The analysis of overall survival in a changing clinical world: the methodological implications of relapsing cancers in randomised controlled trials

Overall survival (OS) is defined as the time from randomisation to death from any cause. It is a definitive endpoint in phase III clinical trials to determine whether an experimental intervention improves participant outcomes compared to a control intervention. However, during clinical trial follow-up, participants, in any arm, may discontinue their trial intervention and begin subsequent anti-cancer therapy, which may improve outcomes.

This PhD thesis aimed to understand whether researchers were considering subsequent anti-cancer therapy in their analysis of OS, what stakeholders (statisticians and other data analysts, payers and industry partners, healthcare professionals, patient, carers and the public) thought about subsequent anti-cancer therapy and OS, and to develop new, or extend existing, methodology which assessed OS whilst accounting for subsequent anti-cancer therapy in randomised clinical trials.

Study One was a systematic literature review investigating how OS is reported. The review found that subsequent anti-cancer therapy was generally acknowledged as a potential limitation, but not addressed through sensitivity analysis. Study Two was a stakeholder consultation incorporating an online questionnaire and focus group. The questionnaire had 103 respondents, and seven participants attended the focus group. From the stakeholder consultation it was concluded that stakeholders were interested in the problem of subsequent anti-cancer therapy. This translated into an alternative hypothetical estimand which was investigated in Study Three. This was a simulation study assessing whether existing statistical analysis methods could be used to estimate the estimand: The treatment effect if everyone who did discontinue before death, went on to receive the same subsequent anti-cancer therapy. The existing methods were also applied to a real data case study. From the simulation and case study, the existing two-stage estimation method was concluded to be the most appropriate. These findings will now be disseminated to the consenting participants from study two and further research will be agreed collaboratively.