What About a Latent Cure Model? Assessing Cure Models’ Performance in Pediatric Acute Lymphoblastic Leukemia Treated With Tisagenlecleucel
Author(s)
Guillem Hopmans Galofré, MSc1, Marti Hopmans, MSc2, Isaac Corro Ramos, PhD3.
1Erasmus University Rotterdam, Rotterdam, Netherlands, 2Independent Researcher, Barcelona, Spain, 3Institute for Medical Technology Assessment (iMTA), Erasmus University Rotterdam, Rotterdam, Netherlands.
1Erasmus University Rotterdam, Rotterdam, Netherlands, 2Independent Researcher, Barcelona, Spain, 3Institute for Medical Technology Assessment (iMTA), Erasmus University Rotterdam, Rotterdam, Netherlands.
OBJECTIVES: Latent cure models (LCM) have been brought up as a resourceful yet underused method to extrapolate survival of curative treatments. We validated extrapolations from an early data cut-off (DCO) of tisagenlecleucel in haemato-oncology via comparison to a later DCO.
METHODS: Mixture cure models (MCM), generalised mixture cure models (GenMCM), LCMs, and spline-based models (SM) were fitted to overall survival (OS) and event-free survival (EFS) data from the ELIANA trial, using a 38.8-month median follow-up (1st DCO). Results were compared to 79.4-month median follow-up (2nd DCO) and clinical expert opinion.
RESULTS: For OS, only SMs with ≥3 degrees of freedom (DF) predicted the accelerated decline observed in the 2nd DCO. For cure models, differences from the 7-year Kaplan-Meier curve ranged from 4.8% to 10.6%. Those with smaller differences were LCMs assuming cure at 7 or 10 years (5.1% to 6.9%) and MCMs with exponential, Weibull-exponential, and GenMCM-4DF (4.8% to 5.9%). Conversely, LCMs’ difference from 5-year EFS estimates ranged 0.5% to 2.26%, outperforming GenMCMs (-3.1% to 1.47%), MCMs (-17.3% to 2%), and SMs (-6.37% to -1.5%).When assessing long-term extrapolations, cure models overestimated clinicians’ most likely 20-year OS rates (16% to 21.9%), whereas SMs with 3-6 DF were closely aligned (-0.57% to -2.58%). Compared to highest plausible estimates, all cure model predictions improved similarly (3% to 8.9%). Most cure models overestimated clinicians’ preferred 20-year EFS rates. Long-term OS estimates for GenMCMs and MCMs were notably lower with 2nd DCO-fitted models, while PFS estimates remained consistent.
CONCLUSIONS: LCMs performed similarly to other cure models when compared to the 2nd DCO and overestimated long-term extrapolations when considering clinicians’ most likely estimates. Long-term extrapolations were mostly unaffected by the choice of DCO. Despite novel methods, uncertainty persists with immature data. Robust estimates require clinical validation, incorporating external data, and longer trial follow-ups.
METHODS: Mixture cure models (MCM), generalised mixture cure models (GenMCM), LCMs, and spline-based models (SM) were fitted to overall survival (OS) and event-free survival (EFS) data from the ELIANA trial, using a 38.8-month median follow-up (1st DCO). Results were compared to 79.4-month median follow-up (2nd DCO) and clinical expert opinion.
RESULTS: For OS, only SMs with ≥3 degrees of freedom (DF) predicted the accelerated decline observed in the 2nd DCO. For cure models, differences from the 7-year Kaplan-Meier curve ranged from 4.8% to 10.6%. Those with smaller differences were LCMs assuming cure at 7 or 10 years (5.1% to 6.9%) and MCMs with exponential, Weibull-exponential, and GenMCM-4DF (4.8% to 5.9%). Conversely, LCMs’ difference from 5-year EFS estimates ranged 0.5% to 2.26%, outperforming GenMCMs (-3.1% to 1.47%), MCMs (-17.3% to 2%), and SMs (-6.37% to -1.5%).When assessing long-term extrapolations, cure models overestimated clinicians’ most likely 20-year OS rates (16% to 21.9%), whereas SMs with 3-6 DF were closely aligned (-0.57% to -2.58%). Compared to highest plausible estimates, all cure model predictions improved similarly (3% to 8.9%). Most cure models overestimated clinicians’ preferred 20-year EFS rates. Long-term OS estimates for GenMCMs and MCMs were notably lower with 2nd DCO-fitted models, while PFS estimates remained consistent.
CONCLUSIONS: LCMs performed similarly to other cure models when compared to the 2nd DCO and overestimated long-term extrapolations when considering clinicians’ most likely estimates. Long-term extrapolations were mostly unaffected by the choice of DCO. Despite novel methods, uncertainty persists with immature data. Robust estimates require clinical validation, incorporating external data, and longer trial follow-ups.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MSR220
Topic
Methodological & Statistical Research
Disease
Oncology