Investigating the Application of Causal Inference Methods for Modelling the Impact of Treatment Sequences in Health Economic Evaluations

This thesis explores the complexities of evaluating treatment sequence effectiveness in health technology assessments (HTA), focusing on challenges posed by limited evidence. Clinical trials typically assess individual lines of treatment (LOTs) rather than entire treatment sequences. Existing research recommends merging LOT-specific evidence from different sources to estimate overall treatment sequence effectiveness. However, this relies on a strong assumption about exchangeability between LOT-specific populations, which may not always hold.

Real-world data (RWD) provide a valuable alternative for assessing the comparative effectiveness of treatment sequences. However, generating reliable real-world evidence (RWE) presents significant challenges, notably confounding, which is exacerbated by the longitudinal nature of treatment sequences, leading to time-varying confounding. Conventional statistical methods (e.g. simple outcome regressions) may yield biased estimates, whereas advanced methods grounded in causal inference principles could offer more reliable estimates, provided their assumptions hold and data sources are of sufficient quality and breadth.

This thesis reviews advanced statistical methods and proposes a series of innovative, interconnected proof-of-concept studies to assess the feasibility of deriving unbiased RWE for comparing treatment sequences. It evaluates suitable RWD sources pertinent to English HTA that support these methods. Further, it leverages the Target Trial Emulation framework, a methodology endorsed by the National Institute for Health and Care Excellence (NICE) RWE framework, to mitigate biases in real-world study designs.

This thesis contributes to the field by delineating the challenges associated with treatment sequences and the landscape of English HTA practices. It complements existing treatment-sequencing modelling frameworks by proposing ways to leverage RWD to inform treatment sequence effectiveness and summarising challenges. Additionally, it extends the NICE RWE framework, particularly regarding practical applications and necessary adaptations for treatment sequence comparisons. A case study employing advanced inverse probability weighting methods demonstrates the feasibility of deriving unbiased treatment effectiveness estimates from Flatiron data, benchmarked against an existing trial.