BRIDGING THE GAP: DEMOGRAPHIC REWEIGHTING OF REAL-WORLD SURVIVAL DATA VERSUS RCTS IN ACUTE MYELOID LEUKEMIA
Author(s)
Emma van Eijndhoven, MSc MA1, Dhanish Revanth Rangaswamy Nandakumar, MPH2, Ambarish J. Ambegaonkar, PhD2;
1APPERTURE LLC, Los Angeles, CA, USA, 2APPERTURE LLC, Marlboro, NJ, USA
1APPERTURE LLC, Los Angeles, CA, USA, 2APPERTURE LLC, Marlboro, NJ, USA
OBJECTIVES: Differences in demographic composition between randomized controlled trials (RCTs) and real-world populations pose challenges for assessing trial generalizability. Given the widespread use of demographic reweighting, this study explores its impact on survival differences between RCT and SEER cohorts across three pivotal AML trials (VIALE-A [venetoclax], AG120-C [ivosidenib], ADMIRAL [gilteritinib]).
METHODS: Three cohorts were constructed for each trial: (1) an RCT cohort, (2) an unadjusted SEER cohort, and a (3) demographically reweighted SEER cohort. The RCT cohorts were derived from reconstructed individual patient-level data for SOC/placebo arms using the Guyot et al. method. SEER cohorts included AML patients diagnosed in 2014-2016 with up to 5-year survival follow-up. SEER data were reweighted to each trial’s baseline age, sex, and race distribution using raking. Overall survival (OS) was summarized using Restricted Mean Survival Time (RMST) and compared across cohorts.
RESULTS: Demographic composition substantially differed between trial populations and SEER. Patients were older in VIALE-A and AG120 but younger in ADMIRAL relative to SEER. All three trials enrolled a higher proportion of Asian and a lower proportion of African American patients. RMST was significantly longer in unadjusted SEER cohorts than in RCT cohorts (12.3 months vs 9.4-10.9). Demographic reweighting materially altered SEER demographic profiles but did not consistently attenuate OS differences. Reweighted SEER RMST moved closer to RCT estimates for AG120-C (11.1 months) compared to the unweighted SEER data (12.3 months), while the discrepancy increased for ADMIRAL and VIALE-A (13.6 and 12.5 months, respectively). Age reweighting was not feasible for VIALE-A due to missing age categorizations.
CONCLUSIONS: Demographic composition varied across trials, and reweighting of real-world data did not consistently alter survival differences relative to RCTs. These findings suggest that demographic alignment alone is insufficient to ensure meaningful comparability of survival outcomes, underscoring the need for caution when relying on reweighted real-world data to assess RCT generalizability.
METHODS: Three cohorts were constructed for each trial: (1) an RCT cohort, (2) an unadjusted SEER cohort, and a (3) demographically reweighted SEER cohort. The RCT cohorts were derived from reconstructed individual patient-level data for SOC/placebo arms using the Guyot et al. method. SEER cohorts included AML patients diagnosed in 2014-2016 with up to 5-year survival follow-up. SEER data were reweighted to each trial’s baseline age, sex, and race distribution using raking. Overall survival (OS) was summarized using Restricted Mean Survival Time (RMST) and compared across cohorts.
RESULTS: Demographic composition substantially differed between trial populations and SEER. Patients were older in VIALE-A and AG120 but younger in ADMIRAL relative to SEER. All three trials enrolled a higher proportion of Asian and a lower proportion of African American patients. RMST was significantly longer in unadjusted SEER cohorts than in RCT cohorts (12.3 months vs 9.4-10.9). Demographic reweighting materially altered SEER demographic profiles but did not consistently attenuate OS differences. Reweighted SEER RMST moved closer to RCT estimates for AG120-C (11.1 months) compared to the unweighted SEER data (12.3 months), while the discrepancy increased for ADMIRAL and VIALE-A (13.6 and 12.5 months, respectively). Age reweighting was not feasible for VIALE-A due to missing age categorizations.
CONCLUSIONS: Demographic composition varied across trials, and reweighting of real-world data did not consistently alter survival differences relative to RCTs. These findings suggest that demographic alignment alone is insufficient to ensure meaningful comparability of survival outcomes, underscoring the need for caution when relying on reweighted real-world data to assess RCT generalizability.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
RWD155
Topic
Real World Data & Information Systems
Disease
SDC: Oncology