Synthesize marginal survival summaries within ML-NMR when individual event times are unavailable

Develop a method to synthesize reported marginal survival summaries—specifically marginal median survival times and marginal restricted mean survival times—within the multilevel network meta‑regression (ML‑NMR) framework for studies where individual event/censoring times are unavailable, by constructing appropriate aggregate‑level likelihood contributions based on the mathematical relationships between these summaries and the population‑average marginal survival function.

Background

The paper extends ML‑NMR to general likelihoods, allowing integration of individual‑level survival models over aggregate covariate distributions. In many practical applications, only aggregate outcomes are available rather than reconstructed individual event/censoring times, which limits synthesis options.

When only conditional log hazard ratios are reported, the authors note these can be synthesized with a Normal likelihood, but this requires that the reported adjustments match those in the ML‑NMR model. They suggest that in theory one could instead synthesize marginal summaries (median survival and restricted mean survival) using established functional relationships, but acknowledge that this has not yet been developed and remains unresolved.