VALIDATION OF EARLY-STAGE CANCER CLINICAL OUTCOMES FOR MELANOMA AND BREAST CANCER WITH REAL WORLD DATA THROUGH A TRIAL EMULATION FRAMEWORK
Author(s)
Carole R. Berini, PhD1, Jessica K. Paulus, ScD1, Malcolm Charles, M.S.1, Zhaohui Su, PhD1, Paul R. Conkling, MD1, Amin Haiderali, MBA, MPH2, Jagadeswara Rao Earla, MBA, PharmD, PhD2, Shan Jiang, Ph.D.2, Kaushal Desai, PhD2;
1Ontada, Boston, MA, USA, 2Merck & Co., Inc., Rahway, NJ, USA
1Ontada, Boston, MA, USA, 2Merck & Co., Inc., Rahway, NJ, USA
OBJECTIVES: Reliable real-world (RW) outcome measures are needed to support clinical and regulatory decision-making. This study aimed to confirm the concordance between RW endpoints and clinical trial estimates in early-stage triple-negative breast cancer (eTNBC) and melanoma with the application of a trial emulation framework.
METHODS: This retrospective observational study utilized electronic health record data from patients in US community oncology practices. RW cohorts were defined in alignment with the control arm of two KEYNOTE (KN) trials (KN-522 in eTNBC, KN-716 in melanoma). RW constructs for clinical endpoints [early Melanoma: recurrence-free (rwRFS), distant-metastasis-free (rwDMFS), eTNBC: event-free (rwEFS), pathological complete response (pCR)] were implemented. Matching Adjusted Indirect Comparison (MAIC) adjusted differences in 5 available baseline characteristics. Kaplan-Meier curves estimated RW endpoints, with an intention-to-treat (ITT) and an as-treated (AT) approach, censoring patients starting immunotherapy. Cox proportional hazard ratios (HR) and generalized linear model risk ratios (RR) compared RW with corresponding KN control arm estimates.
RESULTS: We identified 75 melanoma and 311 eTNBC RW patients. The highest concordance for RFS was observed in the adjusted AT, HR 0.98 (95% CI: 0.59, 1.64; p=0.938). For other endpoints, the highest concordance was observed in the adjusted ITT: DFMS HR 0.83 (95% CI: 0.51, 1.37; p=0.462); EFS HR 1.00 (95% CI: 0.72, 1.38; p=0.980), pCR RR 0.96 (95% CI 0.80, 1.14; p=0.627). Overall, positive confounding was mitigated through MAIC, with change in HRs ranging from 9% to 22%. The AT analysis did not significantly impact HRs, indicating that censoring was non-informative.
CONCLUSIONS: A trial emulation framework provided a robust methodological approach to minimize sources of bias associated with applying clinical trial design to RW populations. Emulation sensitivity to methodology highlights specific disease and measurement characteristics for each endpoint. Current findings confirm the reliability of rwRFS, rwDMFS in melanoma, and rwEFS, rwpCR in TNBC when leveraging careful design and fit-for-purpose data.
METHODS: This retrospective observational study utilized electronic health record data from patients in US community oncology practices. RW cohorts were defined in alignment with the control arm of two KEYNOTE (KN) trials (KN-522 in eTNBC, KN-716 in melanoma). RW constructs for clinical endpoints [early Melanoma: recurrence-free (rwRFS), distant-metastasis-free (rwDMFS), eTNBC: event-free (rwEFS), pathological complete response (pCR)] were implemented. Matching Adjusted Indirect Comparison (MAIC) adjusted differences in 5 available baseline characteristics. Kaplan-Meier curves estimated RW endpoints, with an intention-to-treat (ITT) and an as-treated (AT) approach, censoring patients starting immunotherapy. Cox proportional hazard ratios (HR) and generalized linear model risk ratios (RR) compared RW with corresponding KN control arm estimates.
RESULTS: We identified 75 melanoma and 311 eTNBC RW patients. The highest concordance for RFS was observed in the adjusted AT, HR 0.98 (95% CI: 0.59, 1.64; p=0.938). For other endpoints, the highest concordance was observed in the adjusted ITT: DFMS HR 0.83 (95% CI: 0.51, 1.37; p=0.462); EFS HR 1.00 (95% CI: 0.72, 1.38; p=0.980), pCR RR 0.96 (95% CI 0.80, 1.14; p=0.627). Overall, positive confounding was mitigated through MAIC, with change in HRs ranging from 9% to 22%. The AT analysis did not significantly impact HRs, indicating that censoring was non-informative.
CONCLUSIONS: A trial emulation framework provided a robust methodological approach to minimize sources of bias associated with applying clinical trial design to RW populations. Emulation sensitivity to methodology highlights specific disease and measurement characteristics for each endpoint. Current findings confirm the reliability of rwRFS, rwDMFS in melanoma, and rwEFS, rwpCR in TNBC when leveraging careful design and fit-for-purpose data.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
RWD88
Topic
Real World Data & Information Systems
Topic Subcategory
Reproducibility & Replicability
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Oncology