University of Southern California Dissertations and Theses

Evaluating treatment options for metastatic, castration-resistant prostate cancer: a comprehensive value assessment

(USC Thesis Other)

Evaluating treatment options for metastatic, castration-resistant prostate cancer: a comprehensive value assessment

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content i Evaluating Treatment Options for Metastatic, Castration-Resistant Prostate Cancer: A Comprehensive Value Assessment by Samuel Crawford A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (HEALTH ECONOMICS) August 2022 Copyright 2022 Samuel Crawford ii Dedication For Miso Summer Reign Sasaki – for reminding me to breath over the past two years. iii Acknowledgements To my family, particularly my mother and father, for your unrelenting support from an early age, it led me to have the confidence to take a leap during a period where I felt lost by applying and enrolling in graduate school in a field, dominated by a topic and subject matter I had avoided during my bachelor’s degree and work experience prior to starting graduate school. To my girlfriend, Christine. When in a partnership, stress is shared, and these past few years have been among the most stressful I have ever faced. Thanks for taking on that burden, while being supportive of my goals. To Dr. J, about four and a half years ago you took a call from a random neuroscience lab tech and provided guidance to a stranger. I appreciate what you did then, and I appreciate the support you continued to provide throughout my time at USC. I hope you enjoy a well-earned retirement. To Michelle Ton. When I needed a hand navigating graduate school application, you played a huge role in getting me to where I am today. To the assistants of my committee members (Hanh, Alexis, and Laura), your availability helped keep this ship sailing as smooth as it could. Thank you to Karen Mulligan, Ian Davis, Drishti Baid for being a sounding board on different approaches to the problems examined in this dissertation. iv To my dissertation committee, Dr. William Padula, Dr. Darius Lakdawalla, and Dr. Mitchell Gross, thank you for your valuable insight and your support as I fumbled my way through this. To Dr. Jason Doctor and Dr. Charles Phelps, your consistent and timely assistance played a critical role in the development of this project. To my classmates, I am sorry we missed each other during the last 2 years, but I am excited to work with each other in the future. Lastly, to the University of Southern California School of Pharmacy, University of Southern California Graduate School, and AbbVie, thank you for your institutional support. v Table of Contents Dedication ....................................................................................................................................... ii Acknowledgements ........................................................................................................................ iii List of Tables ................................................................................................................................ vii List of Figures ................................................................................................................................ ix List of Abbreviations ..................................................................................................................... xi Abstract ........................................................................................................................................ xiii Chapter 1: Introduction ................................................................................................................... 1 1.1 Scientific Background and Rationale .................................................................................... 1 1.2 Research Objectives .............................................................................................................. 9 1.3 References ........................................................................................................................... 12 Chapter 2: Medical Utilization Associated with Post-Docetaxel Treatment Alternatives Among Patients with Metastatic, Castrate-Resistant Prostate Cancer ......................................................... 1 2.1 Introduction ......................................................................................................................... 18 2.2 Methods and Study Design ................................................................................................. 20 2.3 Results ................................................................................................................................. 27 2.4 Discussion ........................................................................................................................... 30 2.6 References ........................................................................................................................... 37 2.7 Appendix ............................................................................................................................. 53 Chapter 3: Higher Order Individual Utility Risk Parameters Associated with Health, among a United States Representative Sample ........................................................................................... 57 3.1 Introduction ......................................................................................................................... 58 3.2 Methods and Study Design ................................................................................................. 61 vi 3.3 Results ................................................................................................................................. 70 3.4 Discussion ........................................................................................................................... 79 3.5 Conclusion ........................................................................................................................... 85 3.6 References ........................................................................................................................... 86 3.7 Appendix .......................................................................................................................... 113 Chapter 4: Cabazitaxel vs. an Alternative Androgen-Signaling Target Inhibitor in Metastatic, Castrate-Resistant Prostate Cancer Patients Previously Treated with Docetaxel and an Alternative Androgen-Signaling Target Inhibitor: Applications of the Novel Generalized Risk- Adjusted Cost-Effectiveness (GRACE) Framework .................................................................. 124 4.1 Introduction ....................................................................................................................... 126 4.2 Methods and Study Design ............................................................................................... 128 4.3 Results ............................................................................................................................... 135 4.4 Discussion ......................................................................................................................... 139 4.5 Conclusion ......................................................................................................................... 147 4.6 References ......................................................................................................................... 148 4.7 Appendix ........................................................................................................................... 165 Chapter 5: Conclusions and Policy Implications ........................................................................ 174 5.1 Conclusions ....................................................................................................................... 174 5.2 Aims and Hypotheses Revisited ........................................................................................ 175 5.3 Policy Implications ............................................................................................................ 178 5.4 References ......................................................................................................................... 182 vii List of Tables Chapter 2 Tables ........................................................................................................................... 41 Table 2.1 Baseline Patient Demographics .................................................................................. 41 Table 2.2a Baseline Healthcare Resource Utilization ................................................................ 43 Table 2.2b Baseline Cost Burden ............................................................................................... 44 Table 2.3a All-Cause Cost Burden ............................................................................................. 45 Table 2.3b Cancer-Related Cost Burden .................................................................................... 46 Table 2.4 Average Standard Cost and Patient-Borne Therapy Cost .......................................... 47 Table 2.5 Clinical Outcomes per Treatment Regimen ............................................................... 48 Table 2.6 Adjusted Costs per Treatment Regimen 2L ............................................................... 49 Appendix Table 2.1. Adjusted Outcomes among Third Line Patients ....................................... 56 Chapter 3 Tables ........................................................................................................................... 92 Table 3.1 List of Choice Tasks ................................................................................................... 92 Table 3.2 Descriptive Statistics by Treatment Condition ........................................................... 93 Table 3.3 Descriptive Statistics by Treatment Condition ........................................................... 94 Table 3.4 Prevalence of Relative Risk Aversion and Relative Prudence ................................... 95 Table 3.5 Mean Approximated Certainty Equivalents ............................................................... 96 Table 3.6 Correlation Between Relative Risk Aversion and Relative Prudence ....................... 97 Table 3.7 Relative Prudence by Risk Averse Choices among Health Score Responses ............ 98 Table 3.8 Identifying Demographic Correlates with Relative Risk Aversion and Relative Prudence ..................................................................................................................................... 99 Table 3.9 A Parametric Estimates of Higher Order Parameters, Constant Relative Risk Aversion and Hyperbolic Absolute Risk Aversion (Simplified) .............................................. 102 viii Table 3.9 B Parametric Estimates of Higher Order Parameters, Hyperbolic Absolute Risk Aversion (Expanded) ................................................................................................................ 103 Table 3.9 C Parametric Estimates of Higher Order Parameters, Power Expo (Holt & Laury) 104 Table 3.9 D Parametric Estimates of Higher Order Parameters, Expo-Power (Saha) ............. 105 Table 3.10 Distribution of Higher Order Risk Estimates over Health Score and Cash – Relative Risk Aversion, Relative Prudence, and Relative Temperance ................................................. 106 Table 3.11 Ratio of Higher Order Risk Estimates over Health Score relative to Cash ............ 107 Appendix Table 3.1 Hyperbolic Absolute Risk Aversion Estimates Across Treatment Conditions ................................................................................................................................. 121 Appendix Table 3.2 Impact of Higher Order Estimates on Willingness-to-Pay Thresholds ... 122 Chapter 4 Tables ......................................................................................................................... 152 Table 4.1 Classic Model Parameters ........................................................................................ 152 Table 4.2 Model Clinical Outcomes ......................................................................................... 154 Table 4.3 Expected Base Case Results ..................................................................................... 155 Table 4.4 Two-Way Sensitivity Analysis ................................................................................. 156 Table 4.5 Base Case Scenario Analyses Results ...................................................................... 157 Table 4.6 Expected Base Case Results, Classic and GRACE Model ...................................... 158 Table 4.7 Hyperbolic Absolute Risk Aversion Higher Order Estimates Impact on GRACE Output ....................................................................................................................................... 159 Appendix Table 4.1 Generalized Risk-Adjusted Cost-Effectiveness Modeling, Literature Dependent Values ..................................................................................................................... 169 Appendix Table 4.2 Generalized Risk-Adjusted Cost-Effectiveness, Key Calculations ......... 170 ix List of Figures Chapter 2 Figures .......................................................................................................................... 50 Figure 2.1. Study Schematic ....................................................................................................... 50 Figure 2.2 Patient Disposition .................................................................................................... 51 Figure 2.3 Unadjusted Time to Next Treatment Kaplan-Meier Estimates ................................. 52 Chapter 3 Figures ........................................................................................................................ 108 Figure 3.1 Aversion Task ......................................................................................................... 108 Figure 3.2 Relative Prudence Task ........................................................................................... 108 Figure 3.3 Histogram of Risk Averse and Prudent Selections across Health Score Questions 109 Figure 3.4 Cumulative Distribution Functions (CDFs) of Responses on a Representative Risk Averse and Risk Prudent Questions ......................................................................................... 110 Figure 3.4 Distribution of Health Score Higher Order Parameter Estimates for Relative Risk Aversion, Relative Risk Prudence, and Relative Risk Temperance ......................................... 111 Figure 3.5 Histogram of Relative Risk Aversion over Health Score and over Cash ............... 112 Appendix Figure 3.1 Distribution of Hyperbolic Absolute Risk Aversion Coefficient Estimates .................................................................................................................................................. 123 Chapter 4 Figures ........................................................................................................................ 160 Figure 4.1 Markov Model ......................................................................................................... 160 Figure 4.2 Univariate Sensitivity Analyses .............................................................................. 161 Figure 4.3 Incremental Cost-effectiveness Plane. .................................................................... 162 Figure 4.4 Cost-Effectiveness Acceptability Curve. ................................................................ 163 x Figure 4.5 Expected Value of Perfect Information at Varying Willingness-to-Pay Thresholds. .................................................................................................................................................. 164 Appendix Figure 4.1 Impact of HARA Estimates on GRACE ................................................ 171 Appendix Figure 4.2 Impact of 𝜔𝐻 on GRACE Output .......................................................... 172 Appendix Figure 4.3 Impact of DEALE Method vs. Estimating Weibull Survival Curves .... 