Incorporating Real-World Evidence Into Time-to-Event Extrapolations From Clinical Trial Data
Author(s)
Shubhram Pandey, MSc1, Supreet Kaur, MSc2, Parampal Bajaj, B. Tech3, Barinder Singh, RPh4, Akanksha Sharma, MSc5.
1Senior Consultant and Head, Modeling and Advanced Analytics, Pharmacoevidence, Mohali, India, 2Senior Statistician, HEOR, Heorlytics, Mohali, India, 3Senior Modeler, HEOR, Heorlytics, Mohali, India, 4Director, HEOR, Pharmacoevidence, London, United Kingdom, 5Lead Statistician, HEOR, Pharmacoevidence, Mohali, India.
1Senior Consultant and Head, Modeling and Advanced Analytics, Pharmacoevidence, Mohali, India, 2Senior Statistician, HEOR, Heorlytics, Mohali, India, 3Senior Modeler, HEOR, Heorlytics, Mohali, India, 4Director, HEOR, Pharmacoevidence, London, United Kingdom, 5Lead Statistician, HEOR, Pharmacoevidence, Mohali, India.
OBJECTIVES: Extrapolation of time-to-event data beyond clinical trials remains challenging in health economic evaluations. While numerous methodologies exist for long-term projections, most lack formal mechanisms to incorporate real-world evidence (RWE). This study presents a methodological framework for integrating RWE into survival extrapolations while maintaining mathematical consistency and clinical plausibility.
METHODS: The proposed framework fits multiple parametric distributions to trial data and integrates external RWE through: (1) assigning data-driven weights to parametric models based on fit to RWE points using distance-weighted mean squared error, and (2) applying inverse distance-weighted influence from RWE data points to create blended extrapolations. This approach preserves Kaplan-Meier estimates within the trial period while allowing controlled RWE influence on long-term projections. The methodology was applied to simulated 5-year trial data with external survival estimates up to 40 years.
RESULTS: Across all parametric distributions, RWE integration substantially modified extrapolations and life-year estimates. The weighted average of parametric models without RWE integration yielded 14.29 and 10.15 life years for intervention and control arms, respectively (incremental benefit: 4.14 years). With RWE integration, estimates increased to 15.85 and 11.24 years, respectively (incremental benefit: 4.61 years), representing an 11.4% difference in estimated treatment effect. The best-fitting distributions shifted from log-normal in the standard approach to generalized gamma with RWE integration. Survival probability estimates at distant time points showed more plausible alignment with external evidence while maintaining mathematical consistency.
CONCLUSIONS: Integration of RWE into parametric survival extrapolations significantly impacts life-year estimates and treatment benefits. This model-averaging approach with distance-weighted RWE influence offers value to health technology assessment bodies seeking more realistic long-term projections, especially for novel therapies with limited trial data but available real-world evidence. By reducing uncertainty in extrapolated survival curves, this methodology can improve decision-making reliability for reimbursement and resource allocation.
METHODS: The proposed framework fits multiple parametric distributions to trial data and integrates external RWE through: (1) assigning data-driven weights to parametric models based on fit to RWE points using distance-weighted mean squared error, and (2) applying inverse distance-weighted influence from RWE data points to create blended extrapolations. This approach preserves Kaplan-Meier estimates within the trial period while allowing controlled RWE influence on long-term projections. The methodology was applied to simulated 5-year trial data with external survival estimates up to 40 years.
RESULTS: Across all parametric distributions, RWE integration substantially modified extrapolations and life-year estimates. The weighted average of parametric models without RWE integration yielded 14.29 and 10.15 life years for intervention and control arms, respectively (incremental benefit: 4.14 years). With RWE integration, estimates increased to 15.85 and 11.24 years, respectively (incremental benefit: 4.61 years), representing an 11.4% difference in estimated treatment effect. The best-fitting distributions shifted from log-normal in the standard approach to generalized gamma with RWE integration. Survival probability estimates at distant time points showed more plausible alignment with external evidence while maintaining mathematical consistency.
CONCLUSIONS: Integration of RWE into parametric survival extrapolations significantly impacts life-year estimates and treatment benefits. This model-averaging approach with distance-weighted RWE influence offers value to health technology assessment bodies seeking more realistic long-term projections, especially for novel therapies with limited trial data but available real-world evidence. By reducing uncertainty in extrapolated survival curves, this methodology can improve decision-making reliability for reimbursement and resource allocation.
Conference/Value in Health Info
2025-09, ISPOR Real-World Evidence Summit 2025, Tokyo, Japan
Value in Health Regional, Volume 49S (September 2025)
Code
RWD157
Topic Subcategory
Reproducibility & Replicability
Disease
SDC: Oncology