ISPOR Europe 2018
Barcelona, Spain
November, 2018
MO2
Cancer
Research on Methods (RM)
Modeling Methods (MS)
MODELLING OVERALL SURVIVAL IN IMMUNOTHERAPY USING PARAMETRIC TECHNIQUES: AVELUMAB IN PREVIOUSLY TREATED METASTATIC MERKEL CELL CARCINOMA
Bullement A1, Amin A2, Stapelkamp C2, Willis A3, Lilley C3, Hatswell A4, Pescott C5, Bharmal M5
1BresMed Health Solutions, Nottingham, UK, 2Merck Serono, Ltd, Feltham, UK, 3BresMed Health Solutions, Sheffield, UK, 4Delta Hat Limited and University College of London, Nottingham, UK, 5Merck KGaA, Darmstadt, Germany
OBJECTIVES: Avelumab (an anti-PD-L1 immune-checkpoint inhibitor) was recently approved in the US, EU, Japan, and other jurisdictions for the treatment of patients with metastatic Merkel cell carcinoma (mMCC). Data demonstrating the efficacy of avelumab are available from the registrational phase 2 JAVELIN Merkel 200 trial (NCT02155647) in previously treated mMCC from three data-cuts considering minimum patient follow-up periods of 12, 18, and 24 months. This analysis compares observed and extrapolated survival estimates from multiple data-cuts using standard parametric and spline-based approaches.

METHODS: Standard parametric and spline-based models were fitted to overall survival (OS) data. Goodness of fit was determined following published guidance using visual inspection, statistical fit (via Akaike’s information criterion [AIC]), and plausibility of long-term survival estimates, with advice from clinical and statistical experts. Within each data-cut, the best-fitting standard parametric and spline-based extrapolations were compared to establish which provided the most accurate survival estimation.

RESULTS: The 24-month OS rate from the best-fitting spline-based model (1-knot-odds, 34.1%, AIC, 379.26) fitted to the earliest data-cut provided a closer fit to the observed 24-month OS (35.8%) than the best-fitting standard parametric model (log-normal, 31.8%, AIC, 377.70). 1-knot-odds spline fitted to the latest data-cut provided a lower AIC (453.81) than log-normal (455.31). 5-year OS extrapolations across all data-cut ranged between 16.9%-22.5% (1-knot-odds spline) and 12.4%-16.2% (log-normal).

CONCLUSIONS: Spline-based models provided a more accurate estimation of the observed 24-month OS than standard parametric approaches. Long-term survival estimates from the spline-based models are more aligned with clinical expectations of immunotherapy: an emergent plateau in OS associated with the immune-response effect of treatment. Landmark or cure-based models may also reflect the expected immune-response tail in OS but require explicit assumptions about the estimation of long-term OS, such as the OS for cured patients and the prognostic importance of response. Longer-term data are required to validate OS extrapolations.