Vanderbilt University School of Medicine
Dissertation Title: "Decision Modeling to Inform Resource Prioritization: Methods and Applications"
This dissertation is comprised of three chapters where I investigated the following policy and methodological questions in health decision science: the potential value of improvement in acute stroke quality measures in the US, the health and economic implications of TB prevention for recent cohorts initiating antiretroviral therapy (ART) in Tanzania, and the comparative performance of four model validation techniques in the context of disease simulation modeling.
In Chapter 1, I evaluated the cost-effectiveness and value of potential improvement in acute stroke quality measures in the US. Quality measures are valuable tools to guide improvement in quality of care and health outcomes. Excessive quality measurement, however, can cause inefficiency in the forms of administrative costs and misaligned incentives. My analyses using a stroke microsimulation model showed substantial variation and high concentration in the potential net benefit of quality improvement across ten quality measures. Early carotid imaging and tissue plasminogen activator treatment were the largest contributors to the total maximum value of quality improvement, accounting for 57% of the total value. The top five quality measures accounted for 90% of the total potential value, with the bottom five accounting for the other 10%. These findings can assist providers and payers in setting priorities for quality improvement and value-based payment in acute stroke care.
In Chapter 2, I examined the health impact and cost-effectiveness of isoniazid preventive therapy (IPT) for recent cohorts of people living with HIV initiating ART in Tanzania. IPT is the major TB prevention intervention for individuals receiving ART. As individuals are now initiating ART with better average immune function than in past years, they face TB incidence rates that are substantially lower than in historical cohorts. It is unclear whether earlier conclusions about the cost-effectiveness of IPT still hold. Using a TB/HIV coinfection model parametrized from detailed clinical data, I found that among more recent ART initiation cohorts, IPT resulted in greater health benefits and relatively stable incremental costs. These results show that the case for IPT provision among ART patients is even stronger now than in the past, and highlight the urgency of improving IPT coverage in contemporary ART programs.
In Chapter 3, I compared the relative performance of four validation methods in the context of simulation models: apparent performance, split-sample, cross-validation, and bootstrapping. In statistical modeling, the apparent performance (i.e., using the same sample for model development and validation) is known to result in an optimistic bias in estimating model performance, and more advanced methods have been developed to correct for this bias. However, advanced validation methods have been underused and understudied in simulation modeling. Using a stroke microsimulation model as a case study, I showed that bootstrapping achieved the lowest bias and variance among the four methods and should be considered good practice when developing and validating a simulation model based on individual-level datasets.