Comparability of external and internal control patients for the prospective randomized HOVON‐103 trial in older AML patients

• DOI: 10.1111/bjh.20185
• Publication Date: 2025-05-27T00:08:38Z
• Authors: Sjoerd J. F. Hermans; Jurjen Versluis; Erik D. van Werkh...; Yvette van Norden; Jeroen J. W. M. J...; Gerwin A. Huls; Thomas Pabst; Dimitri A. Breems; Elizabeth Berkx; Avinash G. Dinmoh...; Peter C. Huijgens; Eric Sträng; Jesús María Herná...; Marta Sobas; Rosa Ayala Diaz; Joaquin Martinez ...; Klaus H. Metzeler; Torsten Haferlach; Christian Thiede; Carin A. Uyl‐de G...; Lars Bullinger; Bob Löwenberg; Gert J. Ossenkoppele; Francesco Pignatti; Jan J. Cornelissen

SummaryReal‐world data (RWD) previously contributed to post‐marketing regulatory decision‐making, but are currently also considered as external controls to single‐arm trials. The use of RWD control data may be compromised by methodological issues, urging validation of RWD control cohorts. Two external control cohorts of newly diagnosed acute myeloid leukaemia patients, one registered by the HARMONY Alliance (HA) and one by the Netherlands Cancer Registry (NCR), were compared to the control arm of the randomized HOVON‐103 trial (H103 controls). All patients, aged >65 years with a WHO performance score of 0–2 (or missing), received standard induction chemotherapy. 1:1 propensity score calliper matching (PSM) was applied to improve comparability, and overall (OS) and relapse‐free survival (RFS) were assessed. Fewer data elements were available in external cohorts compared to H103 controls, specifically in the NCR cohort. Baseline characteristics of the external cohorts differed from H103 controls; missing data were also more frequent and predominantly concerned WHO performance score. After PSM, HA patients demonstrated non‐significantly different OS and RFS to H103 controls at 2 years (26 ± 4% vs. 31 ± 5%, p = 0.59; 24 ± 5% vs. 30 ± 6%, p = 0.52), while NCR patients had 12% lower OS (28 ± 4% vs. 40 ± 4%, p = 0.21). Validation of external control cohorts is needed before incorporating RWD control data into comparative analyses, as missing data, specifically comorbidities, and residual confounding may limit comparability.