173 xi List of Abbreviations Abbreviation ANOVA One-Way Analysis of Variance ASTi Androgen-Signaling Targeted Inhibitor (Enzalutamide or Abiraterone) Cab Cabazitaxel CARA Constant Absolute Risk Aversion CCI Charlson Comorbidity Index CDFs Cumulative Distribution Functions CDM Optum Clinformatics Data Mart CE Certainty Equivalent CI Confidence Interval CO Certain Outcome CRPC Castrate-Resistant Prostate Cancer CRRA Constant Relative Risk Aversion Doc Docetaxel EVPI Expected Value of Perfect Information FDB First Databank GDP Gross Domestic Product GLM Generalized Linear Models GRACE Generalized Risk-Adjusted Cost-Effectiveness GRA-QALY Generalized Risk-Adjusted Quality Adjusted Life Year HA_RA Hyperbolic Absolute Risk Aversion - Relative Risk Aversion HA_RP Hyperbolic Absolute Risk Aversion - Relative Temperance HA_RT Hyperbolic Absolute Risk Aversion - Relative Prudence HARA Hyperbolic Absolute Risk Aversion ICD-10 International Statistical Classification of Diseases - Tenth Revision ICD-9 International Statistical Classification of Diseases - Ninth Revision ICER Incremental Cost-Effectiveness Ratio IGRACER Incremental Generalized Risk-Adjusted Cost-Effectiveness Ratio LOT Line of Therapy mCRPC Metastatic, Castrate-Resistant Prostate Cancer NDC National Drug Code NMB Net Monetary Benefit Obs Observations OOP Out-of-Pocket PC Prostate Cancer PD Progressed Disease PFS Progression Free Survival PPPM Per Patient Per Month PSA Prostate-Specific Antigen QALY Quality-Adjusted Life Year QoL Quality of Life Ref Reference xii RMSE Root Mean Squared Error RRA Relative Risk Aversion RRP Relative Risk Prudence SD Standard Deviation US United States WTP Willingness-to-Pay Threshold xiii Abstract Advanced prostate cancer is a leading killer of American males. Despite the development of new treatments, cost and clinical burden of metastatic, castrate-resistant prostate cancer (mCRPC) is staggering among patients and our health system. Recent work identifies optimal treatment sequences for those with mCRPC. In 2019, investigators leading the CARD clinical trial identified cabazitaxel as a potential treatment alternative to the standard of care, an alternative androgen-signaling targeted inhibitor (ASTi, abiraterone or enzalutamide), among mCRPC patients previously treated with docetaxel and an ASTi, enzalutamide or abiraterone. This dissertation had three primary research objectives, aimed at better informing an economic analysis of the treatment decision problem facing clinicians, patients, and other healthcare decision-makers. In the first chapter, we provide context regarding this specific decision problem and outline the necessary steps to conduct a thorough economic evaluation, using both the classic incremental cost-effectiveness ratio (ICER) analysis and the novel, generalized risk-adjusted cost-effectiveness (GRACE) framework. In the second chapter, we evaluate the economic burden facing patients treated with the therapies often seen in the decision problem presented here. In the third chapter, we present the results of a representative survey aimed at eliciting higher order risk estimates of patients facing treatment decisions over general health and a variety of treatment conditions pertaining to health and financial-based lotteries. In the fourth chapter, we present classic and novel economic evaluations of cabazitaxel relative to an ASTi among mCRPC patients treated with docetaxel and an alternative ASTi. Finally in the fifth chapter, we present our key findings in the context of hypotheses presented in chapter one and discuss critical policy implications of this research. xiv In this dissertation, we find that the economic burden for the treatment of mCRPC is substantial for patients undergoing either with an oral or intravenous-based chemotherapy regimen, with the intravenous-based regimen leading to greater overall per-patient per month costs. Despite this, patient out-of-pocket expenses are still greater for oral-based regimens, suggesting the need for a re-evaluation of medical and prescription benefits by insurers. From there, we characterize higher order risk estimates pertaining to health-care decision making. These estimates provide values that can play a critical role in re-appraising an appropriate willingness-to-pay threshold and feed into the novel GRACE framework for economic evaluations of medical technology in the United States. Finally, using values generated in the economic burden analysis, our representative survey, and published literature, we find that cabazitaxel is not a cost-effective alternative to an ASTi, among mCRPC patients treated with docetaxel and an alternative ASTi in most instances. However, the GRACE framework, with the inclusion of new parameters surrounding patient preference and drug effectiveness does provide some instances in which there may be a net benefit from cabazitaxel treatment in our modeled patient population. Publication of these higher order risk estimates and this early execution of GRACE will hopefully inform future executions of this novel cost-effectiveness analysis framework. 1 Chapter 1: Introduction Samuel A. Crawford 1.1 Scientific Background and Rationale Prostate cancer (PC) is one of the leading causes of death among U.S. males. 1 Despite significant advancements in diagnostics and treatment, over 190,000 men develop PC, and 33,000 die each year. 1 Of all male cancers, 21% are PC, which creates a high financial toll and economic impact on United States (US) healthcare. Total costs of PC diagnosis and treatment exceeds $10 billion per year. 2 Individuals with nonmetastatic castration-resistant PC average $35,000 per year, while patients with metastatic castration-resistant PC (mCRPC) face an astounding $155,000 per year per course of treatment. 2,3 MCRPC patients are usually treated initially with docetaxel. Patients may be treated with a hormonal therapy such as, androgen-signaling targeted inhibitor (ASTi) of either abiraterone or enzalutamide, either before or after docetaxel. 4 However, if the disease progresses, decision makers are tasked with prescribing an alternative ASTi (abiraterone or enzalutamide) or another chemotherapeutic agent. Recent evidence has suggested patients develop a tolerance for ASTi’s after an initial prescription. 4 This increased tolerance may lead to worse outcomes if a mCRPC patient is prescribed an alternative ASTi. Cabazitaxel represents a suitable subsequent chemotherapeutic. Cabazitaxel is an effective alternative to abiraterone and enzalutamide in patients with mCRPC. Despite clinical trial evidence that cabazitaxel is efficacious for mCRPC patients, the field could benefit from more generalizable, comprehensive economic evaluation of these findings. Furthermore, the higher cost of cabazitaxel can be questioned given the limited 2 marginal clinical outcomes (several months of progression-free and overall survival) it provides over abiraterone or enzalutamide in the treatment of post-docetaxel, post alternative ASTi mCRPC. Finally, patients may favor side-effects of ASTi (abiraterone or enzalutamide) over cabazitaxel. Patients with mCRPC tend to have poor outcomes with a low life expectancy, relative to patients with other forms of PC. This provides an excellent case-study for applying the novel generalized risk-adjusted cost-effectiveness (GRACE) modelling framework. The GRACE framework is an extension of classical modelling, accounting for patients’ risk-aversion over health-related quality of life (QoL). However, incorporation of the GRACE framework is dependent on three new parameters: relative risk aversion, relative prudence, and relative temperance. To incorporate this novel modelling framework, this dissertation elicited these values through a US representative survey. The field of oncology lacks comprehensive value assessment between preferential treatments for post-docetaxel mCRPC. We conducted a comprehensive value assessment using an observational patient cohort sampled from big data who received cabazitaxel and alternative ASTi for mCRPC patients treated with docetaxel, and alternative ASTi. GRACE adjusts the traditional cost-effective estimate by relaxing the assumption that patients are risk-neutral in health-related QoL and incorporating novel parameters such as value of hope or value of insurance. GRACE also provides a systematic update to the willingness-to-pay threshold. This in turn will allow us to create a more robust decision threshold estimate for cabazitaxel relative to an alternative ASTi in post-docetaxel, post alternative ASTi mCRPC patients. 3 Prostate Cancer PC is one of the leading killers of American men, disproportionally impacting marginalized communities and those with a family history of the disease. 5 Treatment for PC, similar to other cancers, is contingent on the following: 1) how far and likely the cancer will spread and 2) how will the patient respond to treatment. 5 This first component is captured in different tests, such physical and pathologic examination at the time of diagnosis as well as the results of laboratory tests, and the second component is captured through the generation of patient risk scores. PC is known is unique in that it can be diagnosed as a slow-growing disease. Because of this, some older and asymptomatic patients are initially followed with “watchful waiting” or observation. 5 During this phase, a patient is monitored through a series of tests, to ensure growth continues to remain slow. However, if the patient starts to by symptomatic or if a patient is diagnosed with a more aggressive, higher risk disease, patients are then treated with either local or systemic therapy. 5 Local therapy ranges from surgery to localized radiotherapy, whereas systemic therapy may include androgen-deprivation therapy (“hormonal therapy”). Manipulation of hormones can be instrumental in PC growth. As this cancer progresses, hormonal therapy becomes an important intervention. Common hormonal therapies, such as leuprolide acetate can block the major source of androgen production from the testicles while enzalutamide and abiraterone can block activating of the testosterone axis at the molecular level, limiting the downstream PC growth due to hormone activation from testosterone. 5 During the various treatment interventions and throughout care, physicians will evaluate prostate-specific antigen (PSA) levels in the blood. If a patient displays consecutive increases in PSA levels in the presence of medications meant to block testosterone, the cancer may be 4 deemed recurrent and castration-resistant, due to previous unsuccessful, hormonal-specific therapies. 5 At this point, physicians will then evaluate a series body scans, such as bone and compute tomography scans to evaluate possible metastatic spread of the cancer. 5 Findings from these tests will then tell the patient and provider whether the patient now has mCRPC. Clinical and Economic Burden Newly diagnosed PC was incident in 191,930 men in the U.S. in 2020. 6–8 It is the second leading cause of death among U.S. men with cancer, having resulted in 33,330 deaths, or 5.5% of all cancer deaths. 7 PC can be costly to health systems as well. Mean inpatient admissions cost $12,324 for metastatic PC and $10,987 for nonmetastatic disease. 9 Outpatient service costs averaged $1,627 for metastatic and $909 for nonmetastatic. 9 Mean per-patient-per-year all- caused annualized healthcare costs significantly increase from $35,000 among nonmetastatic PC patients to $155,000 among those with the metastatic form of the disease, highlighting the economic burden of mCRPC, relative to the non-metastatic form of the disease. 3 The total U.S. impact of PC exceeds $10 billion, with a large portion of that cost coming from patients with the metastatic, castrate-resistant form. 2 Metastatic Castrate-Resistant Prostate Cancer Castrate-resistant PC (CRPC) is hallmarked by disease progression despite androgen- deprivation therapy, and presents a combination of rise in serum levels related to prostate- specific antigen (PSA) as well as the appearance of new metastases. 10 A common treatment for more severe cases of PC are either hormonal or surgery-based castration. In the event castration is not successful and the patient experiences metastases, the patient then considered as mCRPC. 5 Metastases are usually accompanied with significant debilitation and poor prognoses. Of those patients with progressive CRPC, 90% will advance into bone metastases, which can produce significant morbidity, pain, and pathologic fractures. 10 Annual costs of care for men with mCRPC far exceeds that for non-castrate resistant PC, and outcomes are significantly worse, with estimated overall survival of 16 to 35 months. 11–13 Treatment Options Patients with mCRPC have several treatment options. For patients with new onset mCRPC, abiraterone, cabazitaxel, docetaxel and enzalutamide are the preferred options, with docetaxel taking priority. 5 In the event one of these options is not effective, clinicians may prescribe another of these four popular treatments or another less common therapy such as olaprib, prembrolizumab, radium-233 or mitoxantrone for pain relief. 5 However, despite having several potentially effective treatment options, cross-resistance may limit the effect of different treatment sequences. For example, patients may not have a subsequent response to abiraterone or enzalutamide after already having received an ASTi (abiraterone or enzalutamide). 14–17 The CARD clinical trial results, as presented by de Wit et al. (2019) suggests cabazitaxel is effective relative to an alternative ASTi (abiraterone or enzalutamide) in patients with mCRPC, post-docetaxel, post-ASTi (abiraterone or enzalutamide, Figure 1). 4 However effectiveness of cabazitaxel, relative to an alternative ASTi, following docetaxel and an alternative ASTi, still needs to be validated in a real-world setting. 6 Expanding Breadth of Value Assessment Cost-effectiveness analysis historically has focused on two key attributes, incremental costs and incremental benefit of one therapy relative to another comparator of interest. Both attributes are determined by evaluating the net costs and net clinical benefit scaled by quality, per arm of comparison. From there, the incremental cost is divided by the incremental gain and compared to willingness-to-pay threshold to gauge if a therapy is deemed valuable (or cost- effective). Analysts and researchers have done considerable amount of work to improve and standardize this framework. 18–20 Despite cost-effectiveness analysis commonly using the incremental cost-effectiveness ratio as the core analysis for conducting economic evaluations, not all pieces of value are captured within this routine analysis. The value flower was published in 2018, as an effort to identify previously overlooked value components when evaluating new medical technology. 21 Some attributes presented in the value flower are core to economic analyses, and that include the aforementioned costs and clinical benefit. Other commonly included value attributes presented in the flower include productivity and adherence-improving factors. However, the flower also presents some novel elements, such as value of reduction of uncertainty due to a new diagnostic, fear of contagion, insurance value, severity of disease, value of hope, real option value, equity, and scientific spillovers. Each of these items can be included in some capacity as a sensitivity or scenario analysis. Until now, there has not been an update to the classic incremental cost- effectiveness ratio framework which rigorously and systematically includes some of these novel elements. For the context of the decision problem being explored in this dissertation, the following elements will be explored or discussed in some capacity over the next few aims: net 7 costs, QALYs gained, adherence-improving factors, insurance value, severity of disease, and value of hope. Net costs and QALYs gained are central to any economic evaluation and will be central to the analysis conducted in this dissertation. Costs will be evaluated during an evaluation of a large claims database, Optum Clinformatics DataMart (CDM). QALYs will be determined using published estimates from the CARD clinical trial, the basis for this study, and published estimates of quality-of-life adjustments for mCRPC patients treated with an ASTi or cabazitaxel, following treatment of an alternative ASTi and docetaxel. Adherence-improving factors are often reflected in the net value of the new medical technology and can be teased out through a real- world analysis of the decision problem being studied. However, in this context, transition probabilities and clinical outcomes will be informed using a published randomized clinical trial. For patient clinical outcomes captured in a randomized clinical trial, analysts may struggle to identify if an outcome is due to adherence-improving factors due to a new medical technology or if adherence improvement is due to the clinical trial setting. Because of this, we cannot opine as to if the clinical outcomes we see in the later economic evaluation are due to adherence- improving measures or not. However, financial toxicity can have an impact on medication adherence for patients taking oral agents such as abiraterone and enzalutamide. 