Real-world cost-effectiveness studies: How to deal with confounding?

Cost-effectiveness research is an essential tool in healthcare decision-making, enabling the comparison of interventions based on their incremental costs and effects. While traditionally conducted alongside Randomized Controlled Trials (RCTs), there is a growing interest in complementary Real-World Data (RWD) analysis for Cost-Effectiveness Analysis (CEA). In order for healthcare decision-makers to make valid reimbursement decisions, it is essential that the results of CEAs are both valid and applicable to a broader context. While RCTs offer high internal validity, their generalizability can be limited. Therefore, RWD is increasingly being recognized as a valuable source of information that can complement RCT data. RWD offers advantages over RCTs, such as better generalizability because of less strict patient selection and reflection of actual treatment patterns, but come with complexities, like confounding and missing observations. Confounding, stemming from the non-random allocation of treatment, presents a significant challenge in RWD analysis and requires careful adjustment to avoid biased estimates. In this thesis, the concepts of time-fixed and time-dependent confounding are distinguished, with the latter being more common in longitudinal studies. Various methods for dealing with confounding have been proposed, but their performance in different scenarios remains unclear. Within the context of CEA, this thesis aims to clarify terminology, review methods for adjusting confounding, and compare the performance of different techniques in CEA studies using RWD. Firstly, this thesis aimed to bridge the gap between epidemiologists and economists by providing a glossary that equips participants in interdisciplinary collaborations with the essential shared vocabulary (Chapter 2). Next, the thesis provides a scoping review of methods available for addressing time-invariant confounding (Chapter 3). Following the identification of a wide range of methods, the aim was to evaluate the most promising ones and compare them with current research practices in the context of CEA (Chapter 4). This exploration expanded the research question by building upon the time-invariant confounding study. It introduced time-dependent confounding to the simulation setup by adding an extra time point (Chapter 5). Finally, in the general discussion, insights and recommendations for improving the validity and reliability of CEA using RWD are provided based on the scoping review and simulation studies included in this thesis (Chapter 6).