OBJECTIVES: Meta-analyses (MA), as part of systematic literature reviews (SLRs), may be used to support regulatory approval of treatments assessed in single-arm studies. However, for reimbursement submissions, comparator information is crucial. Indirect treatment comparison approaches are common, but adjustment for covariates is not always feasible. In this setting, other approaches can be used to address between-study variability. Naïve pooling approaches often depend on the proportional hazards (PH) assumption, which can be problematic in comparisons of immunotherapies and chemotherapies, which have different efficacy response profiles. We will illustrate a workflow for pooling time-to-event (TTE) data from SLRs for use in health economic models which allows us to avoid proportionality assumptions. This incorporates MA for TTE data which feeds into parametric survival curve extrapolation. We will compare this new approach to a approach of evidence synthesis which is not robust to PH violations.

METHODS: Our proposed approach utilizes a MA method to pool survival curves extracted from SLRs of the comparator interventions. The conditional survival across all studies at a set of timepoints are modelled simultaneously, in a multivariate random-effects model treating study as a random effect. Model inference follows the DerSimonian and Laird approach. The workflow is illustrated in a simulation study framework in which six standard parametric distributions were fitted to samples from the posterior distribution of the MA model. Model selection included statistical fit and visual inspection per the National Institute for Health and Care Excellence (NICE) guidelines.

RESULTS: Results of the simulation study showed that the random-effects methodology allowed for the recapture of simulated parameters and distributions through parametric survival curve inference.

CONCLUSIONS: This workflow provides an evidence synthesis approach for TTE endpoints which can accommodate proportional hazards violations while considering between-study variation. The output of this approach can be used to directly parameterize health economic models.