22 Thus, those adherence-improving factors, or in this case adherence-worsening factors will be discussed in aim 1 as treatment features clinicians and patients should consider when prescribing an ASTi. Over the last few years, the GRACE framework has been introduced as a novel value assessment framework which can rigorously account for some elements of the value flower that have previously been overlooked in base economic evaluation. 23–25 The GRACE framework was derived under the reasonable assumption that people are risk averse over health-related quality- 8 of-life. With this new value assessment framework, insurance value, severity of disease, and value of hope now are included within a base evaluation using relative risk parameters determined over health-related quality-of-life. Insurance value reflects protection from both health and monetary risk. Severity of disease reflects how the valuations for new therapies for severe diseases may be higher than new therapies for less severe diseases. Value of hope reflects a risk-seeking patients desire for a high variance drug due to the off-chance they stand to be on the benefitting end of that variance. Each of these elements are captured using the GRACE method in a rigorous fashion and are also critical to the treatment decision problem seen in this study. Value of insurance examined using the GRACE framework refers to reducing the variance, and thus health risk associated with uncertain outcomes. Value of hope refers to the potential patient or subject desire for variable, but improved patient outcomes for a therapy for a severe disease. Disease severity is self-explanatory, however is implemented in the GRACE framework using a function of higher order generalizable risk parameters and signals for disease therapy. Again, each of these elements are important features of the decision problem between cabazitaxel and an alternative ASTi for the treatment of post-docetaxel, post-ASTi mCRPC patients. Through the execution of this study, this will likely have a role as a potential scaffold for future GRACE analysis which now incorporate these previously, not commonly or systematically included value elements. Innovation This proposal is one of the first initiatives to implement the GRACE modelling framework for an economic evaluation. Clinical trials by themselves may inform treatment decision-making for patients with mCRPC. However, we can use multiple methods from the 9 decision sciences, grounded in real-world evidence, economic modeling, and patient preferences to further validate and substantiate the findings seen by de Wit et al. (2019). Real-world evidence is already used by the U.S. Food and Drug Administration to substantiate drug effectiveness and safety within its own Sentinel System 26 , but there is considerable momentum to expand real- world evidence analyses and studies to other available datasets, such as Optum Claims Database, available through the Schaeffer Center servers. This study will utilize real-world evidence to determine economic burden of mCRPC patients treated with either cabazitaxel or an alternative ASTi, post-docetaxel, post-ASTi. Real-world evidence analysis will also then be directly applied to a classical and the novel GRACE modelling framework to determine the cost-effectiveness of cabazitaxel relative to alternative ASTi in the target patient population. Furthermore, higher order risk parameters estimated in our survey can then be used to populate future GRACE applications. This study will represent a culminative and innovative effort combining data science and decision sciences to inform a complex decision-making process for patients and providers, specifically for patients with mCRPC. If successful, the GRACE application of this study can be used as a standard guide for future GRACE evaluations. 1.2 Research Objectives AIM 1: To determine the medical costs and healthcare resource utilization associated with different post-docetaxel treatment alternatives among patients with progressed PC. A1.1: We will calculate drug standard cost using Optum Claims Database. H1.1: Cabazitaxel comes at a significantly higher average list price per course of treatment relative to an ASTi (p<0.05). 10 A1.2: We will analyze health system treatment costs including, disease-related and - unrelated costs for PC. H1.2: Medical costs vary significantly depending on the associated PC therapeutic (p<0.05). AIM 2: Develop and deploy a survey instrument to calculate precise estimates of higher order risk attributes with respect to health-state utility. Analyze the responses captured in a US representative sample, to calculate relative risk aversion, relative risk prudence, and relative risk temperance. A2.1: We will develop and pilot an investigational survey to elicit relative risk aversion and relative prudence, evaluating five health components. H2.1: Pilot survey participants will demonstrate relative risk aversion and relative prudence over health-related states. A2.2: After successful piloting, we will deploy this instrument to a representative, US sample. H2.2: A US, representative sample will demonstrate greater relative risk aversion and relative prudence regarding health-related QoL than a similar population in regard to financial compensation. 11 AIM 3: Conduct a classical and novel economic evaluation, comparing cabazitaxel to an alternative ASTi (abiraterone or enzalutamide) in patient previously treated with docetaxel and an alternative ASTi (enzalutamide or abiraterone), using the GRACE modelling framework to account for higher order relative risk preferences over general health values, value of information, value of hope, and value of uncertainty. A3.1: We will calculate and determine health state utilities per progressed disease and progression-free survival for each comparator in this evaluation, using published literature. H3.1: Health-related QoL is higher for post-docetaxel, post-alternative ASTi mCRPC patients treated with an alternative ASTi than those patients treated with cabazitaxel. A3.2: We will develop an economic model of mCRPC treatment alternatives to the assess incremental cost-effectiveness ratio (ICER) from a U.S. societal perspective and healthcare sector perspectives at a cost-effectiveness threshold of $100,000 per quality-adjusted life year (QALY). H3.2: Cabazitaxel is cost-effective in the treatment of post-docetaxel, post-alternative ASTi mCRPC. A3.3: We will develop a GRACE model to account and adjust for patient preferences which may be risk averse over health related QoL, for patients with mCRPC being treated with an alternative ASTi or cabazitaxel. H3.3: For patients with mCRPC, the willingness-to-pay for effective treatments will be higher than conventionally and classically used thresholds, and cabazitaxel will be a cost- effective comparator to alternative-ASTi for post-docetaxel, post-alternative ASTi mCRPC patients. 12 1.3 References 1. Street W. Cancer Facts & Figures 2020. Published online 1930:76. 2. Yabroff KR, Lund J, Kepka D, Mariotto A. Economic Burden of Cancer in the United States: Estimates, Projections, and Future Research. Cancer Epidemiol Biomarkers Prev. 2011;20(10):2006-2014. doi:10.1158/1055-9965.EPI-11-0650 3. Appukkuttan S, Tangirala K, Babajanyan S, Wen L, Simmons S, Shore N. A Retrospective Claims Analysis of Advanced Prostate Cancer Costs and Resource Use. PharmacoEconomics Open. Published online October 22, 2019. doi:10.1007/s41669-019- 00185-8 4. de Wit R, de Bono J, Sternberg CN, et al. Cabazitaxel versus abiraterone or enzalutamide in metastatic prostate cancer. New England Journal of Medicine. 2019;381(26):2506-2518. 5. NCCN Guidelines for Patients Prostate Cancer. Prostate Cancer. Published online 2019:106. 6. Prostate cancer statistics. World Cancer Research Fund. Published August 22, 2018. Accessed November 10, 2020. https://www.wcrf.org/dietandcancer/cancer-trends/prostate- cancer-statistics 7. Cancer of the Prostate - Cancer Stat Facts. SEER. Accessed November 10, 2020. https://seer.cancer.gov/statfacts/html/prost.html 8. Street W. Cancer Facts & Figures 2018. Published online 1930:76. 9. Tangirala K, Appukkuttan S, Simmons S. Costs and Healthcare Resource Utilization Associated with Hospital Admissions of Patients with Metastatic or Nonmetastatic Prostate Cancer. Am Health Drug Benefits. 2019;12(6):306-312. 10. Hotte SJ, Saad F. Current management of castrate-resistant prostate cancer. Curr Oncol. 2010;17(Suppl 2):S72-S79. 11. Mottet N, Bellmunt J, Bolla M, Cornford P, De Santis M. EAU - ESTRO - SIOG Guidelines on Prostate Cancer. Published online 2016. 12. Yabroff KR, Lund J, Kepka D, Mariotto A. Economic Burden of Cancer in the United States: Estimates, Projections, and Future Research. Cancer Epidemiol Biomarkers Prev. 2011;20(10):2006-2014. doi:10.1158/1055-9965.EPI-11-0650 13. Appukkuttan S, Tangirala K, Babajanyan S, Wen L, Simmons S, Shore N. A Retrospective Claims Analysis of Advanced Prostate Cancer Costs and Resource Use. PharmacoEconomics Open. Published online October 22, 2019. doi:10.1007/s41669-019- 00185-8 13 14. Loriot Y, Bianchini D, Ileana E, et al. Antitumour activity of abiraterone acetate against metastatic castration-resistant prostate cancer progressing after docetaxel and enzalutamide (MDV3100). Annals of oncology. 2013;24(7):1807-1812. 15. Maines F, Caffo O, Veccia A, et al. Sequencing new agents after docetaxel in patients with metastatic castration-resistant prostate cancer. Critical reviews in oncology/hematology. 2015;96(3):498-506. 16. Attard G, Borre M, Gurney H, et al. Abiraterone alone or in combination with enzalutamide in metastatic castration-resistant prostate cancer with rising prostate-specific antigen during enzalutamide treatment. Journal of Clinical Oncology. 2018;36(25):2639. 17. Oh WK, Cheng WY, Miao R, et al. Real-world outcomes in patients with metastatic castration-resistant prostate cancer receiving second-line chemotherapy versus an alternative androgen receptor-targeted agent (ARTA) following early progression on a first- line ARTA in a US community oncology setting. In: Urologic Oncology: Seminars and Original Investigations. Vol 36. Elsevier; 2018:500. e1-500. e9. 18. Neumann PJ, Kim DD, Trikalinos TA, et al. Future Directions for Cost-effectiveness Analyses in Health and Medicine. Med Decis Making. 2018;38(7):767-777. doi:10.1177/0272989X18798833 19. Sanders GD, Neumann PJ, Basu A, et al. Recommendations for Conduct, Methodological Practices, and Reporting of Cost-effectiveness Analyses: Second Panel on Cost- Effectiveness in Health and Medicine. JAMA. 2016;316(10):1093-1103. doi:10.1001/jama.2016.12195 20. Garber AM, Phelps CE. Economic foundations of cost-effectiveness analysis. Journal of Health Economics. 1997;16(1):1-31. doi:10.1016/S0167-6296(96)00506-1 21. Lakdawalla DN, Doshi JA, Garrison LP, Phelps CE, Basu A, Danzon PM. Defining Elements of Value in Health Care—A Health Economics Approach: An ISPOR Special Task Force Report [3]. Value in Health. 2018;21(2):131-139. doi:10.1016/j.jval.2017.12.007 22. Caram MEV, Oerline MK, Dusetzina S, et al. Adherence and out-of-pocket costs among Medicare beneficiaries who are prescribed oral targeted therapies for advanced prostate cancer. Cancer. 2020;126(23):5050-5059. doi:10.1002/cncr.33176 23. Lakdawalla DN, Phelps CE. Health Technology Assessment With Diminishing Returns to Health: The Generalized Risk-Adjusted Cost-Effectiveness (GRACE) Approach. Value in Health. 2021;24(2):244-249. doi:10.1016/j.jval.2020.10.003 24. Lakdawalla DN, Phelps CE. Health technology assessment with risk aversion in health. Journal of Health Economics. 2020;72:102346. doi:10.1016/j.jhealeco.2020.102346 25. Lakdawalla DN, Phelps CE. Evaluation of Medical Technologies with Uncertain Benefits. National Bureau of Economic Research; 2019. doi:10.3386/w26058 14 26. Corrigan-Curay J, Sacks L, Woodcock J. Real-World Evidence and Real-World Data for Evaluating Drug Safety and Effectiveness. JAMA. 2018;320(9):867-868. doi:10.1001/jama.2018.10136 15 Chapter 2: Medical Utilization Associated with Post-Docetaxel Treatment Alternatives Among Patients with Metastatic, Castrate-Resistant Prostate Cancer Samuel A. Crawford, Mitchell E. Gross, William V. Padula, Abstract Objective: Treatments for second and third line metastatic, castrate resistant prostate cancer (CRPC) have undergone a shift based on results of a recent clinical trial exploring cabazitaxel (Cab) or androgen-signaling targeted inhibitors (ASTi, including abiraterone or enzalutamide in docetaxel-treated CRPC. Evidence now suggests Cab is superior to a second androgen-signaling targeted inhibitor (ASTi) in these patients. However, as healthcare decision-makers adopt decision pathways, the associated costs with these regimens must be evaluated to appropriately assess cabazitaxel as an effective treatment option. Methods: This retrospective claims analysis used administrative data from Optum Clinformatics™ Data Mart database. Patients were identified as having metastatic, CRPC if they had a claim indicating confirmation of a prostate cancer diagnosis, evidence of metastatic disease, and two or more line of therapy within our treatment cohorts of interest. Using our inclusion and exclusion criteria, we identified 961 patients treated with different sequences of ASTi, docetaxel (Doc), and cabazitaxel (Cab) for CRPC. Specifically, the category “ASTi- ASTi” is used to capture subjects treated with one ASTi (abiraterone or enzalutamide) followed by the other ASTi in any order. “ASTi-Doc” and “ASTi-Cab” captures subjects treated with an ASTi followed by docetaxel or cabazitaxel or vice versa, respectively. As an additional 16 comparator, we also identified 35 patients who were treated with ASTi-Doc-ASTi and ASTi- Doc-Cab, to examine cost-burden associated with treatment sequalae laid out in the CARD clinical trial. Patients treated with ASTi-Doc-ASTi or ASTi-Doc-Cab are inclusive of those captured in the ASTi-Doc second line cohort. However, among third line captured patients, only costs and clinical outcomes relative to the third line were observed (i.e. . from 3 rd line start to 3 rd line end). Whereas, for second line patients, only costs and clinical outcomes were abstracted relative to the second line of therapy (i.e., from 2 nd line start date to 2 nd line end date). Baseline healthcare resource utilization and cost-burden, treatment line cost-burden, standard costs for treatment during specific line of therapy, and adjusted costs were examined in this analysis. Results: Cost-burden was high for all patient cohorts evaluated. However, ASTi-Doc and ASTi- Cab were associated with a significant increase in cost-burden relative to ASTi-ASTi. Total per-member per month cost burden was $35,747, $56,854, and $53,165 for patients in the ASTi-ASTi, ASTi-Doc, and ASTi-Cab cohorts. Total cost burden was $211,248 and $185,188 among patients treated with ASTi-Doc-ASTi and ASTi-Doc-Cab, in third-line treatment cohorts. This significant increase was not passed down to patients, as the patient borne costs was significantly less for the ASTi-Cab cohort relative to other cohorts. Differences in third lines treatment cohorts were not observed, which may be due to lack of sample size. Conclusions: This large administrative claims-based data analysis showed significant total medical treatment cost-burden for patients treated with for later line metastatic, CRPC. Despite, significant total medical costs differences, out-of-pocket costs did not rise with total medical 17 costs. However, patient-responsible expenses were high and should be examined in future studies due to their potential impact on treatment adherence, persistence, and in-turn clinical outcomes. 18 2.1 Introduction Prostate cancer is one of the most common cancers in the United States and is a leading cause of death. In 2021, 248,530 men were diagnosed with prostate cancer and 34,130 died of prostate cancer. 1 While most patients are cured with surgery or radiation, metastatic prostate cancer develops in a subset of patients. Regional forms of metastatic, castrate-sensitive prostate cancer are associated with a very high (>98%) 5-year survival rate. However, progression to distant castrate-resistant prostate cancer (CRPC) is associated with a much lower 5-year survival rate, around 30.7%. 2–4 In 2019, de Wit et al. published the results of the pivotal CARD clinical trial. 