Innovation, Value, and Uncertainty in Oncology Precision Medicine

Abstract

Abstract Innovation, Value, and Uncertainty in Oncology Precision Medicine Devender Singh Dhanda Chair of Supervisory Committee: Professor David L. Veenstra Professor of Pharmacy Department of Pharmacy Background: There is substantial enthusiasm for precision medicine (PM), and its impact is more apparent in oncology as most new oncology drugs and diagnostics approved over the past decade are based on tumor genetics. PM research endeavors and developments have transformed the way we diagnose, prevent, and treat cancer. However, two major issues are yet to be addressed: 1) How the reimbursement policies have impacted the indication subdivisions versus the technological advances, and 2) Given the vast information that comes from current Next-Generation Sequencing (NGS) technologies, how much information should be used in clinical decision-making context. In this dissertation, I, therefore, developed 1) the quasi-experimental design to estimate the impact of insurance coverage expansions (Medicare part D implementation) on the potential indication size for oral oncology drugs; and 2) a decision-analytical modeling approach to estimate the uncertainty of adding the moderate/low risk genes to the targeted hereditary breast cancer screening (HBCS) gene-panels. Methods: In Chapter 2, I designed a quasi-experimental study design by exploiting the exogenous variation in the age at the cancer diagnosis for various cancer sites. This variation in age at diagnosis for different cancer sites helped us isolate the impact of Medicare part D from other contemporary developments such as genomics advances. I used several sources to build a unique dataset including the US Food and Drug Administration (FDA) website, Surveillance, Epidemiological, and End Results (SEER) program database, published and gray literature. I used a difference-in-difference approach to estimate the effect of Medicate part D on the indication size. In Chapter 3, I developed a decision-analytic model to estimate the long-term outcomes of using HCBS gene panels for breast cancer screening in high-risk women. I created three hypothetical gene panels and compared them to no testing strategy and amongst each other. I populated the model using clinical and economic inputs from published epidemiological literature, Surveillance, Epidemiological, and End Results (SEER) program website, and other published cost-effectiveness studies. The outcomes assessed were costs, life years gained (LEs), quality-adjusted life years (QALYs), incremental cost-effectiveness ratio (ICER), the incidence of breast cancer, and ovarian cancer. To characterize uncertainty, I performed one-way and probabilistic sensitivity analyses. Results: In Chapter 2, I demonstrated that the indication size for oral oncology drugs decreased by 37.5% in the post-Medicare part D period. I also estimated that for a one percent increase in the proportion of Medicare-eligible patients for a cancer site, the indication size for oral cancer drugs decreased by 3.28% (p-value=0.001). In evaluating the covariates as potential mediators, I demonstrated that neither the companion diagnostics nor the line of therapy at the time of drug approval contributed significantly to the causal effect. In Chapter 3, I demonstrated that the inclusion of moderate/low risk genes on the targeted panels increased the overall decision uncertainty. Specifically, all three screening strategies (Panel 1-3) dominated the no screening strategy. Both Panel 2 (BRCA1/2 plus other high-risk genes) and panel 3 (panel 2+ moderate/low-risk genes) dominated (better outcomes, lower cost) panel 1 (BRCA1/2 only) in the deterministic analyses. However, in probabilistic analysis accounting for uncertainty, panel 2 (BRCA1/2 and other high-risk genes) was the optimal strategy at a willingness to pay (WTP) threshold of $100,000 or less. Conclusions: I found that the labeled indication size for oral cancer drugs decreased in the post-Medicare part D period, and the effect was heterogeneous based on the proportion of Medicare-eligible patients for cancer sites at the time of diagnosis. Using the decision-analytical modeling approach, I found that the evidence is likely sufficient for the inclusion of high-risk genes onto the BRCA1/2 gene-panels; however, the evidence is not sufficient for the inclusion of moderate/low risk genes on the BRCA1/2 gene-panels.