Comparing Alternative Extrapolation Methods Using Standard Partitioned Survival Model Functionality in the Presence of Converging Survival Data: A Case Study in Renal Cell Carcinoma
Author(s)
Paulina Kazmierska, MSc1, Neda Aminnejad, PhD2, George Bungey, MSc3.
1PPD™ Evidera™ Health Economics & Market Access, Thermo Fisher Scientific, London, United Kingdom, 2PPD™ Evidera™ Health Economics & Market Access, Thermo Fisher Scientific, Toronto, ON, Canada, 3Research Scientist, PPD™ Evidera™ Health Economics & Market Access, Thermo Fisher Scientific, London, United Kingdom.
1PPD™ Evidera™ Health Economics & Market Access, Thermo Fisher Scientific, London, United Kingdom, 2PPD™ Evidera™ Health Economics & Market Access, Thermo Fisher Scientific, Toronto, ON, Canada, 3Research Scientist, PPD™ Evidera™ Health Economics & Market Access, Thermo Fisher Scientific, London, United Kingdom.
OBJECTIVES: In specific circumstances, combinations of parametric survival extrapolations may produce implausible crossings between two comparators due to the characteristics of the underlying Kaplan-Meier (KM) data. The objective of this case study was to explore the impact of combining typical functionality (treatment effect waning, KM+parametric fit) included in partitioned survival models (PSMs) with standard parametric models as alternative extrapolation approaches compared to conservatively assuming equivalence at the crossing point, using multiple overall survival (OS) data cuts from an existing renal cell carcinoma (RCC) trial.
METHODS: Published OS data from the CLEAR trial comparing lenalidomide in combination with pembrolizumab (LEN+PEM) versus sunitinib (SUN) in RCC was digitised. A series of standard parametric distributions were then fitted in R to the August 2020 data cut, which showed convergence between study arms at 33 months. LEN+PEM and SUN OS was then extrapolated beyond the first data cut, considering the following methods: (1) assuming equivalence at the crossing point; (2) assuming equivalent efficacy to SUN for LEN+PEM at the start of convergence between the two treatments, after which LEN+PEM OS hazards are set equal to SUN; (3) exploring a combined KM+parametric extrapolation approach using a truncated KM curve. Long-term extrapolations were then assessed based on comparison with UK clinical expert expectations for SUN from NICE TA858 (<20% at 10 years), as well as visual comparison with the final data cut KM curve (July 2022).
RESULTS: Approaches 2 and 3 appeared to produce reasonable long-term extrapolations of the earlier data cut that both aligned with clinical expert feedback from NICE TA858 and better matched KM curves from the final data cut compared to approach 1.
CONCLUSIONS: The results of our case study suggest that methods using relatively straightforward functionality commonly implemented in PSMs may provide reasonable alternative extrapolations in the presence of converging OS KM data between study arms.
METHODS: Published OS data from the CLEAR trial comparing lenalidomide in combination with pembrolizumab (LEN+PEM) versus sunitinib (SUN) in RCC was digitised. A series of standard parametric distributions were then fitted in R to the August 2020 data cut, which showed convergence between study arms at 33 months. LEN+PEM and SUN OS was then extrapolated beyond the first data cut, considering the following methods: (1) assuming equivalence at the crossing point; (2) assuming equivalent efficacy to SUN for LEN+PEM at the start of convergence between the two treatments, after which LEN+PEM OS hazards are set equal to SUN; (3) exploring a combined KM+parametric extrapolation approach using a truncated KM curve. Long-term extrapolations were then assessed based on comparison with UK clinical expert expectations for SUN from NICE TA858 (<20% at 10 years), as well as visual comparison with the final data cut KM curve (July 2022).
RESULTS: Approaches 2 and 3 appeared to produce reasonable long-term extrapolations of the earlier data cut that both aligned with clinical expert feedback from NICE TA858 and better matched KM curves from the final data cut compared to approach 1.
CONCLUSIONS: The results of our case study suggest that methods using relatively straightforward functionality commonly implemented in PSMs may provide reasonable alternative extrapolations in the presence of converging OS KM data between study arms.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
EE141
Topic
Economic Evaluation, Health Technology Assessment, Methodological & Statistical Research
Disease
Oncology