ESTIMATING LONG-TERM SURVIVAL IN THE FACE OF IMMATURE DATA- A CASE STUDY OF NIVOLUMAB IN SECOND LINE RENAL CELL CARCINOMA

Author(s)

Klijn SL¹, Torkilseng EB², Tyas D³, Sowdani A³, Malcolm B³, Mudd A¹, Johannesen KM⁴
¹Pharmerit International, Rotterdam, The Netherlands, ²Bristol-Myers Squibb, Lysaker, Norway, ³Bristol-Myers Squibb, Uxbridge, UK, ⁴Bristol-Myers Squibb AB, Stockholm, Sweden

Presentation Documents

CN7--strong-u-klijn-sl-u-sup-1-sup-strong-torkilseng-eb-sup-2-sup-tyas-d-sup-3-sup-sowdani-a-sup-3-sup-malcolm-b-sup-3-sup-mudd-a-sup-1-sup-johannesen-km-sup-4-sup-br-sup-1-sup-pharmerit-international-rotterdam-the-netherlands-sup-2-sup-bri ...

OBJECTIVES: The clinical trial of nivolumab in second line renal cell carcinoma (CheckMate‑025) was stopped early after meeting prespecified overall survival (OS) gains in an interim analysis. This study aims to illustrate the added benefit of long-term registry data in the selection of the survival extrapolation curves in a case with 14 months’ minimum follow-up.

METHODS: We compared different OS extrapolations on statistical fit criteria and alignment with RCC registry data. NICE (UK), NoMA (Norway), and TLV (Sweden) respectively selected the single generalized gamma (single-GG), independent gamma (indep‑G), and single log-logistic (single‑LL) curves based on the 14‑months dataset. We used a 26‑months dataset of OS, unavailable at the time of the original extrapolation decisions, to assess the predictive value and stability of the extrapolations.

RESULTS: All HTA agency extrapolations underestimated OS compared to the Norwegian and Swedish registry data, though the single‑LL curve matched the registry data most closely. Predicted OS estimates based on the 14‑months dataset, using single‑GG, indep‑G, and single‑LL models, matched the observed survival recorded in the 26‑months dataset. Refitting the curves to the 26‑months data showed substantial differences in mean OS estimates compared to the 14-month data for the indep‑G curve (Δ=7.7%). The other two curves provided similar OS estimates, irrespective of the dataset used (Δ<1%).

CONCLUSIONS: The single-LL curve matched external long‑term registry data most precisely and it had a high predictive value as evaluated by a comparison to data from CheckMate‑025’s 26‑months dataset. Though registry data may be based on treatments with alternative mechanisms of action, they may provide a lower boundary for the OS estimate in the absence of long-term trial data. This study is an example of how utilization of data from external registries provided valuable information and more stable OS estimates across different datasets in case of limited trial survival data.

Conference/Value in Health Info

2017-11, ISPOR Europe 2017, Glasgow, Scotland

Value in Health, Vol. 20, No. 9 (October 2017)

Code

CN7

Topic

Methodological & Statistical Research

Topic Subcategory

Modeling and simulation

Disease

Oncology, Urinary/Kidney Disorders

Explore Related HEOR by Topic

Methodology

Presentation