Presentation (CTI)

OBJECTIVES: Multilevel Network Meta-Regression (ML-NMR) is an increasingly utilized population-adjusted indirect treatment comparison method. However, incorporating ML-NMR into Excel-based cost-effectiveness models (CEMs) using full marginal posterior curves via G-computation across every MCMC draw is computationally infeasible. We present a streamlined approach using conditional (adjusted) M-spline survival curves evaluated at individual mean covariate values.
METHODS: Extending the simulated newly diagnosed multiple myeloma (NDMM) survival dataset from the R package multinma, we constructed a partitioned survival model comparing lenalidomide, thalidomide, and placebo. To replicate the ML-NMR framework in Excel, we utilized the multinma-estimated study-specific spline coefficients, integrated M-spline basis, covariate-specific linear predictors, and conditional log-hazard ratios at mean target covariate values. These conditional “average individual” curves were validated against population-marginalized survival curves. ML-NMR was applied to progression-free (PFS) survival. Costs, utilities, and overall survival were sourced from the literature, applying a hypothetical discount to the lenalidomide list price. Cost-effectiveness was expressed using incremental cost-effectiveness ratios (ICERs), estimated over a lifetime horizon using a UK NHS/PSS perspective.
RESULTS: In the validation analysis, the conditional curves aligned fully with the population-marginalized curves, with the largest absolute difference at any time point being 1.6 percentage points. Operationally, evaluating a single mean conditional vector instead of full G-computation loops enabled a standard 1,000-run PSA to execute within seconds in Excel. Mean costs and QALYs were £72,799 (95% CrI: £64,761, £81,780) and 4.57 (3.62, 5.51) for lenalidomide; £36,014 (£30,189, £41,858) and 3.68 (2.92, 4.47) for thalidomide; and £10,392 (£9,589, £11,259) and 3.45 (2.8, 4.09) for placebo. The estimated ICER for lenalidomide was £41,093 and £55,702 versus thalidomide and placebo, respectively. Due to simulated/hypothetical inputs, results are illustrative only.
CONCLUSIONS: Utilizing conditional M-spline survival curves at mean covariate values offers a valid and computationally efficient framework to integrate advanced Bayesian ML-NMR outputs into Excel-based CEMs.