Methodologies for complex health economic modeling

Making effective decisions in health and medicine is crucial, as each decision affects the well-being of others. Making cost-effective decisions in health and medicine is even more critical as economic resources are not infinite. Health economic modeling refers to the process of evaluating the costs and effects of healthcare interventions. Decision analytic models are mathematical tools to account for and model the uncertainty around each decision, and to determine the best decision after collecting different sources of evidence. There are different open challenges and interesting problems related to statistical techniques to analyze complex and realistic evidence sources, as well as the uncertainty surrounding decision-making. In this thesis, we focus on the development of various Bayesian methodologies in complex and realistic frameworks within the context of health economic modeling. In the first chapter, we provide an extensive introduction to the context of health economic modeling and cost-effective analysis, and outline the open challenges we will address throughout the thesis. In Chapter 2, we define Inverse target trial emulation, a novel Bayesian methodology to generate realistic observational data. The central idea of this methodology is to start with an initial (preliminary) Randomized Clinical Trial (RCT) and use the initial information to generate different types of observational data in various contexts and under different assumptions. Target trial emulation (TTE) is a methodology that links an observational dataset to a targeted RCT, emulating experimental data by solving all the different forms of bias in observational data. In this context, we reverse this process, aiming to simulate (not only emulate) observational data from an initial trial. This methodology proves to be very useful for performing different forms of sensitivity analysis, research prioritization, and testing of the robustness of the methods typically employed to analyze observational data. In the third and fourth chapters, we focus our attention on VoI (Value of Information) analysis in complex and realistic scenarios. Given an underlying health-economic decision model, VoI analysis is a technique to quantify the expected benefit that may result from reducing uncertainty in the economic model. In particular, the Expected Value of Sample Information (EVSI) is a measure that estimates the expected benefit of performing additional data to reduce the uncertainty in the parameters of the underlying health economic model. Until now, the EVSI methodology has been applied only to fairly simple data collection exercises. In most cases, it has been used to understand the value of randomized clinical trial (RCT) data, meaning that it measured the value of collecting additional RCTs to reduce uncertainty. In these chapters, relying also on the results from Chapter 1, we design and apply a novel methodology to use EVSI when we plan to collect more complex and realistic data, i.e., data affected by missingness and confounding. In the final chapter, we develop a Bayesian version of the subpopulation treatment effect pattern plot (STEPP) methodology and apply it to real scenarios. STEPs is a methodology that enables researchers to properly analyze the heterogeneity of treatment effects (HTE) in experimental studies, providing the necessary information to customize treatment for individuals to maximize benefits. STEPP constructs overlapping subpopulations along the continuum of a continuous covariate of interest (e.g., a biomarker), thus improving the precision of the estimated treatment effects within the subgroups. In that chapter, we introduce a Bayesian version of the STEPP method (B-STEPP) and demonstrate that a Bayesian approach enables flexible modeling of the dependence among the relevant parameters, providing good control over the joint distribution of the parameters and their associated uncertainty.