AML risk stratification models utilizing ELN-2017 guidelines and additional prognostic factors: a SWOG report

Abstract

Background: The recently updated European LeukemiaNet risk stratification guidelines combine cytogenetic abnormalities and genetic mutations to provide the means to triage patients with acute myeloid leukemia for optimal therapies. Despite the identification of many prognostic factors, relatively few have made their way into clinical practice.

Methods: In order to assess and improve the performance of the European LeukemiaNet guidelines, we developed novel prognostic models using the biomarkers from the guidelines, age, performance status and select transcript biomarkers. The models were developed separately for mononuclear cells and viable leukemic blasts from previously untreated acute myeloid leukemia patients (discovery cohort, N = 185) who received intensive chemotherapy. Models were validated in an independent set of similarly treated patients (validation cohort, N = 166).

Results: Models using European LeukemiaNet guidelines were significantly associated with clinical outcomes and, therefore, utilized as a baseline for comparisons. Models incorporating age and expression of select transcripts with biomarkers from European LeukemiaNet guidelines demonstrated higher area under the curve and C-statistics but did not show a substantial improvement in performance in the validation cohort. Subset analyses demonstrated that models using only the European LeukemiaNet guidelines were a better fit for younger patients (age < 55) than for older patients. Models integrating age and European LeukemiaNet guidelines visually showed more separation between risk groups in older patients. Models excluding results for ASXL1, CEBPA, RUNX1 and TP53, demonstrated that these mutations provide a limited overall contribution to risk stratification across the entire population, given the low frequency of mutations and confounding risk factors.

Conclusions: While European LeukemiaNet guidelines remain a critical tool for triaging patients with acute myeloid leukemia, the findings illustrate the need for additional prognostic factors, including age, to improve risk stratification.

Keywords: AML; Acute myeloid leukemia; Biomarkers; ELN; Elderly; European LeukemiaNet guidelines; Mathematical modeling; Model development and validation; Prognostic factors.

Fig. 1

Performance of ELN₂₀₁₇ model in Mononuclear cells. Overall Survival probability over time by ELN₂₀₁₇ risk group in MNCs from the validation cohort (n = 166). C-statistics are for the ELN₂₀₁₇ model fit to the validation cohort for all patients (age 18.5–88.8, a), patients younger than 55 years old (N = 86, b), and patients 55 years and older (N = 80, c). The total number of patients who were at risk of death (alive and uncensored) are shown for each year of follow-up

Fig. 2

Performance of ELN₂₀₁₇ model in Viable Leukemic Blasts. Overall Survival probability over time by ELN₂₀₁₇ risk group in VLBs from the validation cohort (n = 166). C-statistics are for the ELN₂₀₁₇ model fit to the validation cohort for all patients (age 18.5–88.8, a), patients younger than 55 years old (N = 86, b), and patients 55 years and older (N = 80, c). The total number of patients who were at risk of death (alive and uncensored) are shown for each year of follow-up

Fig. 3

Performance of AGE + ELN₂₀₁₇ Model in Mononuclear Cells. Overall Survival probability over time as predicted by the AGE + ELN₂₀₁₇ models developed using the discovery cohort in MNCs. The continuous risk score from the AGE + ELN₂₀₁₇ model in the discovery cohort was divided into quartiles and the boundaries of these quartiles were used to define a four-level categorical variable. A model was fit using this categorical variable in the validation cohort for all patients (N = 166, age 18.5–88.8, a), patients younger than 55 years old (N = 86, b), and patients 55 years and older (N = 80, c). There were no patients younger than 55 in 3rd and 4th quartiles (b) or patients older than 55 in 1st quartile (c). The total number of patients who were at risk of death (alive and uncensored) are shown for each year of follow-up

Fig. 4

Performance of AGE + ELN₂₀₁₇ Model in Viable Leukemic Blasts. Overall Survival probability over time as predicted by the AGE + ELN₂₀₁₇ models developed using the discovery cohort in MNCs. The continuous risk score from the AGE + ELN₂₀₁₇ model in the discovery cohort was divided into quartiles and the boundaries of these quartiles were used to defined a four-level categorical variable. A model was fit using this categorical variable in the validation cohort for all patients (N = 166, age 18.5–88.8, a), patients younger than 55 years old (N = 86, b), and patients 55 years and older (N = 80, c). There were no patients younger than 55 in 4th quartiles (b) or patients older than 55 in 1st quartile (c). The total number of patients who were at risk of death (alive and uncensored) are shown for each year of follow-up