2 The basis of this trial focused on the drug mechanisms of the current early treatment options for metastatic, CRPC, the androgen-signaling targeted inhibitors, abiraterone and enzalutamide. Given both have a similar mechanism of action, the CARD clinical trial found that for patients being treated with late lines of therapy, post-docetaxel and post-ASTi (abiraterone or enzalutamide) use, cabazitaxel is a clinical effective alternative option to an alternative ASTi (enzalutamide or abiraterone). A leading theory behind this finding relates to cross-resistance between ASTi agents. 5,6 Among CRPC patients treated with an ASTi or intravenous therapy, such as cabazitaxel or docetaxel, payment coverage differs considerably. Cabazitaxel and docetaxel are covered through medical coverage by payers due to being intravenous therapies administered within an outpatient clinic, while enzalutamide and abiraterone are covered under prescription benefits due to being oral agents taken at by the patient at home. Costs for all four therapies are well- documented as being expensive. 7 However, prescription benefit shortfalls lead to a larger financial toxicity among patients and worse adherence among CRPC patients prescribed 19 enzalutamide or abiraterone. 8 We aim to better characterize the cost differences born by both patient and payer considering differences in medical and prescription coverage. Treatments for metastatic, CRPC are notoriously expensive in total cost from both the payer and patient perspectives. 8,9 Some studies have examined the impact on treatment options for patients with metastatic, CRPC on healthcare resource utilizations and costs. 10–14 However, the dynamic at which a CARD trial-based treatment regimen for metastatic castration-resistant prostate cancer could impact patient and payer cost burden relative to the original standard of care (abiraterone or enzalutamide, to enzalutamide or abiraterone) has not been recently characterized in a United States setting. Recent estimates on the medical economic burden given standard of care treatment sequalae relative to CARD trial-inspired treatment sequences can benefit the field, by better characterizing the economic and patient financial toxicity associated with each regimen, considering improvements in clinical outcomes as published by de Wit. This study aims to refresh estimated healthcare resource utilization and cost burden associated with treatments among a vulnerable population of metastatic, CRPC patients, particularly those treated with an alternative ASTi in a second- or third-line setting following the use of an early ASTi relative to those who are treated with docetaxel or cabazitaxel, two chemotherapeutic agents. In this study, healthcare resource utilization and cost burden were identified in patients with advanced metastatic prostate cancer (CRPC) and their corresponding using medical administrative claims captured in Optum Clinformatics™ Data Mart (CDM) database. Patients were identified with metastatic CRPC based on the presence of a diagnosis claim for prostate cancer and treatment with 2 or more of the following: enzalutamide, abiraterone, cabazitaxel or docetaxel. Unadjusted and adjusted cost burden associated with second and third lines of therapy 20 were aggregated and calculated per comparator arms in this study. In this study, we also identify treatment list costs per specific lines of therapy and potential impact of associated medical or treatment costs on potential patient outcomes. 2.2 Methods and Study Design Data Source and Sample Selection This study was a retrospective cohort claims analysis using comprehensive medical claims from CDM. CDM is one of the largest payer-owned integrated claims databases of a commercial insurer in the United States. This dataset has been used in several retrospective, real- world pharmacoeconomic studies. 15–19 At the time of this analysis, CDM covered tens of millions of unique members with captured medical and pharmacy claims. Furthermore, United Healthcare’s coverage through Medicare Advantage is also well covered in CDM, translating to a representative sample of older patients covered with commercial and government-based insurance coverage. Several demographic measures, such as race, gender, state, and plan type are also captured in this database. Available data also includes patient diagnoses as captured by International Statistical Classification of Diseases, Ninth and Tenth Revisions (ICD-9 and ICD- 10). Commercial medical claims from many healthcare sites are captured in this dataset, including hospital visits, inpatient admissions, outpatient visits, emergency room visits, pharmacy claims, and other medical based encounters. In this study, the following patients were identified from the CDM within a 11-year time- period from January 1, 2010, to December 31, 2020 (See Figure 2.1 for an overview Schematic of this study design): 21 (1) Total prostate cancer population: including patients with at least 1 prostate cancer diagnosis (ICD-9 code: 185.XX; ICD-10 code: C61), who were enrolled in the database for at least 6 continuous months. (2) Patients with at least 1 claim for metastatic disease, as identified using the following codes (ICD-9 code: 196-199, 2097; ICD-10 code: C77-C80, C78). The first claim for metastatic disease must have occurred following the first captured diagnosis for prostate cancer. (3) Patients with no recorded flag for hormone sensitivity (ICD-10: Z191) 1 . (4) Pharmacy claims for either an androgen-signaling targeted inhibitor (ASTi, abiraterone or enzalutamide), docetaxel, and/or cabazitaxel. Comparators Pharmacy claims were delineated by line of therapy. Lines of therapy were identified using drug start date, days supplied, and a grace period. End of line of therapy was identified using: 1) a 60-day treatment gap for all drugs in treatment regimen; 2) no gap, but an end in patient data capture; and 3) line is interrupted by a qualifying refill of a non-line drug. A 14-day grace period was used to identify qualifying drugs within a treatment regimen for a specific line of therapy. That is, if cabazitaxel was started on day 1, and a second antineoplastic was started on day 7, that was considered the second drug within the cabazitaxel treatment cocktail. This study centered on mono-therapeutic lines. In the event a patient had multiple therapies along with our treatments of interest (abiraterone, enzalutamide, docetaxel or cabazitaxel), those patients were excluded. 1 Claims for hormone sensitivity is likely not a consistent or reliable measure for identifying prostate cancer patients who are castrate resistant. Furthermore, it is unclear as to if clinicians routinely or reliably will list a condition necessitating this claim from a medical bill coder. However, in the event a patient is listed as being “hormone sensitive,” analysts for this study wanted to be sure not to include those patients. 22 Lines of therapy were used to identify 2 groups of key cohorts of interest, as detailed below. Group 1 – 2 nd Line of Therapy Cohorts a. ASTi-ASTi: Patients treated with an 1) ASTi (abiraterone or enzalutamide) and then 2) an alternative ASTi (enzalutamide or abiraterone). b. ASTi-Doc: Patients treated with an 1) ASTi or Docetaxel and then 2) the alternative (Docetaxel or an ASTi). c. ASTi-Cab: Patients treated with an 1) ASTi or Cabazitaxel and then 2) the alternative (Cabazitaxel or an ASTi). Group 2 – 3 rd Line of Therapy Cohorts a. ASTi-Doc-ASTi: Patients treated with 1) an ASTi or Docetaxel, followed by 2) Docetaxel (if treated with an ASTi) or an ASTi (if treated with Docetaxel), followed by 3) an alternative ASTi (abiraterone if first ASTi was enzalutamide or enzalutamide if first ASTi was abiraterone). b. ASTi-Doc-Cab: Patients treated with 1) an ASTi or Docetaxel, followed by 2) Docetaxel (if treated with an ASTi) or an ASTi (if treated with Docetaxel), followed by 3) an Cabazitaxel. Patients treated with an alternative ASTi or cabazitaxel in the third-line setting are a subset of those captured in the second-line cohorts. However, costs and clinical outcomes will be aggregated according to their respective cohort line (e.g. outcomes aggregated over second line start to second line end for patients in the second line cohort, outcomes aggregated over third line start to third line end for patients in the third line cohort.) and First Databank (FDB) drug data was used to flag therapies of interest after matching descriptors from FDB to national drug code (NDC) in CDM patient claims. 23 Patient Baseline Characteristics Baseline characteristics during the 6-month baseline period prior to the first date of metastatic disease were captured and presented for each cohort in this study. Key demographic variables captured included the following: age, gender, region (as coded from patient state), race, plan type (Medicare or Commercial), time to first treatment (captured in months), starting year of first line of therapy, time between first identified prostate cancer diagnosis date and first captured treatment for metastatic disease, and Charlson comorbidity index. Outcomes Baseline Healthcare Resource Utilization Healthcare resource utilization measures were assessed during the baseline period to contextualize economic burden assessed per cohorts of interest. The following measures were captured as baseline proxies for healthcare resource utilizations: inpatient admissions, outpatient visits, emergency room visits, hospice admissions, and other medical encounters. Measures were captured using American Medical Associated place-of-service codes. Healthcare resource utilizations measures were scaled per patient per month (PPPM), for ease of comparison between study comparators. PPPM counts are presented as mean (standard deviation [SD]) for each type of utilization, per comparator of interest. Cost Measures PPPM healthcare costs were captured in the 6-month baseline period and per highlighted line of treatment for comparators of interest (line 2 for 2 line of therapy cohorts and line 3 for 3 line of therapy cohorts). PPPM costs were determined for all-cause and prostate cancer related. 24 Prostate cancer-related medical costs were medical costs assigned to a prostate cancer diagnosis flag. Cancer-related pharmacy costs were aggregated based on FDB flags for antineoplastics. Costs were stratified by medical and pharmacy related. Medical costs were the aggregate of outpatient visit, inpatient admission, emergency room, hospice-related, or other costs. Costs are presented as comprehensive, payer-based, and patient-borne. Comprehensive costs are the aggregate of all costs. Payer-based are generated from reported medical charges. Patient-borne costs are the aggregate of copay, coinsurance, and deductible. Comprehensive pharmacy and medical costs were based on all available claims for each cohort and each treatment of interest. PPPM cost outcomes are presented as mean (SD). Average wholesale and patient-borne costs per drug per group of interest (two or three lines) were also calculated to determine the standard drug cost captured in CDM and the patient- borne costs for the therapies of interest. Costs per drug per group of interest is reported by drug of interest, PPPM. All costs were scaled to 2021 dollars using bureau of labor statistics consumer purchasing index. 20 Clinical Outcomes Clinical outcomes were also examined in these data. Mortality was captured using the DOD version of CDM and were calculated from the end of second line of therapy to death in months among the second line cohorts and third line of therapy to death in months among the third line cohort. Time to next treatment was used as proxy to evaluate progressed disease, under the assumption that a patient treated with another line of an antineoplastic has some form of a progressed disease. 21 Several studies have used time to next treatment as a proxy for progressed disease. 22–24 25 Statistical Analysis Healthcare resource utilization and comparator arm-related costs were calculated using patient treatment claims and resource utilization measures. Mean and SD were calculated for continuous demographic variables (e.g. age), healthcare resource utilization, and costs. Proportions were calculated for discrete demographic variables. One-way analysis of variance (ANOVA) was used to evaluate differences in unadjusted baseline characteristics and post- baselines economic burden and clinical outcomes between line 2 treatment sequences and a t-test was used to compare outcomes among line 3 treatment sequences. Factors impacting differences in costs are assessed using generalized linear models (GLM) with gamma distributions, controlling for a vector of the following demographic variables: age, gender, region (as coded from patient state), race, plan type (Medicare or Commercial), time to first treatment (captured in months), starting year of first line of therapy, time between first identified prostate cancer diagnosis date and first captured treatment for metastatic disease, and Charlson comorbidity index (Equation 2.1) GLM with gamma distributions and log-link is widely used for producing adjusted healthcare cost burden estimates due to skewness and variability of healthcare cost data. 25–27 Among regression and other GLM functions, GLM with gamma distribution and log link is favored due to also having flexibility when dealing with bounded or categorical outcomes. 28,29 Generalized Linear Model, with Gamma Distribution and Log Link ln ⁡( 𝐸 ( 𝑌 𝑗 ) ) = 𝑋𝛽 , 𝑌 𝑗 ~ 𝐺𝑎 𝑚𝑚 𝑎 ⁡( 𝘀 , 𝜃 ) Eq. 2.1 • 𝑌 𝑗 represents a patient vector of outcomes • 𝑋 represents a matrix of demographic or clinical characteristics • 𝛽 measure the impact or effect size of covariates 26 • 𝘀 is the location parameter of the gamma distribution • 𝜃 is the rate parameter of the gamma distribution Pairwise correlations were conducted to examine the association between comparator group of interest drug cost and clinical outcomes (time to next treatment and mortality). Outcomes with fewer than 12 observations were censored for patient data safety. 11 observations is commonly used as the cut-off when presenting clinical outcomes, due to the reasonable ease at which a bad actor can infer a patient’s identifiable information within a study when clinical outcomes at that level are reported. 30,31 To determine the sample size necessary to find a significant effect of therapeutic costs and treatment sequalae effect on outcomes using a linear regression model, we completed a power calculation using Stata’s “power cox” command, we calculated the necessary sample size to demonstrate a difference between patients treated by one of the established treatment sequalae, assuming no censoring, a default power of 0.8, an odds ratio of 0.54, and a significance level of α = 0.05. 32 The estimated sample size to identify a significant difference is 83 subjects. The formal calculation behind the Stata command is based off of study design of clinical trials and observational analysis, published by Dupont and Plummer (1998). 33 More than 83 subjects exist for identifying differences between second lines of therapy. However, less than 83 are available for comparison of third line outcomes. Due to this limitation, some outcomes will only be evaluated for second lines of therapy. Comparisons with limited sample size will be considered exploratory and hypothesis generating for future analyses on either larger datasets or refreshed CDM with more patients that meet the third line of therapy criteria. 27 Statistical significance was defined at the level of 5% ( 𝛼 = 0 . 05 ). All statistical analysis was conducted using SAS software, version 9.4 (SAS Institute Inc., Cary, NC), and STATA software version 14.0 (StataCorp LP, College Station, TX). 2.3 Results Patient Disposition With the criteria listed above, 530,364 patients from the CDM were identified as having prostate cancer (Figure 2.2). 54,885 of those patients had more than one claim for metastatic disease. Of those with metastatic disease, 715 were treated with an ASTi (enzalutamide or abiraterone) than an alternative ASTi (ASTi-ASTi); 211 were treated with an ASTi then docetaxel or vice versa (ASTi-Doc); and 35 were treated with an ASTi and then cabazitaxel or vice versa (ASTi-Cab). 35 patients were treated with an ASTi or docetaxel in the first- and second-line setting, and then an alternative ASTi in the third line. 21 patients were treated with an ASTi or docetaxel in the first- and second-line setting, and then cabazitaxel in the third-line setting. Mean time between a diagnosis for prostate cancer and first captured line of therapy is 19.8 (24.9) months for the ASTi-ASTi cohort, 15.1 (20.6) months for the ASTi-Doc cohort, 17.5 (24.5) months for the ASTi-Cab cohort. For three lines of therapy cohorts, mean time between a diagnosis for prostate cancer and first captured line of therapy is 15.6 (16.9) months and 20.9 (28.4) months for the ASTi-Doc-ASTi and ASTi-Doc-Cab cohorts respectively (Table 2.1). 