Distinguishing AML from MDS: a fixed blast percentage may no longer be optimal

Abstract

Patients with acute myeloid leukemia (AML) have conventionally received more intense therapy than patients with myelodysplastic syndrome (MDS). Although less intense therapies are being used more often in AML, the dichotomy between AML and MDS remains, with the presence of ≥20% myeloblasts in marrow or peripheral blood generally regarded as defining AML. Consequently, patients with 19% blasts are typically ineligible for AML studies, and patients with 21% blasts are ineligible for MDS studies. Here we cite biologic and clinical data to question this practice. Biologically, abnormalities in chromosome 3q26 and mutations in NPM1 and FLT3, regarded as AML associated, also occur in MDS. The genetic signatures of MDS, particularly cases with 10% to 19% blasts (MDS-EB2), resemble those of AML following a preceding MDS (secondary AML). Mutationally, secondary AML appears at least as similar to MDS-EB2 as to de novo AML. Patients presenting with de novo AML but with secondary-type AML mutations seem to have the same poor prognosis associated with clinically defined secondary AML. Seattle data indicate that after accounting for European LeukemiaNet 2017 risk, age, performance status, clinically secondary AML, and treatment including allogeneic transplantation, patients with World Health Organization-defined AML (n = 769) have similar rates of overall survival, event-free survival, and complete remission (CR)/CR with incomplete hematologic recovery as patients with MDS-EB2 (n = 202). We suggest defining patients with 10% to 30% blasts (AML/MDS) as eligible for both AML and MDS studies. This would permit empiric testing of the independent effect of blast percentage on outcome, allow patients access to more therapies, and potentially simplify the regulatory approval process.

Figure 1.

OS, EFS, and RFS for CR/CRi subgroup and CR without MRD subgroup. OS (A); EFS (B); RFS for CR/CRi subgroup (C); CR without MRD subgroup (D). The y-axes show probabilities of indicated outcomes. Univariate log-rank P values are as indicated.