OBJECTIVES: Syntheses of time-to-event data often rely on published hazard ratios (HR) therefore assuming proportional hazards (PH). Methods now exist to reconstruct individual patient data (IPD) from published Kaplan-Meier (KM) plots. This enables the PH assumption to be formally tested and syntheses to be conducted on the accelerated failure time (AFT) scale using a time ratio (TR) as the measure of relative treatment effect. While TRs do not require the PH assumption, these are rarely used in evidence synthesis. We compare relative treatment effect and mean survival estimates obtained from a network meta-analysis (NMA) conducted on the PH and AFT scales. METHODS: KM curves for overall survival, from a subset of an evidence network analysed in a previous NICE appraisal, were digitised. The PH and AFT assumptions were assessed for each study using log-cumulative hazard, Schoenfeld residual and Q-Q plots. HRs and TRs were estimated for each study using a PH and AFT Weibull model and synthesised in a Bayesian NMA. Mean survival was estimated for each treatment using the NMA results and assuming a Weibull distribution for baseline survival. RESULTS: The PH and AFT assumptions were potentially violated in 8/12 and 4/12 studies respectively based on the plots assessed. The choice of scale influenced mean survival; the difference in mean survival between scales ranged from 0.72 to 7.38 months across treatments, compared to mean survival of 20.15 and 23.42 across treatments for the PH and AFT scales respectively. CONCLUSIONS: Relative treatment effects expressed as TRs are not commonly presented in evidence synthesis due to unfamiliarity around interpretation and lack of reporting in published literature. However, the choice of scale can influence mean survival estimates. Syntheses of time-to-event data for cost-effectiveness analysis should therefore no longer rely on published HRs and should consider the AFT scale when the PH assumption is violated.