28 Patient Characteristics Mean age was 72.3 (8.5), 67.4 (9.1), and 67.1 (8.0) for the ASTi-ASTi, ASTi-Doc, and ASTi-Cab cohorts, respectively, and 67.6 (8.0) and 66.3 (7.5) for ASTi-Doc-ASTi and ASTi- Doc-Cab comparators (Table 2.1). Most subjects, regardless of treatment sequence resided in the Southeast. Patients treated with ASTi-Cab, ASTi-Doc-ASTi, and ASTi-Doc-Cab were more often covered with a commercial insurance option. Patients treated with an ASTi-Doc or ASTi- Cab were more likely to have started their first line of therapy in a later year than those who were treated with an ASTi-ASTi. Patients treated with ASTi-Doc-ASTi more often started their first line of therapy in an earlier year than patients treated with ASTi-Doc-Cab. Baseline Healthcare Resource Utilization and Economic Burden Among all 2L cohorts, there were no observed differences in baseline inpatient admissions, outpatient visits, emergency room visits, or hospice admissions (Table 2.2a). There was an observed difference in utilized other services, with ASTi-Cab and ASTi-Doc recording more other service visits than ASTi-ASTi. Among PPPM baseline cost burden, there were also no significant differences between each 2L cohort for each category examined, beyond total patient-borne medical costs. ASTi-Doc and ASTi-Cab patients experienced greater patient-borne costs medical costs than those in the ASTi-ASTi arm. Among 3L cohorts there were also no observed significant differences between medical service utilization or patient PPPM cost burden. 29 Treatment Cost Burden Unadjusted all-cause cost burden differed in comprehensive total medical costs, payer total medical costs, and patient total medical costs among patients treated in the two lines of therapy cohort (Table 2.3a). In each comprehensive, payer, or patient-borne total medical costs, ASTi-Doc and ASTi-Cab was higher than ASTi-ASTi. Same cost categories did not differ among patients in the three lines of therapy cohort. In the 2L cohort, patients differed in every category (Table 2.3b). ASTi-Doc and ASTi-Cab had higher comprehensive total medical expenses and payer total medical than ASTi-ASTi. ASTi-Cab was more expensive than ASTi- ASTi and ASTi-Doc for comprehensive pharmacy and payer pharmacy cost burden. ASTi-Cab has a lower patient-born total medical and pharmacy cost burden relative to ASTi-Doc and ASTi-ASTi. Patients treated with either ASTi-Doc-ASTi or ASTi-Doc-Cab did not have different associated medical costs, all-cause or cancer-related in all categories. Average Standard Cost and Patient-Borne Therapy Cost Standard costs for each therapy significantly differed, regardless of line of therapy. In a second line setting, average standard cost and patient-borne costs were higher for abiraterone and enzalutamide, relative to docetaxel and cabazitaxel (Table 2.4). This was also reflected in a third line setting. Cabazitaxel was the most expensive in terms of mean standard, with cabazitaxel representing a 10-fold increase on docetaxel standard cost in a 2L setting, and close to a 2-fold increase on abiraterone and enzalutamide in a 3L setting. However, docetaxel and cabazitaxel had a lower patient cost-burden than abiraterone and enzalutamide, regardless of 2L or 3L setting. 30 Clinical Outcomes per Treatment Regimen Unadjusted time to third line treatment for patients in the 2L cohort differed between comparators. Those treated with an ASTi-ASTi had the longest time before next treatment (373.1 days), whereas those treated with an ASTi-Doc had the shortest time before third line treatment (266.8 days; Table 2.5). Time to mortality and clinical outcomes examined for patients treated with an ASTi-Doc-ASTi or ASTi-Doc-Cab was censored due to data availability. Adjusted Outcomes per Cohort Controlling for key characteristics yield adjusted outcomes in line with the unadjusted. ASTi-Cab still tends to be more expensive than ASTi-ASTi and ASTi-Doc. Third line adjusted outcomes appear to be significantly skewed by an outlying observation and small sample size (<30 observations, See Appendix Table 2.1). Third line adjusted outcomes should be reevaluated during the next published data refresh by CDM. 2.4 Discussion Treatments for metastatic, CRPC are notoriously expensive, regardless of line of therapy or treatment sequaelae. 8,9,34,35 Several studies have characterized treatment-associated expenses. However, these studies either came from early samples of data, focused on specific sub-patient populations, or did not tease out economic expenses associated with specific lines of therapy considering updates in preferred treatment regimens for this patient population. 9,35–38 In this study we examined patient, payer, and comprehensive cost-burden by second or third line treatment options, assuming the following treatment sequalae: ASTi-ASTi, ASTi-Doc, ASTi- Cab, ASTi-Doc-ASTi, and ASTi-Doc-Cab. 31 Among all patients, cancer-related and all-cause economic burden was high from both the payer and patient perspective. Relative to ASTi-ASTi, ASTi-Doc and ASTi-Cab were associated with significantly higher costs, which may be in part due to medical costs brought on my chemotherapy. In a third line setting, costs again were high, however there was not a significant difference between ASTi-Doc-ASTi or ASTi-Doc-Cab arms. This may in part be due to limited sample size. Despite significant expense differences in the second line cohorts, out-of-pocket expenses borne by patients tend to be greater among oral regimens (ASTi’s) relative to intravenous-based regimens. Difference in medical and pharmacy benefits may play a large role in the patient-borne price difference. These findings are consistent with previously published research, that associated medical cost-burden, both medical and cancer related are substantial for this patient population. 12,39,40 Furthermore, the increase in PPPM medical expenses from 2 nd line therapies to 3 rd line therapies, also aligns with previous research that has found associated medical costs increase as disease progresses. Considering these captured costs, relative cost per treatment sequalae is critical for further economic evaluation of ASTi’s, docetaxel or cabazitaxel following an ASTi in a second- or third-line setting. In a second line setting, patients treated with cabazitaxel had greater comprehensive, total medical expenses and payer total medical expenses. However, those costs did not lead to greater financial toxicity among patients. Patient-borne costs during second-line therapy was greater among those treated with an alternative ASTi or docetaxel, as opposed to cabazitaxel. Pharmacy expenses were not significantly different between the three, second line comparators, suggesting increases in expenses can be largely attributed to greater healthcare resource utilization associated with downstream consequences tied to infusion-based regimens. 32 Since abiraterone and enzalutamide are dispensed as oral agents taken in a home-setting as opposed to an infusion, the increased patient cost responsibility despite lower overall cost, might be a critical consideration for insurers. Furthermore, evidence presented here regarding ASTi time to next treatment, used as a proxy for disease progression, may highlight the importance of persistence and adherence to abiraterone and enzalutamide, which could be negatively impacted by higher patient cost burden. This may be critical for payers to account for when determining patient cost obligation for hormonal-based regimens. These themes in general cost difference were also reflected when accounting for cancer-related costs, specifically, which may suggest the differences observed in a second line setting are largely driven by costs and resource utilization associated with cancer treatment or management. Despite evidence that chemotherapy-based regimens following an ASTi lead to higher overall cost burdens for payers, significant differences were not observed in a third line setting, evaluating ASTi-Doc-ASTi and ASTi-Doc-Cab. This could be due to patient disease severity. In the event of third line treatments for metastatic, CRPC, these patients are among the most severe, which may necessitate greater healthcare resource utilization and thus higher overall costs for both payers and patient. Among mCRPC patients and their providers, OOP costs likely represents the extent of the financial consideration for deciding between cabazitaxel and an ASTi. However, in the context of a large system in which healthy individuals also pay premiums to avoid the financial risk associated with becoming ill, these findings should also be discussed in the context of insurance value. With the CARD clinical trial, cabazitaxel demonstrated greater progression free survival and overall survival relative to an alternative ASTi among mCRPC patients, treated with an alternative ASTi and docetaxel. Now that a treatment option with greater survival benefit has 33 been identified, foreseeably more healthy individuals are willing to pay for an instrument to gain access to cabazitaxel and the ancillary medical needs associated with cabazitaxel if they need to be treated for later line mCRPC. That demand for cabazitaxel, in the event of becoming sick (not solely when sick), is indicative of “physical risk protection” among these patients. 41 However, these healthy individuals cannot be insured against the sickness. They can only be insured against the financial burden of management and treatment. So, in the context of total medical costs for cabazitaxel far exceeding treatment for an ASTi, there is a possibility that the physical risk protection exceeds that of the general financial risk borne by the total patient population seen by a large insurer, since cabazitaxel represents a clinical improvement, among a severely sick patient population, in which the incidence for this progressed disease is likely small. Economic evaluations expanding the implications of this research should take insurance value into account when considering whether cabazitaxel is cost-effective relative to an alternative ASTi as seen in this treatment decision problem. Given this critical dynamic between financial and physical risk protection determining value of insurance, this highlights the importance of benefit transparency for patients and providers. Despite worst clinical outcomes as captured in the CARD clinical trial, ASTi has a higher patient burden, which translates to a greater financial risk for higher physical risk, lowering the impact of patient insurance value. However, simultaneously, the per member per month impact from ASTi treatment for patients in this context is likely lower than for cabazitaxel treatment, which could in turn raise health member insurance value. Each of these are critical considerations when conducting future economic evaluations of cabazitaxel and an alternative ASTi for mCRPC patients treated with docetaxel and an alternative ASTi. 34 Treatments for metastatic, CRPC have recently come under increased scrutiny due to significant cost burden for patients and payers. 8,42 In this study, we find that this significantly expands beyond direct therapeutic cost to total medical expenses. Payer and patient costs impact on adherence and persistence should be further examined. Poor patient adherence on these therapies can impact treatment efficacy. 36,43,44 In this analysis, a correlation between patient- borne treatment costs and clinical outcomes was observed. Given these two elements, impact of patient-responsible costs on treatment adherence and persistence should be further examined. In the event costs have a large impact on treatment uptake, payers and healthcare decision-makers should heavily weigh that knowledge before suggesting treatment. Other future directions should include validating these findings through a qualitative survey eliciting cost-burden from the payer and patient perspective or secondary dataset. This study had several limitations. First, our analysis was completed using a retrospective claims database, which may limit the available clinical information, which would then limit how we stratify treatment groups by relevant risk groups. Second, healthcare resource utilization pertaining to outside of the hospital are likely not captured here, which will may lead to an underrepresentation of costs and outcomes contingent on treatment costs and economic burden related to different treatment sequelae. Third, classification of disease and treatment depend on ICD-9-CM and ICD-10-CM codes. There is a possibility of under- or over-coding claims when determining the clinical characteristics and economic burden associated with patients captured in this analysis. Fourth, CDM may not be generalizable to other patient populations of interest that are better captured in datasets dedicated to Medicare or publicly insured individuals. Of those included in this analysis, more than 70% are covered with Medicare, suggesting Medicare- centric dataset, like SEER-Medicare or a similar dataset may be more suitable for future 35 analyses. Fifth, time to next treatment is not a perfect analog for disease progression. Previous studies indicated that time to next treatment may even overestimate disease progression in some cases, suggesting patients captured in this study may have had longer disease free intervals. 21 Sixth, clinical outcomes among patients treated with a third line of therapy were limited in analysis due to poor sample size. Adjusted outcomes and costs associated with cabazitaxel among the third line cohorts (ASTi-Doc-ASTi and ASTi-Doc-Cab) may have skewed outcomes due to poor sample size and the existence of outliers. As these treatment sequalae become more popular in usage, and as more data is collected, this outcome should also be reexamined to further evaluate the findings explored in the CARD clinical trial. This study confirms previously published literature that treatment and care for patients with CRPC is expensive, both from the patient and payer perspective. 8,12,13,45 In light of this confirmatory analysis and expansion evaluating docetaxel and cabazitaxel in the context of abiraterone and enzalutamide, there are several future directions researchers should examine. First, researchers should examine the impact of patient-borne costs on docetaxel and cabazitaxel treatment. Previous work has indicated that high patient-cost burden has an impact on adherence to oral therapies, enzalutamide and abiraterone. 8 However, more research can be done to examine how the difference in pharmacy and medical benefits impact treatment adherence and persistence for patients that are receiving intravenous regimens. Second, this study focused on observed monotherapies, or instances in which only one of our four therapies of interest are the line of therapy. If an ASTi leads to cross-tolerance to the alternative ASTi, then patient outcomes during combination therapies should be stunted. Future researchers can expand the inclusion criteria and evaluate clinical outcomes from that perspective. Third, cabazitaxel as an alternative to an alternative ASTi, following docetaxel and ASTi use was only recently found to be an 36 effective alternative. 2 Analysts replicating or expanding this analysis should do so on more recent datasets to allow for a greater sample aggregation. Fourth, patient quality-of-life is a key decision attribute when deciding between a hormonal, oral-based regimen, such as abiraterone or enzalutamide, relative to an intravenous regimen, such as cabazitaxel or docetaxel. Future research should aim to better elucidate the impact and patient preference between hormonal- based, oral regimens and chemotherapeutic, and intravenous regimens. Finally, CRPC patients were primarily identified in this study by examining treatment regimens, due to a desire to not limit patient sample size by clinical markers that may be missing but are necessary for more popular identifying algorithms. Using a published CRPC identifying algorithm can help better identify patients with CRPC and evaluating the impact on patient-born costs on CRPC patients. 46 However, this may be contingent on the sample size available in future published datasets and analyses. 