Background: Upon approval of new treatments in subsequent lines of therapy, quantifying efficacy of a new therapy in the settings for earlier lines can be challenging. As demonstrated in the clinical trials of immuno-oncology (IO) treatments, switching to IO treatments is allowed in the standard of care arm and more switching has been observed over time across different trials. This poses challenges in assessing comparative effectiveness and in health economic evaluations.

Objective: The objective is to assess statistical and modelling approaches that could potentially be adopted in these situations.

Method/Approach: Statistical approaches for adjustment for treatment switching to allow comparisons across different comparators in health economics evaluation were reviewed and their pros and cons including HTA’s critiques will be presented. The approaches reviewed included: rank preserving structural failure time model, assuming same treatment effect before and after progression; inverse-probability-of-censoring weighting, assuming no unmeasured confounders; 2-stage methods, making fewer assumptions but subject to availability of data at a secondary baseline; excluding centres allowing subsequent treatments, which maintains randomization but results in reduced sample size; censoring at subsequent treatment, which could be prone to selection bias/informative censoring; and unanchored matching-adjusted indirect comparison, which removes the issue of different extent of subsequent therapies between common comparator arms however is subject to potential bias due to residual confounding.

Implications/Recommendations

All statistical approaches have limitations and different methods may result in different results. The choice of modelling approach should depend on data availability and be informed by methods used in comparator economic analyses as well as HTA preferences. Economic modelling should reflect what happens in clinical practice and thus comparison of switching patterns observed in clinical trials and clinical practice is required to judge whether adjustment is needed in health economic analysis. Economic models should be built with flexibility to include all possible approaches.