OBJECTIVES: This study employed a derivation approach to identify rw disease progression in AML, as defined by induction failure and relapse events.

METHODS: Patients with AML diagnosed between 1/1/2014 and 3/31/2022 from the nationwide Flatiron Health EHR-derived de-identified database were included in the study. Using AML consensus guidelines, dIFR events were determined when one or more of the following was present: bone marrow biopsy (BMBx) blasts >5%, documented provider assessment of BMBx was “Failure to Respond'' or “Relapse”, or peripheral blood blasts >0%. Performance of the dIFR variable was assessed through inter-rater agreement (i.e., event and date agreement via duplicate abstraction), association with downstream events (i.e., new therapy start/current therapy stop, drug change within a line, or death) and rw overall survival (rwOS) and rw event-free survival (rwEFS) and their correlation in the overall cohort and clinically distinct subgroups (e.g., age, cytogenetic risk).

RESULTS: Of the total 6781 patients, demographic and clinical characteristics were comparable to the AML population in the US SEER database. Inter-rater agreement for dIFR event and date were strong (>97%, >90%). The proportion of patients with a dIFR event after the first line initiation (64%) and those with downstream event(s) (53%, predominantly being a new therapy start) align with clinical expectations. Median rwOS, rwEFS were 14, 4.7 months, respectively. There was alignment among poor prognostic subgroups having a shorter median rwOS and rwEFS vs better prognostic subgroups. Spearman’s rank correlation between rwOS and rwEFS was strong (⍴ = 0.7).

CONCLUSIONS: Performance of the novel dIFR variable was strong, as evidenced by inter-rater agreement and alignment with clinical expectations of downstream events and rw endpoint analyses. This study demonstrates the feasibility of a derived rw progression approach, which can unlock the ability to conduct rw outcome studies across large AML cohorts.