2.5 Conclusion Using CDM, over a ten-year time horizon, we found a significant increase in comprehensive and payer, total and cancer-related medical expenditures for metastatic, CRPC patients treated with an ASTi-Doc or ASTi-Cab in second line settings relative to those patients treated with an ASTi-ASTi regimen. Despite high overall costs, patient borne costs (copays + deductibles) were significantly less for the ASTi-Cab cohort. Differences in total and cancer-related expenses were not observed between ASTi-Doc-ASTi or ASTi-Doc-Cab. Further analyses should be aimed at untangling why these differences occur and if they impact clinical outcomes for patients’ burden with a large expense. 37 2.6 References 1. American Cancer Society | Cancer Facts & Statistics. American Cancer Society | Cancer Facts & Statistics. Accessed June 6, 2021. http://cancerstatisticscenter.cancer.org/ 2. de Wit R, de Bono J, Sternberg CN, et al. Cabazitaxel versus abiraterone or enzalutamide in metastatic prostate cancer. New England Journal of Medicine. 2019;381(26):2506-2518. 3. Siegel DA. Prostate Cancer Incidence and Survival, by Stage and Race/Ethnicity — United States, 2001–2017. MMWR Morb Mortal Wkly Rep. 2020;69. doi:10.15585/mmwr.mm6941a1 4. Tombal B, Lecouvet F. Modern Detection of Prostate Cancer’s Bone Metastasis: Is the Bone Scan Era Over? Advances in Urology. 2011;2012:e893193. doi:10.1155/2012/893193 5. Poon DMC, Wong KCW, Chan TW, et al. Survival Outcomes, Prostate-specific Antigen Response, and Tolerance in First and Later Lines of Enzalutamide Treatment for Metastatic Castration-resistant Prostate Cancer: A Real-World Experience in Hong Kong. Clinical Genitourinary Cancer. 2018;16(5):402-412.e1. doi:10.1016/j.clgc.2018.07.008 6. Lavaud P, Gravis G, Foulon S, et al. Anticancer Activity and Tolerance of Treatments Received Beyond Progression in Men Treated Upfront with Androgen Deprivation Therapy With or Without Docetaxel for Metastatic Castration-naïve Prostate Cancer in the GETUG- AFU 15 Phase 3 Trial. European Urology. 2018;73(5):696-703. doi:10.1016/j.eururo.2017.09.022 7. Pollard ME, Moskowitz AJ, Diefenbach MA, Hall SJ. Cost-effectiveness analysis of treatments for metastatic castration resistant prostate cancer. Asian J Urol. 2017;4(1):37-43. doi:10.1016/j.ajur.2016.11.005 8. Caram MEV, Oerline MK, Dusetzina S, et al. Adherence and out-of-pocket costs among Medicare beneficiaries who are prescribed oral targeted therapies for advanced prostate cancer. Cancer. 2020;126(23):5050-5059. doi:10.1002/cncr.33176 9. Rana D, Geue C, Baillie K, et al. Healthcare Costs for Metastatic Castration-Resistant Prostate Cancer Patients Treated with Abiraterone or Enzalutamide. PharmacoEconomics Open. 2022;6(2):303-313. doi:10.1007/s41669-021-00307-1 10. Mehra M, Wu Y, Dhawan R. Healthcare resource use in advanced prostate cancer patients treated with docetaxel. J Med Econ. 2012;15(5):836-843. doi:10.3111/13696998.2012.681718 11. Schultz NM, Flanders SC, Wilson S, et al. Treatment Duration, Healthcare Resource Utilization, and Costs Among Chemotherapy-Naïve Patients with Metastatic Castration- Resistant Prostate Cancer Treated with Enzalutamide or Abiraterone Acetate: A Retrospective Claims Analysis. Adv Ther. 2018;35(10):1639-1655. doi:10.1007/s12325- 018-0774-1 38 12. Appukkuttan S, Tangirala K, Babajanyan S, Wen L, Simmons S, Shore N. A Retrospective Claims Analysis of Advanced Prostate Cancer Costs and Resource Use. PharmacoEconomics Open. Published online October 22, 2019. doi:10.1007/s41669-019- 00185-8 13. Valderrama A, Tangirala K, Babajanyan S, et al. Treatment, healthcare resource utilization, and costs associated with non-metastatic and metastatic castration-resistant prostate cancer: A claims analysis. JCO. 2017;35(15_suppl):e18341-e18341. doi:10.1200/JCO.2017.35.15_suppl.e18341 14. Wen L, Yao J, Valderrama A. Evaluation of Treatment Patterns and Costs in Patients with Prostate Cancer and Bone Metastases. JMCP. 2019;25(3-b Suppl):S1-S11. doi:10.18553/jmcp.2019.25.3-b.s1 15. Johnston SS, Morton JM, Kalsekar I, Ammann EM, Hsiao CW, Reps J. Using Machine Learning Applied to Real-World Healthcare Data for Predictive Analytics: An Applied Example in Bariatric Surgery. Value in Health. 2019;22(5):580-586. doi:10.1016/j.jval.2019.01.011 16. Chandran U, Reps J, Stang PE, Ryan PB. Inferring disease severity in rheumatoid arthritis using predictive modeling in administrative claims databases. PLOS ONE. 2019;14(12):e0226255. doi:10.1371/journal.pone.0226255 17. Suchard MA, Schuemie MJ, Krumholz HM, et al. Comprehensive comparative effectiveness and safety of first-line antihypertensive drug classes: a systematic, multinational, large-scale analysis. The Lancet. 2019;394(10211):1816-1826. doi:10.1016/S0140-6736(19)32317-7 18. Jin MC, Sussman ES, Feng AY, et al. Hemorrhage risk of direct oral anticoagulants in real- world venous thromboembolism patients. Thrombosis Research. 2021;204:126-133. doi:10.1016/j.thromres.2021.06.015 19. Ye Y, Manne S, Bennett D. Identifying Patients With Inflammatory Bowel Diseases in an Administrative Health Claims Database: Do Algorithms Generate Similar Findings? INQUIRY. 2019;56:0046958019887816. doi:10.1177/0046958019887816 20. U.S. Bureau of Labor Statistics. Consumer Price Index for All Urban Consumers: Medical Care in U.S. City Average. FRED, Federal Reserve Bank of St. Louis. Published January 1, 1947. Accessed March 10, 2022. https://fred.stlouisfed.org/series/CPIMEDSL 21. Walker MS, Herms L, Miller PJE. Performance of time to discontinuation and time to next treatment as proxy measures of progression-free survival, overall and by treatment group. JCO. 2020;38(15_suppl):e19135-e19135. doi:10.1200/JCO.2020.38.15_suppl.e19135 22. Beachler DC, Lamy FX, Russo L, et al. A real-world study on characteristics, treatments and outcomes in US patients with advanced stage ovarian cancer. Journal of Ovarian Research. 2020;13(1):101. doi:10.1186/s13048-020-00691-y 39 23. Huang Q, Borra S, Li J, et al. Time to Next Treatment, Health Care Resource Utilization, and Costs Associated with Ibrutinib Use Among U.S. Veterans with Chronic Lymphocytic Leukemia/Small Lymphocytic Lymphoma: A Real-World Retrospective Analysis. JMCP. 2020;26(10):1266-1275. doi:10.18553/jmcp.2020.20095 24. Jalbert JJ, Wu N, Chen CI, Ambati S, Ge W, Arnason JE. Real-World Treatment Patterns After CD19-Directed CAR T Cell Therapy Among Patients with Diffuse Large B Cell Lymphoma. Adv Ther. Published online April 9, 2022. doi:10.1007/s12325-022-02087-4 25. Mazumdar M, Lin JYJ, Zhang W, et al. Comparison of statistical and machine learning models for healthcare cost data: a simulation study motivated by Oncology Care Model (OCM) data. BMC Health Serv Res. 2020;20:350. doi:10.1186/s12913-020-05148-y 26. Barber J, Thompson S. Multiple regression of cost data: use of generalised linear models. J Health Serv Res Policy. 2004;9(4):197-204. doi:10.1258/1355819042250249 27. Ng VKY, Cribbie RA. Using the Gamma Generalized Linear Model for Modeling Continuous, Skewed and Heteroscedastic Outcomes in Psychology. Current Psychology. 2017;36(2):225-236. doi:10.1007/s12144-015-9404-0 28. Dodd S, Bassi A, Bodger K, Williamson P. A comparison of multivariable regression models to analyse cost data. Journal of Evaluation in Clinical Practice. 2006;12(1):76-86. doi:10.1111/j.1365-2753.2006.00610.x 29. Deb P, Norton EC. Modeling Health Care Expenditures and Use. Annual Review of Public Health. 2018;39(1):489-505. doi:10.1146/annurev-publhealth-040617-013517 30. Bonomi L, Jiang X, Ohno-Machado L. Protecting patient privacy in survival analyses. J Am Med Inform Assoc. 2019;27(3):366-375. doi:10.1093/jamia/ocz195 31. Klann JG, Joss M, Shirali R, et al. The Ad-Hoc Uncertainty Principle of Patient Privacy. AMIA Jt Summits Transl Sci Proc. 2018;2018:132-138. 32. Hosmer DW, Lemeshow S, May S. Applied Survival Analysis: Regression Modeling of Time to Event Data, Second Edition. Wiley; 2008. 33. Dupont WD, Plummer WD. Power and sample size calculations for studies involving linear regression. Control Clin Trials. 1998;19(6):589-601. doi:10.1016/s0197-2456(98)00037-3 34. Shah A, Shah R, Kebede N, et al. Real-world incidence and burden of adverse events among non-metastatic prostate cancer patients treated with secondary hormonal therapies following androgen deprivation therapy. Journal of Medical Economics. 2020;23(4):330- 346. doi:10.1080/13696998.2019.1705313 35. Alemayehu B, Buysman E, Parry D, Becker L, Nathan F. Economic burden and healthcare utilization associated with castration-resistant prostate cancer in a commercial and Medicare Advantage US patient population. J Med Econ. 2010;13(2):351-361. doi:10.3111/13696998.2010.491435 40 36. Hwang SS, Chang VT, Alejandro Y, et al. Study of hormone refractory prostate cancer: hospital care and palliative care resource use at a VA medical center. Cancer Invest. 2004;22(6):849-857. doi:10.1081/cnv-200039643 37. Crawford ED, Black L, Eaddy M, Kruep EJ. A retrospective analysis illustrating the substantial clinical and economic burden of prostate cancer. Prostate Cancer Prostatic Dis. 2010;13(2):162-167. doi:10.1038/pcan.2009.63 38. Ha YS, Kim SY, Chung JI, et al. Trends in End-of-Life Resource Utilization and Costs among Prostate Cancer Patients from 2006 to 2015: A Nationwide Population-Based Study. World J Mens Health. 2021;39(1):158-167. doi:10.5534/wjmh.200113 39. Roehrborn CG, Black LK. The economic burden of prostate cancer: ECONOMIC BURDEN OF PROSTATE CANCER. BJU International. 2011;108(6):806-813. doi:10.1111/j.1464- 410X.2011.10365.x 40. Seal B, Sullivan SD, Ramsey SD, et al. Comparing Hospital-Based Resource Utilization and Costs for Prostate Cancer Patients With and Without Bone Metastases. Appl Health Econ Health Policy. 2014;12(5):547-557. doi:10.1007/s40258-014-0101-1 41. Lakdawalla D, Malani A, Reif J. The insurance value of medical innovation. Journal of Public Economics. 2017;145:94-102. doi:10.1016/j.jpubeco.2016.11.012 42. $150K Prostate Cancer Drug Draws New Attention to “March-In” Rights. Published April 6, 2022. Accessed April 12, 2022. https://www.medpagetoday.com/special- reports/exclusives/98091 43. Banna GL, Urzia V, Benanti C, et al. Adherence to abiraterone or enzalutamide in elderly metastatic castration-resistant prostate cancer. Support Care Cancer. 2020;28(10):4687- 4695. doi:10.1007/s00520-020-05311-5 44. Iacorossi L, Gambalunga F, De Domenico R, Serra V, Marzo C, Carlini P. Qualitative study of patients with metastatic prostate cancer to adherence of hormone therapy. European Journal of Oncology Nursing. 2019;38:8-12. doi:10.1016/j.ejon.2018.11.004 45. Tangirala K, Appukkuttan S, Simmons S. Costs and Healthcare Resource Utilization Associated with Hospital Admissions of Patients with Metastatic or Nonmetastatic Prostate Cancer. Am Health Drug Benefits. 2019;12(6):306-312. 46. Freedland SJ, Ke X, Lafeuille MH, et al. Identification of patients with metastatic castration-sensitive or metastatic castration-resistant prostate cancer using administrative health claims and laboratory data. Curr Med Res Opin. 2021;37(4):609-622. doi:10.1080/03007995.2021.1879753 41 Chapter 2 Tables Table 2.1 Baseline Patient Demographics 2L 3L VARIABLES ASTi-ASTi ASTi-Doc ASTi-Cab ASTi-Doc-ASTi ASTi-Doc-Cab Obs 715 211 35 35 21 Age - - - - - Mean 72.3 67.4 67.1 67.6 66.3 Min 42 36 50 50 57 Max 90 88 82 82 85 SD 8.5 9.1 8.0 8.0 7.5 Median 73 67 67 68 64 Age Categories - - - - - 35 - 50 0.01 0.03 0.03 0.03 0.00 50 - 65 0.19 0.41 0.37 0.37 0.62 65 - 80 0.62 0.47 0.57 0.54 0.33 >80 0.18 0.09 0.03 0.06 0.05 Gender - - - - - Male 1.00 1.00 1.00 1.00 1.00 Region - - - - - Northeast 0.11 0.09 0.06 0.06 0.10 Midwest 0.20 0.24 0.20 0.20 0.33 Southeast 0.36 0.52 0.43 0.54 0.52 West 0.33 0.16 0.31 0.20 0.05 Race - - - - - Asian 0.02 0.03 0.00 0.03 0.00 Black 0.14 0.18 0.06 0.14 0.33 Hispanic 0.09 0.07 0.09 0.03 0.05 White 0.67 0.60 0.71 0.74 0.43 Unknown 0.08 0.12 0.14 0.06 0.19 Plan Type - - - - - Commercial 0.23 0.44 0.57 0.51 0.52 Medicare 0.77 0.56 0.43 0.49 0.48 Time to first treatment for metastatic disease categories - - - - - 0 - 6 months 0.39 0.44 0.46 0.40 0.48 7 - 12 months 0.10 0.16 0.11 0.06 0.10 1 - 2 years 0.16 0.16 0.26 0.29 0.10 2 - 3 years 0.13 0.11 0.03 0.17 0.10 42 3 - 4 years 0.07 0.04 0.03 0.03 0.10 4 - 5 years 0.06 0.03 0.00 0.03 0.00 > 5 years 0.09 0.06 0.11 0.03 0.14 Year Starting First LOT - - - - - 2010 0.00 0.03 0.06 0.06 0.05 2011 0.03 0.06 0.23 0.11 0.05 2012 0.08 0.06 0.03 0.20 0.05 2013 0.11 0.09 0.03 0.09 0.24 2014 0.17 0.07 0.00 0.09 0.05 2015 0.15 0.10 0.11 0.11 0.14 2016 0.15 0.11 0.14 0.23 0.05 2017 0.17 0.15 0.09 0.00 0.10 2018 0.10 0.18 0.17 0.06 0.19 2019 0.01 0.10 0.06 0.06 0.10 2020 0.01 0.03 0.09 0.00 0.00 Time to first treatment for metastatic disease following PC Diagnosis (Months) - - - - - Mean 19.8 15.1 17.5 15.6 20.9 SD 24.9 20.6 24.5 16.9 28.4 Time on 2L or 3L Therapy (Months) Mean 9.47 7.56 8.38 7.60 3.78 SD 11.60 7.47 8.36 8.60 1.65 CCI - - - - - Mean 2.04 .99 .23 .06 .14 SD 5.33 3.62 .55 .34 .48 Notes: 2L corresponds to cohorts in the 2L comparison group. 3L corresponds to cohorts in the 3L comparison group. ASTi = Androgen-signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel; Obs = Observations; SD = Standard Deviation; LOT = Line of Therapy; PC = Prostate Cancer; CCI = Charlson Comorbidity Index 43 Table 2.2a Baseline Healthcare Resource Utilization 2L 3L VARIABLES ASTi-ASTi ASTi-Doc ASTi-Cab ASTi-Doc-ASTi ASTi-Doc-Cab Observations 717 213 35 35 23 Inpatient Admissions Mean 1.6 2.8 2.16 0.61 1.37 SD 5.39 10.94 5.94 3.59 4.49 p-value 0.10 0.52 Outpatient Visits Mean 10.66 11.43 5.1 8.87 20.5 SD 24.29 30.45 10.36 14.57 73.47 p-value 0.40 0.48 Emergency Room Visits Mean 0.61 0.82 0.37 0.70 0.25 SD 4.51 5.39 1.42 2.42 1 p-value 0.80 0.34 Hospice Admissions Mean 0 0 0 0 0 SD 0.02 0 0 0 0 p-value 0.83 >0.99 Other Use Mean 10.14 15.59 23.73 12.86 12.36 SD 12.96 25.31 34.56 13.67 10.27 p-value 0.00* 0.88 Notes: Baseline healthcare resource utilization contains counts of patient visits or admissions to the above medical facilities. Counts are reported as encounters (admissions or visits) per patient per month (PPPM). ANOVA was used to examine differences in the second line cohorts and t- tests were used to examine differences in the third line cohorts. Significant p-values, < 0.05, are marked with a *. 2L corresponds to cohorts in the 2L comparison group. 3L corresponds to cohorts in the 3L comparison group. ASTi = Androgen-signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel; SD = Standard Deviation 44 Table 2.2b Baseline Cost Burden 2L 3L VARIABLES ASTi-ASTi ASTi-Doc ASTi-Cab ASTi-Doc-ASTi ASTi-Doc-Cab Comprehensive - Total Medical Mean $21,698 $27,466 $27,436 $16,990 $49,152 SD $37,067 $57,277 $37,749 $18,222 $111,609 p-value 0.20 0.20 Comprehensive Pharmacy Mean $11,966 $13,370 $18,084 $11,377 $16,243 SD $15,975 $14,319 $18,922 $9,193 $20,131 p-value 0.05 0.31 Payer - Total Medical Mean $21,321 $26,777 $26,895 $16,399 $48,445 SD $36,750 $57,000 $37,417 $17,829 $111,589 p-value 0.23 0.21 Payer - Pharmacy Mean $11,643 $13,062 $17,777 $11,057 $15,889 SD $15,930 $14,305 $18,796 $9,196 $20,093 p-value 0.05 0.31 Patient Total Medical Mean $377 $689 $541 $591 $707 SD $618 $839 $588 $676 $1,059 p-value 0.00* 0.66 Patient Pharmacy Mean $323 $308 $307 $319 $354 SD $304 $332 $334 $368 $533 p-value 0.80 0.79 Notes: Baseline economic burden contains dollarized amounts of comprehensive cost burden (patient + payer), payer economic burden, and patient cost burden. Amounts are scaled to per patient per month (PPPM). ANOVA was used to examine differences in the second line cohorts and t-tests were used to examine differences in the third line cohorts. Significant p-values, < 0.05, are marked with a *. 2L corresponds to cohorts in the 2L comparison group. 3L corresponds to cohorts in the 3L comparison group. ASTi = Androgen-signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel; SD = Standard Deviation 45 Table 2.3a All-Cause Cost Burden 2L 3L VARIABLES ASTi-ASTi ASTi-Doc ASTi-Cab ASTi-Doc-ASTi ASTi-Doc-Cab Comprehensive - Total Medical Mean $35,747 $56,854 $53,163 $211,248 $185,188 SD $55,178 $120,198 $32,387 $239,927 $99,037 p-value 0.04* 0.69 Comprehensive Pharmacy Mean $24,885 $29,818 $33,204 $120,790 $102,746 SD $23,744 $39,166 $15,333 $154,739 $57,657 p-value 0.18 0.66 Payer - Total Medical Mean $35,263 $56,014 $52,955 $208,437 $183,791 SD $54,882 $119,297 $32,319 $237,127 $99,016 p-value 0.04* 0.70 Payer - Pharmacy Mean $24,551 $29,523 $33,020 $119,311 $102,129 SD $23,704 $39,067 $15,414 $152,864 $57,516 p-value 0.17 0.67 Patient Total Medical Mean $484 $840 $208 $2,811 $1,396 SD $886 $1,264 $372 $3,459 $2,000 p-value 0.00* 0.18 Patient Pharmacy Mean $335 $294 $185 $1,480 $616 SD $438 $290 $210 $2,069 $790 p-value 0.25 0.13 Notes: Line of therapy all-cause cost burden contains dollarized amounts of comprehensive cost burden (patient + payer), payer cost burden, and patient cost burden. All medical and pharmacy costs are included. For 2L therapies, costs are captured during the second line of therapy start and end date. For 3L therapies, costs are captured during the third line of therapy start and end date. Amounts are scaled to per patient per month (PPPM). ANOVA was used to examine differences in the second line cohorts and t-tests were used to examine differences in the third line cohorts. Significant p-values, < 0.05, are marked with a *. 