Background: (Network) meta-analysis of published survival data are often based on the reported hazard ratio, which relies on the proportional hazards assumption. This assumption is implausible when hazard functions intersect and can result in invalid indirect treatment comparisons and network meta-analysis of competing interventions.

Methods: As an alternative to (network) meta-analysis of the single constant hazard ratio, meta-analysis models with a multi-dimensional treatment effect are proposed. Based on published Kaplan-Meier curves the hazard ratio of compared interventions in a randomized controlled trial (RCT) are described as a function of time, and the parameters of these functions are synthesized across studies and comparisons.

Results: The proposed models are illustrated with two example network meta-analyses in oncology.

Conclusion: (Network) meta-analysis of published survival data with models where the hazard ratio is modeled as a function of time can be more closely fitted to the available data than meta-analysis based on the constant hazard ratio.