ELICITING CANCER PROGRESSION RATES FOR LOCALLY/REGIONAL RECURRENT (LR) POPULATIONS USING AGGREGATE LEVEL SURVIVAL DATA: AN ILLUSTRATION FROM MELANOMA
Author(s)
Shubhram Pandey, MSc1, Sameer Mansoori, MSc1, Barinder Singh, RPh1, Murat Kurt, BS, MS, PhD2;
1Pharmacoevidence Pvt. Ltd., SAS Nagar, Mohali, India, 2Iovance Biotherapaeutics, Inc., Philadelphia, PA, USA
1Pharmacoevidence Pvt. Ltd., SAS Nagar, Mohali, India, 2Iovance Biotherapaeutics, Inc., Philadelphia, PA, USA
OBJECTIVES: State-transition models accounting for prognostic differences between LR and distant recurrent (DR) beyond trial follow-up can improve long-term survival projections in early-stage cancer trials. However, early data cuts limit estimation of transitions from LR-state without strong assumptions. We propose an approach to infer these transitions leveraging aggregate-level survival data for LR and DR populations.
METHODS: LR patients were assumed to remain LR, progress to DR, or die at constant monthly rates. Survival rates for LR and DR populations were extrapolated using standard parametric models, with the top three distributions based on AIC blended using relative likelihood weights. Mean LR sojourn time, estimated as area differential between the modeled survival curves for LR and DR populations, informed the probability of remaining in LR state [P(LR)]. Progression from LR-to-DR[P(DR)] was elicited by approximating modeled LR-survival as the convolution of modeled DR-survival and LR-to DR transitions based on monthly survival estimates from LR. Sensitivity analyses examined BIC-based model weights and alternative time horizons. The approach was illustrated in melanoma using publicly available relative survival data from Surveillance, Epidemiology, and End Results (SEER) Program.
RESULTS: Base-case analysis assumed a lifetime horizon of 39 years, corresponding to mean baseline age of 61 years in LR population. Estimated mean LR sojourn time was 143.6 months, corresponding to P(LR) = 0.9934 and P(DR) = 0.0066. Blending survival distributions for LR and DR populations using BIC-based model weights and evaluating them over medium (20-year) and short (5-year) time horizons, the estimates ranged between 0.9177-0.9927 for P(LR) and 0.0072-0.082 for P(DR).
CONCLUSIONS: The proposed approach is scalable and applicable across cancer types to support calibration of cost-effectiveness models distinguishing LR and DR patients. The findings should be interpreted cautiously, as they reflect the U.S. treatment landscape and rely on aggregate-level data from LR and DR populations that may not fully overlap.
METHODS: LR patients were assumed to remain LR, progress to DR, or die at constant monthly rates. Survival rates for LR and DR populations were extrapolated using standard parametric models, with the top three distributions based on AIC blended using relative likelihood weights. Mean LR sojourn time, estimated as area differential between the modeled survival curves for LR and DR populations, informed the probability of remaining in LR state [P(LR)]. Progression from LR-to-DR[P(DR)] was elicited by approximating modeled LR-survival as the convolution of modeled DR-survival and LR-to DR transitions based on monthly survival estimates from LR. Sensitivity analyses examined BIC-based model weights and alternative time horizons. The approach was illustrated in melanoma using publicly available relative survival data from Surveillance, Epidemiology, and End Results (SEER) Program.
RESULTS: Base-case analysis assumed a lifetime horizon of 39 years, corresponding to mean baseline age of 61 years in LR population. Estimated mean LR sojourn time was 143.6 months, corresponding to P(LR) = 0.9934 and P(DR) = 0.0066. Blending survival distributions for LR and DR populations using BIC-based model weights and evaluating them over medium (20-year) and short (5-year) time horizons, the estimates ranged between 0.9177-0.9927 for P(LR) and 0.0072-0.082 for P(DR).
CONCLUSIONS: The proposed approach is scalable and applicable across cancer types to support calibration of cost-effectiveness models distinguishing LR and DR patients. The findings should be interpreted cautiously, as they reflect the U.S. treatment landscape and rely on aggregate-level data from LR and DR populations that may not fully overlap.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR19
Topic
Methodological & Statistical Research
Disease
SDC: Oncology