Dissertation Title: "Decision Analysis in the Evaluation of Breast Cancer Treatment"
Advances in breast cancer treatment have improved the prognosis of many women with this disease. Whether a patient faces a diagnosis of early-stage breast cancer or more advanced disease, she and her doctor must choose a course treatment, weighing the benefits of each therapy against possible adverse effects. Similarly, policy makers must consider the benefits and harms of treatment alternatives, as well as their costs to society. Decision analysis is a method for systematically weighing all of these considerations, in order to identify an optimal treatment strategy.
In Chapter 1 decision analysis is used to estimate the cost-effectiveness of strategies for selecting metastatic breast cancer patients for trastuzumab treatment. In this situation, there is fundamental tradeoff between test cost and test accuracy - a more accurate test has a greater cost per patient. We found that use of the more accurate test is optimal in the long run, since this test minimizes the number of false positive results and their associated treatment costs. Our analysis demonstrates the importance of considering test cost and outcomes in assessing the cost-effectiveness of a targeted therapy.
In Chapter 2 decision analysis is used to examine the tradeoff between the efficacy and side effects of adjuvant therapy for early-stage breast cancer. We compared multi-agent chemotherapy with ovarian suppression in premenopausal women with hormone-responsive disease. We found that small relative differences in treatment efficacy had a substantial impact on outcome, regardless of treatment side effects. When treatments were equally efficacious, assumptions about menopause and quality of life affected the choice of therapy.
Chapter 3 involves the calibration of a decision-analytic adjuvant treatment model with population-based breast cancer survival estimates. We compared survival projected by the model with survival estimates from national cancer registry data, and identified a set of parameter values that provided the best fit between model projections and registry-based estimates. We found that breast cancer recurrence estimates suggested by calibration with population-based data differed from those found in clinical trials and those used in prior decision models. This analysis provides a framework for calibrating decision-analytic models with disease outcomes in the population, in order to improve the applicability of such models.