Accounting for Heterogeneity in Health Decision Analysis

Abstract

This dissertation explores ways to account for heterogeneity in health decision analysis, including: the integration of individual-level risk factors into a chronic disease simulation model, evaluation of personalized blood pressure treatment decisions, and inclusion of varying community risks and preferences in a decision analysis of COVID-19 mitigation measures in schools. Microsimulation models are often used to conduct health decision analyses for chronic disease screening and prevention strategies. These simulations track how individual-level risk factors change over time and predict disease incidence based on those risk factors, which allows decision-makers to model how different risk factor screening strategies (e.g., start atrial fibrillation screening at age 65) could impact incidence and health outcomes. In Chapter 1, we evaluate how accurately methods to parameterize these simulation models replicate observed data. In a case study using an ischemic stroke simulation model, we introduce the use of joint longitudinal and time-to-event models to parameterize the simulations and compare it to existing approaches that use standard regression and survival analysis techniques. We find that simulations parameterized with the joint model approach more accurately predict ischemic stroke incidence than existing parameterization approaches. These results indicate that joint longitudinal and time-to-event models can be used to create more precise simulation models and potentially improve chronic disease screening and prevention decision analyses. Intensive blood pressure treatment significantly reduces the risk of cardiovascular disease events, like heart attack and stroke, but can have significant drawbacks compared to standard options (e.g., side-effects, increased medical costs). Personalized decisions that target intensive treatment towards those most likely to benefit could help mitigate these trade-offs, and previous work has estimated the heterogeneous treatment effects (HTEs) of intensive treatment across different types of patients to guide treatment decisions. In Chapter 2, using data from the SPRINT trial, we estimated the HTEs of intensive treatment using a range of methods and evaluated whether making personalized treatment decisions based on these estimates would improve overall outcomes relative to a non-personalized decision. We found that using HTEs to guide treatment decisions did not improve outcomes if the objective was to only treat individuals who have at least a minimum level of benefit from treatment (e.g., 5% decrease in cardiovascular disease risk). However, when there was a constraint on the number of individuals who could receive intensive treatment and the objective was to allocate those treatments to those most likely to benefit (i.e., no minimum benefit level), using HTEs to assign treatment improved outcomes when individuals were stratified over a few subgroups. These results demonstrate that using HTEs to personalize treatment decisions would likely only improve outcomes in specific contexts and, more broadly, that it is important to account for the decision objective when evaluating HTE estimates. Mitigation measures like masks and ventilation have the potential to reduce COVID-19 incidence in elementary schools, but policies like mask mandates are generally temporary measures and come with drawbacks (e.g., potential impact on learning). When making decisions around mitigation policies, school districts must balance the health benefits of reduced COVID-19 incidence against these drawbacks. In Chapter 3, to help guide these decisions, we used an agent-based simulation model to estimate how COVID-19 incidence in elementary school communities is associated with in-school mitigation measures. We conducted the analysis across varying levels of local incidence in the surrounding area and vaccination within the school community. We found that school community incidence decreased with mitigation and vaccination and increased with local incidence. The local incidence thresholds for when school communities should change mitigation measures depended on their objective (e.g., keeping additional cases from ending mask mandates below 10 vs. 20 cases per month). These findings suggest that appropriate increases and decreases for in-school mitigation depend on policy makers’ goals, and that responsive plans, where mitigation is deployed based on local COVID-19 incidence and vaccine uptake, can help meet these goals.