A comparison of published time invariant Markov models with Partitioned Survival models for cost effectiveness estimation; three case studies of treatments for glioblastoma multiforme.

backgroundCost-effectiveness analyses of treatments for glioblastoma multiforme (GBM) have mostly used state transition Markov models with time invariant transition probabilities (TIMMs). In three case studies of GBM treatments, we compared Partitioned Survival model (PSM) results with published outputs from TIMMs.

methodsPSMs used the same RCT data sources, utility values, time horizons, cycle times and annual discounting used in published TIMMs. Reported overall survival and progression-free survival plots were digitised and fitted with a range of parametric models. Economic model outputs were generated in the same form as reported for the TIMMs. PSM output uncertainty was explored in univariate and in multivariate sensitivity analyses.

resultsPSMs generated incremental cost-effectiveness ratios that were different to the published TIMMs. The magnitude of difference was substantial in two cases. The PSMs were reasonably robust and in sensitivity analyses were sensitive to variations in the same model inputs as were the TIMMs. When compared to the RCT data, the TIMMs tended to generate underestimates of the likely overall survival gain. TIMM estimates for depletion of individuals from the stable disease state and for accumulation in the dead state had relatively poor resemblance to the source RCT data.

conclusionTIMMs delivered different cost-effectiveness estimates to PSMs; in two cases, TIMMs produced substantially lower ICER values than PSMs. Model output differences appear attributable to less realistic cost-and-benefit estimates generated in TIMMs due to rapid depletion from the stable disease state and/or accumulation in the dead state.