The Synergy of Real-World Evidence, Transportability, and Decision-Analytic Modelling in the Evaluation of Disease Screening Programs

Decision Science is an interdisciplinary field that focuses on generating and integrating evidence to understand, inform, and optimize decision-making. In this dissertation, I used decision-analytic modelling to evaluate the cost-effectiveness of an emerging class of tuberculosis (TB) infection diagnostics; I leveraged real-world data and observational study designs to improve our understanding of long-term mortality following TB diagnosis; and, I developed a novel framework that integrates causal inference with decision analysis to quantify bias in value of information analyses arising from transportability issues. Collectively, this dissertation demonstrates the synergy of real-world evidence, transportability methods, and decision-analytic modelling in guiding resource allocation of disease screening programs and advancing healthcare research. In Chapter 1, I evaluated the cost-effectiveness of using host-response-based transcriptional signatures (HrTS) to screen for incipient tuberculosis (TB) among migrants arriving in the United States. HrTS have emerged as promising tools to identify individuals with elevated risks of developing TB disease. In this study, I created an individual-based discrete event simulation model to compare the projected health and economic impact of four post-arrival TB screening strategies. Strategies included no screening, screening using conventional interferon gamma release assays (IGRA-only), IGRA followed by HrTS (IGRA-HrTS), and HrTS-only, based on the WHO Optimal Target Product Profile for HrTS. Cost- effectiveness varied by TB incidence in migrants’ countries of origin, with no screening favored for very low-incidence settings. Overall, HrTS may be cost-effective in specific migrant subgroups, but results are sensitive to several assumptions, including progression risk trends post-entry. In Chapter 2, I leveraged claims data, national TB and mortality registries, and electronic health record to quantify long-term mortality risks associated with pulmonary TB. Individuals with a history of TB have been shown to face elevated long-term mortality, but the impact of TB disease, as opposed to underlying risk factors, on mortality risks remains unclear. In this study, I estimated the long-term mortality risks among individuals with pulmonary TB compared to matched individuals without TB using a retrospective cohort design in Taiwan. Using coarsened exact matching with a risk set sampling approach, we constructed a cohort of 2,038 TB cases and 6,114 matched controls, followed for a median of 7.2 years. Using Cox regression models with time- varying coefficients, the 10-year survival probability among TB exposed individuals was estimated to be 17 percentage point lower than their unexposed counterparts, with one-third of this survival difference attributable to post-TB effects. The findings highlight the importance of early TB detection and prevention. They also indicate that not accounting for long-term TB mortality risks may underestimate the value of TB intervention programs in policy modeling studies, affecting resource allocation decisions. In Chapter 3, I bridged causal inference and decision science methodologies to enhance the quality of evidence translation from clinical trial findings into real-world healthcare decisions. Decisions on whether to adopt new healthcare technologies often rely on evidence from randomized controlled trials (RCTs). However, trial and target populations often differ in important ways that can limit the generalizability and transportability of RCT results, raising questions about the need for additional evidence. Value of information (VOI) analysis can help guide research investment decisions by quantifying the value by reducing uncertainty before making policy decisions. However, conventional applications of VOI methods in health care research do not formally address biases that could arise from transportability issues. To address this gap, we proposed a novel framework that incorporates transportability methods from causal inference literature into VOI estimation to more accurately calculate the value of new research (expected value of sample information (EVSI)). We formally defined metrics to quantify components of systematic errors in EVSI calculations when transportability issues are neglected. Finally, we demonstrated this proposed approach through a simulation study and a case study using data from the National Lung Screening Trial. This novel approach has the potential to enhance the quality of evidence translation and the efficiency of clinical trial designs.