Use of Multilevel Network Meta Regression (ML-NMR) Model in RET Mutation-Positive Medullary Thyroid Cancer (MTC)
Author(s)
Manoj Khanal, PhD, Min-Hua Jen, PhD, Michael Sonksen, PhD, Tarun Puri, MD, Jacek Kiesel, PhD, Urpo Kiiskinen, PhD.
Eli Lilly and Company, Indianapolis, IN, USA.
Eli Lilly and Company, Indianapolis, IN, USA.
OBJECTIVES: The most common indirect comparison includes network meta-analysis (NMA) and matching adjusted indirect comparison (MAIC) both of which have limitations. The standard NMA uses only aggregate data (AgD) and assumes treatment effect modifiers across trials are the same. MAIC is currently the widely used method for population adjustment, however it is limited to the pairwise indirect comparison scenario with one study having individual patient level data (IPD) and one study having AgD. As a result, MAIC cannot be extended to incorporate a larger network of studies. This study evaluates the feasibility of ML-NMR which extends beyond MAIC and NMA in a way that it uses IPD from the trials that have it, and covariate information from all the trials.
METHODS: To illustrate the application of ML-NMR, selpercatinib data from LIBRETTO-531(selpercatinib vs investigator’s choice of cabozantinib or vandetanib) and published data from EXAM (cabozantinb vs placebo), and ZETA (vandetanib vs placebo) trials for MTC were used. Regression models with different likelihood functions such as exponential, gompertz, and weibull with fixed effects for progression free survival (PFS) were considered, and deviance information criterion (DIC) was used to select the best fit. The covariates in the model included age, M918T mutation positive (RET; non-RET), sex (male; female), and ECOG (0; >=1).
RESULTS: The non proportional hazards and more flexible model such as cubic splines or piecewise exponential were also fitted, however there were convergence issues. The best fit was obtained using weibull model among the three that converged. The PFS HR (95% CI) for selpercatinib vs cabozantinib, vandetanib, and placebo were 0.13 (0.08, 0.22), 0.15 (0.09, 0.25), and 0.05 (0.03, 0.10) respectively.
CONCLUSIONS: This case study demonstrates the feasibility of using ML-NMR for comparing time to event outcomes across studies. The findings suggest that selpercatinib is more efficacious than cabozantinib, vandetanib, and placebo for MTC.
METHODS: To illustrate the application of ML-NMR, selpercatinib data from LIBRETTO-531(selpercatinib vs investigator’s choice of cabozantinib or vandetanib) and published data from EXAM (cabozantinb vs placebo), and ZETA (vandetanib vs placebo) trials for MTC were used. Regression models with different likelihood functions such as exponential, gompertz, and weibull with fixed effects for progression free survival (PFS) were considered, and deviance information criterion (DIC) was used to select the best fit. The covariates in the model included age, M918T mutation positive (RET; non-RET), sex (male; female), and ECOG (0; >=1).
RESULTS: The non proportional hazards and more flexible model such as cubic splines or piecewise exponential were also fitted, however there were convergence issues. The best fit was obtained using weibull model among the three that converged. The PFS HR (95% CI) for selpercatinib vs cabozantinib, vandetanib, and placebo were 0.13 (0.08, 0.22), 0.15 (0.09, 0.25), and 0.05 (0.03, 0.10) respectively.
CONCLUSIONS: This case study demonstrates the feasibility of using ML-NMR for comparing time to event outcomes across studies. The findings suggest that selpercatinib is more efficacious than cabozantinib, vandetanib, and placebo for MTC.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MSR213
Topic
Clinical Outcomes, Methodological & Statistical Research, Study Approaches
Disease
Oncology