2L corresponds to cohorts in the 2L comparison group. 3L corresponds to cohorts in the 3L comparison group. ASTi = Androgen- signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel; SD = Standard Deviation 46 Table 2.3b Cancer-Related Cost Burden 2L 3L VARIABLES ASTi-ASTi ASTi-Doc ASTi-Cab ASTi-Doc-ASTi ASTi-Doc-Cab Comprehensive - Total Medical Mean $19,846 $31,709 $32,240 $109,481 $108,048 SD $28,474 $60,782 $19,881 $122,551 $57,717 p-value 0.02* 0.97 Comprehensive Pharmacy Mean $8,177 $8,501 $16,955 $27,173 $33,976 SD $4,410 $6,961 $13,622 $33,518 $23,166 p-value 0.00* 0.53 Payer - Total Medical Mean $19,575 $31,244 $32,096 $107,942 $107,166 SD $28,375 $60,238 $19,811 $121,055 $57,553 p-value 0.02* 0.98 Payer – Pharmacy Mean $7,968 $8,380 $16,854 $26,676 $33,815 SD $4,405 $6,949 $13,672 $33,696 $23,186 p-value 0.00* 0.52 Patient Total Medical Mean $271 $465 $144 $1,538 $881 SD $439 $687 $279 $1,792 $1,275 p-value 0.00* 0.27 Patient Pharmacy Mean $209 $121 $101 $497 $161 SD $310 $196 $156 $638 $334 p-value 0.01* 0.08 Notes: Line of therapy cancer-related cost burden contains dollarized amounts of comprehensive cost burden (patient + payer), payer cost burden, and patient cost burden. All medical and pharmacy costs associated with a patient’s cancer is included here. For 2L therapies, costs are captured during the second line of therapy start and end date. For 3L therapies, costs are captured during the third line of therapy start and end date. Amounts are scaled to per patient per month (PPPM). ANOVA was used to examine differences in the second line cohorts and t-tests were used to examine differences in the third line cohorts. Significant p-values, < 0.05, are marked with a *. 2L corresponds to cohorts in the 2L comparison group. 3L corresponds to cohorts in the 3L comparison group. ASTi = Androgen-signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel; SD = Standard Deviation; SD = Standard Deviation; ANOVA = Analysis of variances. 47 Table 2.4 Average Standard Cost and Patient-Borne Therapy Cost 2L 3L VARIABLES Abiraterone Enzalutamide Docetaxel Cabazitaxel Abiraterone Enzalutamide Cabazitaxel Standard Cost Mean $8,416 $9,331 $1,543 $14,921 $7,695 $11,643 $18,396 SD $4,625 $5,816 $1,179 $11,033 $3,018 $16,550 $16,430 p-value 0.00* 0.00* Patient Cost Mean $192.28 $261.14 $13.22 $19.61 $249.10 $226.62 $0 SD $313.87 $349.70 $79.40 $109.19 $276.27 $281.42 $0 p-value 0.01* 0.00* Notes: Standard cost and patient-borne costs were captured per drug of interest (abiraterone, enzalutamide, docetaxel, and cabazitaxel) per line of therapy. Presented is the average per patient per month costs. ANOVA was used to examine differences in the second line cohorts and t-tests were used to examine differences in the third line cohorts. Significant p-values, < 0.05, are marked with a *. 2L corresponds to cohorts in the 2L comparison group. 3L corresponds to cohorts in the 3L comparison group. ASTi = Androgen-signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel; SD = Standard Deviation; ANOVA = Analysis of variances. 48 Table 2.5 Clinical Outcomes per Treatment Regimen 2L VARIABLES ASTi-ASTi ASTi-Doc ASTi-Cab Time to Next Treatment (Days), Line 2 -> Line 3 Mean 373.1 266.8 321.4 SD 411.5 230.8 307.9 p-value 0.00* Notes: Time to next treatment, from second line of therapy to third line was captured for 2L therapies sequences. Third line to fourth line among 3L therapies was also captured but removed from this analysis due to censoring, due to small sample. Time to mortality for 2L and 3L therapies following end of either 2 nd or 3 rd line of therapy was also censored due to small patient count and availability, particularly due to cabazitaxel. ANOVA was used to examine differences in the second line cohorts. Significant p-values, < 0.05, are marked with a *. 2L corresponds to cohorts in the 2L comparison group. ASTi = Androgen-signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel; SD = Standard Deviation; ANOVA = Analysis of variances. 49 Table 2.6 Adjusted Costs per Treatment Regimen 2L 2L VARIABLES ASTi-ASTi ASTi-Doc ASTi-Cab All-Cause Costs Comprehensive Medical Mean $40,110 $74,073 $95,185 SD $22,081 $49,804 $51,379 Pharmacy Mean $30,294 $36,830 $44,625 SD $10,376 $14,178 $15,944 Patient Medical Mean $569 $1,074 $164 SD $318 $653 $101 Pharmacy Mean $418 $354 $181 SD $222 $208 $92 Cancer-Related Costs Comprehensive Medical Mean $22,931 $41,774 $57,141 SD $11,188 $24,144 $28,195 Pharmacy Mean $10,128 $10,452 $19,337 SD $2,419 $2,443 $4,894 Patient Medical Mean $325 $609 $105 SD $165 $352 $71 Pharmacy Mean $304 $158 $114 SD $333 $235 $115 Clinical Outcome Time to Next Treatment (L2 - L3) Mean 373.16 241.41 293.46 SD 243.58 132.77 159.81 Notes: Adjusted predicted costs were generated using a generalized linear model controlling for a vector of covariates. 2L corresponds to cohorts in the 2L comparison group. ASTi = Androgen- signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel; 50 Chapter 2 Figures Figure 2.1. Study Schematic Notes: Study schematic with overview of study design. 51 Figure 2.2 Patient Disposition Notes: Patient disposition per comparator arm were generated using the inclusion and exclusion criteria outline here and in this article. 52 Figure 2.3 Unadjusted Time to Next Treatment Kaplan-Meier Estimates Notes: Unadjusted Kaplan-Meier progression figures were generated using time to next treatment from line 2 to line 3 to examine impact of 2L treatment sequences on disease progression. ASTi = Androgen-signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel 53 2.7 Appendix Appendix Note 1. Future Analyses and Identifying Patients with Metastatic, Castrate-Resistant Prostate Cancer (Abstracted from Freedland et al., 2021 for future reference) Identification of patients with metastatic, castrate-resistant prostate cancer (CRPC) can rely on capturing several key pieces. First patients need to be identified as having undergone a medical or surgical castration. Medical castration was defined as the following. Continuous androgen deprivation treatment agents (such as leuprolide, triptorelin, goserelin, histrelin, and degarelix) for at least 90 days with no gap in treatment for more than 30 consecutive days. Episodes of androgen deprivation therapy treatment can be constructed based on observed claims for the aforementioned agents and assumed days of supply. The first episode used to establish androgen deprivation therapy was constructed based on the first captured episode started on the date of the first claim for an androgen deprivation therapy agent, after the first observed diagnosis for prostate cancer. Continuous androgen deprivation therapy treatment was considered ended after a gap of more than 30 days. Patients who did not receive a medical castration may have received a surgical castration. Surgical castration was identified by observing claims for bilateral orchiectomy. Bilateral orchiectomy was identified as a procedure for one bilateral orchiectomy or two unilateral orchiectomies (including one in right side and one in left side). If the two unilateral orchiectomies were performed on separate days, the date of the surgical castration was the date of the second unilateral orchiectomy. If available, data prior to the study period (eg before January 1, 2014) were used to identify patients with orchiectomy. Following identification of medical and surgical castration, patients were observed for having evidence of castration resistance. Five criteria were used to identify patients with 54 castration resistance. 1) The ICD-10-CM hormone resistance code and 2) and 3) rising prostate- specific antigen levels after either medical or surgical castrations, or 4) and 5) disease progression to metastatic disease (for patients without metastases prior to castration) after medical or surgical castration were also considered as indicating patients that had metastatic, CRPC. Patients identified as having metastatic CRPC were identified using the algorithm (See Table 6 of Freedland et al., 2021). This algorithm relies on all select mCRPC patients as having metastatic disease and evidence of castration resistance, with no subsequent ICD or HCPCs codes identifying the patient as having a subsequent diagnosis of hormone sensitivity. Drug use exclusively was not used to identify patients with metastatic, CRPC for the purpose of not misidentifying patients with metastatic, castrate-sensitive prostate cancer who may have received an off-label therapy. First observed metastatic disease diagnosis and first evidence of castration resistance was used as the index date of interest, triggering the beginning of identifying specific lines of therapy and their associated costs. Metastatic, CRPC patients were also required to be at least 18 years of age, have at least 6 months of continuous enrollment prior to the index date, and have at least 30 days of follow-up post index date. Baseline period was defined as 6 months prior to index date. Follow-up period was considered that from index date to the earliest among end of data availability, end of insurance coverage or death. This can be an approach for identifying patients with advanced metastatic, CRPC. However, due to limited number of metastatic CRPC patients treated with our treatment sequalae of interest, this identifying algorithm was not applied for the purpose of avoiding culling 55 metastatic, castrate-resistant cancer patients who were likely treated with our treatments of interest but did not have the necessary identifying features laid out in this algorithm. 56 Appendix Table 2.1. Adjusted Outcomes among Third Line Patients VARIABLES ASTi-Doc-ASTi ASTi-Doc-Cab All-Cause Costs Comprehensive Medical Mean $487,457 $519,046 SD $830,591 $451,579 Pharmacy Mean $267,074 $298,515 SD $298,563 $269,465 Patient Medical Mean $226,915 $788,766,477 SD $834,939 $3,471,248,223 Pharmacy Mean $7,505 $2,395 SD $25,268 $2,396 Cancer-Related Costs Comprehensive Medical Mean $238,519 $300,495 SD $308,531 $258,504 Pharmacy Mean $70,564 $88,837 SD $50,417 $46,415 Patient Medical Mean $185,052 $782,467,852 SD $681,875 $3,529,406,002 Pharmacy Mean $345,945,106,290,841 $181,620 SD $2,046,634,543,377,220 $633,999 Notes: Adjusted predicted costs were generated using a generalized linear model controlling for a vector of covariates. Adjusted outcomes may be skewed by small sample size and outlier costs. ASTi = Androgen-signaling, targeted inhibitor; Doc = Docetaxel; Cab = Cabazitaxel. 57 Chapter 3: Higher Order Individual Utility Risk Parameters Associated with Health, among a United States Representative Sample Samuel A. Crawford, Jason N. Doctor, William V. Padula, & Darius N. Lakdawalla Abstract Growing innovations in cost-effectiveness analysis require the elicitation and characterization of individual utility functions to determine higher order risk parameters, such as relative risk aversion, relative prudence, and relative temperance. This study uses a series of risk apportionment tasks to characterize higher order risk parameters over several treatment conditions, in the gain domain of treatment decision-making. This survey was administered among a United States’ representative sample, in the Understanding America Study. Survey responses were then used to approximate certainty equivalents per respondent. Utility parameters were estimated using non-linear least squares using constant relative risk aversion, expo-power, and hyperbolic absolute risk aversion utility functions. Utility parameters were then used to calculate higher order risk parameters, relative risk aversion, relative prudence, and relative temperance. Findings suggest that in response to treatment gains, United States’ respondents overall exhibit relative risk aversion and relative prudence. However, the degree of relative risk aversion and relative prudence is less than what is seen with cash gains in similar treatment gambles. 58 3.1 Introduction Healthcare decision-making is inherently risky. Treatment outcomes are almost always heterogenous and vary widely depending on the disease, patient, caregivers, setting, and treatment. 1 Treatment and patient outcome uncertainty also play a pivotal role in the demand for health insurance, but can be overlooked in value assessments of new therapies. 2,3 As cost- effectiveness analysis frameworks continue to evolve, inclusion of attributes that patients consider important and their specific valuation of those attributes, such as treatment and outcome uncertainty, is critical. 4–6 This study aims to characterize patient higher order risk parameters, such as relative risk aversion, relative prudence, and relative temperance for potential inclusion in innovative health economic evaluations of new interventions of therapies. 3,7,8 Historically, analysis of risk attitudes has focused on characterizing higher order risk parameters and behavior among financial or monetary based decision-making. 9–11 These studies successfully built off previous literature by not only demonstrating risk-averse or risk-seeking behavior among individuals responding to financial-based lotteries, but also estimating the actual degree of the higher order risk parameters, such as relative risk aversion and prudence. Several studies have aimed to elicit risk preferences under the assumptions of expected utility theory and prospect theory, for month monetary and health outcomes. 12–16 The implications of these findings also have direct applications to several important individual behaviors in economics, such as bargaining, sustainable development, personal investment, and legal compliance. Like the importance of higher order risk parameters over financial outcomes, there is growing interest and use-cases for estimating these risk parameters over health-related decision- making, such as treatments. Characterizing these parameters first allows us to test if the canonical quality-adjusted life year (QALY) model for economic evaluation best represents 59 individual preferences and properly values interventions and new therapies. Typical cost- effectiveness analysis assumes the QALY as a vital part of the incremental cost-effectiveness ratio without dynamically allowing for relative risk aversion over health-related quality-of- life. 3,7,8,17 The implications of this assumption suggest that health-care decision makers, patients, and society-at-large value interventions of similar expected benefit as the same, regardless of underlying health of patient. However, there is considerable real-world evidence suggesting decision-makers, patients, and the public do value interventions for less healthy individuals at a higher level than for healthier patients. 18–21 Laboratory-setting evidence also suggests risk- averse, prudent and temperate preferences need to be included in economic evaluations of interventions and treatments. 22–24 Beyond specific applications to validating popular cost- effectiveness analysis frameworks and for populating critical parameters in new, innovative frameworks, there have also been calls to estimate and elucidate relative risk aversion, relative prudence, and relative temperance over health-related conditions or quality-of-life, given the potential impact on patient decision-making as a whole. 25–28 Several studies have evaluated the impact of risk attitudes on health-related decision-making. Early research found higher order risk parameters, such as relative prudence, had an impact on a patient’s ideal level of prevention for a health risk, as well as the impact of higher order risk parameters on attitudes toward treatment decisions that could impact life expectancy with the presence of comorbidities. 25,27,29–31 Other studies have also evaluated the cross-correlation of higher order risk parameters associated with income and health, demonstrating degrees of investment necessary for acute and tertiary care among patients with and without comorbidities. 28,29 60 Research has also characterized relative risk aversion and relative prudence over health domains, however they have been limited in their estimations of relative risk aversion, relative risk prudence and relative temperance. 