Dissertation Title: "Policy Assessment of Medical Imaging Utilization: Methods and Applications"
We developed and applied decision-analytic methods to two controversial uses of diagnostic imaging technologies: diagnosing Alzheimer’s disease (AD) with positron-emission tomography (PET) and screening for lung cancer with computed tomography (CT).
In Chapter 2, we evaluated strategies for the diagnosis of early-stage AD: usual care (a clinical exam plus a structural imaging test) with or without an additional functional imaging exam (PET or functional magnetic resonance imaging [fMRI]). Using our best estimates for inputs, PET yielded fewer benefits (at a higher cost) than usual care, and fMRI had an incremental cost-effectiveness ratio (vs. usual care) of $598,800/quality-adjusted life year (QALY). Our main conclusion was that resources should be allocated towards developing better therapies, not improving diagnostic tests. The published article informed Medicare’s National Coverage Decision, an example of the use of modeling to inform decision making in practice.
Chapter 3 reviews lung cancer screening trials and discusses two limitations of published cost-effectiveness analyses of lung cancer screening: their use of stage-shift models and their methods for accounting for smokers’ elevated competing mortality.
In Chapter 4, we estimated heart disease and other-cause (non-lung cancer, non-heart disease) mortality rates, stratified by gender, race (black/white), age group, and smoking status. These mortality rates were inputs for the model described below. We developed a Bayesian approach to synthesize survey data linked to mortality outcomes, vital statistics data, and published risk ratios from cohort studies, correcting for known inconsistencies between the datasets.
In Chapter 5, we developed a comprehensive microsimulation model of lung cancer, populated with cohorts of the U.S. population. The natural history model simulated lung cancer development, growth, and metastasis, not transitions between disease stages (e.g., local to regional), avoiding biases inherent in stage-shift models. Survival depended on the underlying disease stage and treatment received. The model ( screening) was calibrated to observed incidence rates, survival rates, and characteristics of cancers (cell type, stage, size), and validated by reproducing existing trial results (+ screening). As evidence-based decisions, policies, and guidelines become the standard among health care payers and providers, decision-analytic modeling and Bayesian evidence synthesis will serve as important tools for formally combining information.