‐cell function‐related genes and plasma proteins

Interactions between acute myeloid leukaemia (AML) cells and immune cells are postulated to corelate with outcomes of AML patients. However, data on T-cell function-related signature are not included in current AML survival prognosis models. We examined data of RNA matrices from 1611 persons with AML extracted from public databases arrayed in a training and three validation cohorts. We developed an eight-gene T-cell function-related signature using the random survival forest variable hunting algorithm. Accuracy of gene identification was tested in a real-world cohort by quantifying cognate plasma protein concentrations. The model had robust prognostic accuracy in the training and validation cohorts with five-year areas under receiver-operator characteristic curve (AUROC) of 0.67-0.76. The signature was divided into high- and low-risk scores using an optimum cut-off value. Five-year survival in the high-risk groups was 6%-23% compared with 42%-58% in the low-risk groups in all the cohorts (all p values <0.001). In multivariable analyses, a high-risk score independently predicted briefer survival with hazard ratios of death in the range 1.28-2.59. Gene set enrichment analyses indicated significant enrichment for genes involved in immune suppression pathways in the high-risk groups. Accuracy of the gene signature was validated in a real-world cohort with 88 pretherapy plasma samples. In scRNA-seq analyses most genes in the signature were transcribed in leukaemia cells. Combining the gene expression signature with the 2017 European LeukemiaNet classification significantly increased survival prediction accuracy with a five-year AUROC of 0.82 compared with 0.76 (p < 0.001). Our T-cell function-related risk score complements current AML prognosis models.