Guiding the choice of analytic approach for economic evaluations of oncology treatments

Different modelling approaches are used to address the same decision problem but can lead to different estimates of life years gained and quality-adjusted life years. Three common methods are used in health economics: the partitioned (PSM), the state-transition (STM) and more recently the multi-state model (MSM). Novel methods were also identified to jointly model progression and survival using a copula to jointly model survival outcomes and MSMs with transitions estimated simultaneously. Differences in model predictions may have the propensity to change the conclusions of an economic analysis and the decisions made on the basis of such analyses. A simulation study was conducted to identify whether one approach is consistently superior to others under particular circumstances, or in general. The simulation study suggests that no single method is satisfactory in all circumstances and that approaches cannot be selected based on observed data characteristics alone. Case studies using real trial data also indicated that different assumptions could be made when modelling treatment effects, that PSMs and STMs may be inaccurate to varying degrees when estimating incremental outcomes and that neither is bias-free. This thesis demonstrated that it is not possible to determine with certainty a priori which approach to select, based only on the observed characteristics of the available data; thus, analysts and decision-makers need be careful when relying on predictions from a single approach. Recommendations are formulated to improve the transparency of health economic analyses and increase decision-makers’ confidence in the use of those models. Because it is unknown whether ICERs generated using a single analytic approach are adequate, in some cases, decision-making should consider ICERs from a range of alternative approaches to account for structural uncertainty. This thesis also highlights the importance of clinical input in selecting the most appropriate approach for the extrapolation of survival data.