29,30,32 Despite identifying the presence and sign of relative risk aversion, relative prudence and relative temperance in laboratory settings, previous research has not been able to extrapolate risk apportionment tasks to identify the shape of individual utility functions and the necessary derivatives need to estimate relative risk aversion, relative prudence and relative temperance. Furthermore, studies identifying the presence and sign of higher order risk preferences have primarily been conducted in a lab setting or with young samples, which may limit generalizability to general applications of these findings. 22,29,30 This study utilizes risk apportionment tasks to characterize higher order risk estimates of a representative, general United States (U.S.) sample, over several treatment conditions related to health and a monetary outcome to anchor against studies conducting similar research but in a financial setting. Prevalence and degree of relative risk aversion and relative prudence is examined among a non-student, United States’ representative sample using a series of preference questions pertaining to a hypothetical treatment selection. Furthermore, we characterize responses contingent on using different treatment conditions which may provide insight as to how higher order risk parameters change depending on the treatment or baseline condition. Finally, using answers to our risk apportionment tasks in this survey, we were able to elicit individual utility functions regarding health-related quality of life, which then allows us to estimate higher order risk parameters among a U.S. representative sample. Our results suggest that U.S.-based respondents demonstrate relative risk aversion and relative prudence across treatment conditions and general health status as captured by “health score.”. However, when respondents answer relative risk aversion and relative prudence tasks 61 framed with more serious health conditions, responses tend to be more risk neutral and imprudent. When answering tasks framed as financial-based gambles, respondents appeared to be more risk averse and prudent relative to health conditions. This finding provides an important anchor for contextualizing treatment risk preference estimates relative to previously published risk estimates evaluated among tasks framed as financial-based lotteries. These findings confirm the importance of new cost-effectiveness analysis frameworks that account for risk averse and prudent preferences in treatment decision-making. Furthermore, the estimates of higher order risk parameters can now be used in innovative cost-effectiveness analysis frameworks that take these perspectives into account. 3.2 Methods and Study Design Background and Elicitation Method Eliciting higher order risk parameters (relative risk aversion, relative risk prudence, and relative risk temperance) over health-related treatment decision making required question formation based on designs implemented in studies characterizing higher order risk parameters over decision making related to financial-based lotteries. 9 Relative risk aversion is defined mechanistically as the following 33,34 : 𝐸 [ 𝑢 ( 𝑋 ) ] < 𝑢 ( 𝐸 ( 𝑋 ) ) Eq. 3.1 where the expected utility of an argument (health gain or financial gain in the context of this paper) is less than the actual utility of that argument. This simple equation also implies concavity of utility functions and is the lynchpin behind the logic for why individuals value and purchase insurance. 62 Relative risk prudence and relative temperance are defined mechanistically as derivatives of the above property, respectively 33,34, 2 : 𝐸 [ 𝑢 ′ ( 𝑋 ) ] > 𝑢 ′ ( 𝑋 ) Eq. 3.2 𝐸 [ 𝑢 ′′ ( 𝑋 ) ] < 𝑢 ′′ ( 𝑋 ) Eq. 3.3 Beyond these functions, Eeckhoudt and Schelsinger (2006) were able to define the concepts on observable preferences, in a manner that hinged on pivotal work in the field published by Rothschild and Stiglitz (1970). 35,36 Rothchild and Stiglitz (1970) defined relative risk aversion as an avoidance of mean preserving spreads. Such that when a respondent was faced with an option between a certain outcome or a lottery, with a certain outcome less than the mean value of the lottery, respondents would select the certain outcome. Eeckhoudt and Schelsinger expanded that development by creating questions to characterize relative prudence and relative temperance by identifying individuals who preferred to avoid zero-mean risks occurring in a lower wealth state relative to a higher wealth state. If a respondent preferred the zero-mean risk in a lower wealth state, that would suggest imprudence. Relative temperance- based questions in previous published studies expand relative prudence questions by adding a third, zero-mean risk, beyond the relative prudence zero-mean risk. 9 However, these questions were not examined in this study for the purpose of limiting respondent cognitive load. The survey developed for eliciting higher order risk parameters is based off surveys built off of the foundation of these two key publications, particularly that published by Noussair et al., 2014. 9 However, due to the inherent difficulty in interpreting questions pertaining to imaginary health scenarios, questions in this survey took the form as seen in Figures 1 and 2, previously as 2 These equations are mutually exclusive. The condition for relative risk aversion is not required for relative prudence or vice versa. Equations 3.2 and 3.3 are largely presented to demonstrate the condition at which we would view a respondent or a group of respondents as demonstrating relative risk aversion (Eq. 3.2) or relative risk prudence (Eq. 3.3) 63 designed for relative risk aversion and relative prudence respectively. Relative risk aversion tasks take the form of selecting between a gamble with a low or high gain or a certain outcome. Relative risk prudence tasks take the form of selecting between a gamble with a low or high gain, and a secondary gamble tied to either the low or high gain or a certain outcome. Each task consisted of 9 decisions, where the gamble did not vary, but the certain outcome did. 5 certain outcome options consisted of values that were less than the expected value of the gamble, and 4 certain outcome options consisted of values that were more than the expected value of the gamble. Each higher order parameter (relative risk aversion, relative risk prudence, and relative risk temperance) can be characterized using Arrow-Pratt formulas, such as the following: 𝑅𝑅 ( 𝑋 ) = − 𝑋 𝑈 ′′ ( 𝑋 ) 𝑈 ′ ( 𝑋 ) Eq. 3.4 𝑅𝑃 ( 𝑋 ) = − 𝑋 𝑈 ′′′ ( 𝑋 ) 𝑈 ′′ ( 𝑋 ) Eq. 3.5 𝑅𝑇 ( 𝑋 ) = − 𝑋 𝑈 ′′′′ ( 𝑋 ) 𝑈 ′′′ ( 𝑋 ) Eq. 3.6 Survey Approach To estimate relative risk aversion and relative prudence over health-related QoL, we deployed a lottery-based treatment survey with different outcomes over six specific conditions: health score, proportion of year spent bedridden (bedridden weeks), mobility, cognition, vision, and money. Each of these domains were selected to test a range of inhibited conditions, ranging from less severe or easily treated (vision) to more severe and difficult to treat (bedridden weeks), general health as captured by health score, and cash. Cash was included as a treatment condition given the well-characterized risk preferences of decisions over cash in multiple studies. 9,37–39 64 Domains included in these decision tasks each have ordinal quality rankings, such that one options is clearly greater than the other. Furthermore, surveys were conducted on uncommon scales to limit participant anchoring to known average performance. Health score was asked on a 0 to 100 scale, with 0 being equal to death and 100 equal to perfect health. Bedridden weeks as asked as proportion of year able to perform usual activities with the rest of the year spent bedridden due to illness. Cognition was asked as time necessary to complete the trail-making task, a test asking respondents to connect alternating numbers in letters in ascending and alphabetic order. Vision was asked as eyesight captured using the Snellen eye test. Cash was asked on a scale with values lined up to health score based on $100,000 per a full healthy year. Previous research has shown that depending on the baseline respondent endowment, respondents may demonstrate differing risk seeking or risk averse behavior. 29,40,41 For this reason and given that treatment decision making is often completed under the assumption of utility gain for a subject, we endowed each participant a baseline level health of 20 units before taking part in the survey. A question was also asked to elicit the participants interpretation of the severity of the endowed health state, for the purpose of characterizing participant understanding of starting health state, prior to the decision tasks. Specific instructions can be seen in Appendix Note 1. A visual analog scale, corresponding to health on a scale of 0-100 was also asked prior to administering the survey for the purposes of capturing a proxy for baseline health of the survey respondent. Each task was made up of three sets of nine questions to elicit relative risk aversion and two sets of nine questions to elicit relative risk prudence. Each question set will present two lotteries to survey respondents, either a choice between two gambles, a choice between one 65 gamble and one certain equivalent, or one gamble and an option for the respondent to list a certain equivalent. See Table 3.1 for specific gambles. When conducting this survey, it is assumed that if someone selects a certain outcome, certain outcomes greater than that outcome are also preferred. For example, for Task 1, if a respondent selects the certain outcome 28, then they are prompted to select all certain outcomes greater than 28. Finding the switch point is key for this survey design and for later analyses. Typically, in a lab setting, a researcher is allowed to provide prompts so the respondent fills this correctly. Since this survey was scaled among a representative, large swath of the US population, errors were coded into the survey to prompt the respondent to follow these assumptions. Given this error window, we also allowed the respondent to change an answer at any point during the survey. Lottery payouts were normalized on a scale of 0 to 100, for ease of comparison between treatment conditions. However, tasks were scaled to representative values according to the treatment framing condition (e.g. 100 to 5500 feet during the mobility treatment condition, where 0 corresponds to 100 feet and 100 correspond to 5500 feet). Two attention checks were programmed into the survey, prompting the user to select either the certain gamble or certain outcome depending on the question. One validity check was also programmed into the survey, with a certain outcome option that greatly exceeded either gain from the gamble, under the assumption that a rational respondent would always select that certain outcome. Experimental Design and Treatments For ease of exposition and display, in Table 3.1 and the rest of this article, the following notation will be used to describe the lotteries. Let x_y denote a lottery that yields an outcome of 66 x or y, with equal probability. Similarly, compound lotteries, such as that in Figure 3.2, are written as x_y+(z1_-z1). Subjects were presented with one group of nine decisions at a time. The nine choices measuring relative risk aversion and relative risk prudence were displayed in ascending order, such that the certain treatment outcome improved monotonically. Choices for relative risk aversion and relative prudence differed regarding average expected benefit of treatment gamble. Choices for relative prudence differed in the size of the zero-mean risk, as well as whether the zero-mean risk occurred in the better outcome state or worse outcome state. These variations allowed us to study the effect of changes in risk magnitude and impact of zero mean risk on prudent decision-making regarding health status. No indifference options were provided, i.e., subjects always had to choose one of the lotteries. All risks involved in this experiment were equiprobable lotteries. For interpretation of compound lotteries in terms of relative prudence, it is crucial to emphasize the independence of the multiple risks. Therefore, we presented lotteries to respondents graphically by means of a simple table (Appendix Note 3.1). Appendix Note 3.1 contains an example of the display participants saw for the most complex decision type in the experiment, that for relative risk prudence. Question instructions and priming conditions are given in the Appendix. There were six different treatment conditions, as summarized in Table 2. Each subject participated in the health score (otherwise known as quality-of-life) treatment and in one of bedridden weeks, cognition, mobility, vision, or cash treatments. Each treatment condition was similarly scaled, based off quality-of-life treatment gambles (Table 3.3). Different treatment conditions were tested to determine consistency of risk attitudes depending on baseline health 67 condition and context of decision making. More details on treatment conditions are provided in Table 3.3. Sample sizes for the different treatment conditions are also shown in Table 3.3. Task 1 was considered the reference task among relative risk aversion questions. Using task 1 as the reference, tasks 2 and 3 were set up to test if individual utility functions were consistent with constant absolute risk aversion (CARA) and constant relative risk aversion (CRRA; see below). 42 Under CRRA: 𝐶𝐸 ⁡~ ⁡[ 𝑥 , 𝑝 ⁡; 𝑦 , 1 − 𝑝 ] if and only if 𝛼 𝐶𝐸 ⁡~ ⁡[ 𝛼𝑥 , 𝑝 ⁡ ; 𝛼𝑦 , 1 − 𝑝 ] Under CARA: 𝐶𝐸 ⁡~ ⁡[ 𝑥 , 𝑝 ⁡; 𝑦 , 1 − 𝑝 ] if and only if 𝛼 + 𝐶𝐸 ⁡ ~ ⁡[ 𝛼 + 𝑥 , 𝑝 ⁡ ; 𝛼 + 𝑦 , 1 − 𝑝 ] To test CRRA, 1.15 was multiplied by the values in task 1 to set up task 3. To test CARA, 8 was added to the values in task 1. Wilcoxon sum rank statistical tests were conducted on the linear approximations of the certainty equivalents to identify if respondent followed either CRRA or CARA. Analysis The following are calculated and presented in the following pages. The number of subjects per health condition evaluation (vision, cognition, mobility, bedridden weeks, and health-state utility) will be reported. Sample demographics per arm will also be reported in primary tables. Following reported sample demographics, average comparator arm sample responses will be produced, indicating under which arms do subjects tend to select risk averse or risk prudent health-based lotteries. Prevalence of relative risk aversion and relative risk prudence will be calculated as the sum of risk averse or prudent answers of the 9 decisions per task. Certainty equivalents of responses will be captured using linear approximation at the switch point from the gamble to the certain outcome. For example, if in Task 1 in Table 3.1, a 68 respondent’s switch point from gamble to certain outcome is between certain outcomes 33 and 38, their certainty equivalent of the gamble is approximated to be 35.5. This is computed for each respondent based on their switch point. One-sample t-test will be used to evaluate if mean approximate certainty equivalents or risk aversion or risk prudence answer prevalence are different from specific values. Wilcoxon rank-sum test will be used to evaluate if prevalence or mean approximate certainty equivalents are different from one grouping relative to another. Rank correlation between relative risk aversion and relative risk prudence will also be calculated. These outputs will provide context on how participants responded to each of these lotteries. Demographic correlates will be analyzed relative to each higher risk order. Finally, parametric estimates of relative risk aversion and relative risk prudence under expected utility using constant relative risk aversion (CRRA), expo-power, power-expo, and hyperbolic absolute risk aversion (HARA). A variety of coefficient constraints will also be employed to identify the best model for evaluating higher order risk parameters among our sample responses. Root mean squared error (RMSE) will also be evaluated for each model to examine which had the best fit among our responses. The utility function with the lowest RMSE will be employed for subsequent analyses. Some specific utility conditions will also be tested, such as constant relative risk aversion and constant absolute risk aversion, through question design for each treatment gamble option and listed certainty equivalents. 42 Statistical testing of utility function coefficient estimates will be used to determine the utility function that best reflects subject risk preference over our general measure of health score. 69 Subjects and Background Data In total, 1,283 subjects were recruited and 1,202 participated in this experiment (Table 3.2). Each subject was an active members and participants of the University of Southern California’s Understanding America Study, managed by The Center for Economic and Social Research. Understanding America Study consists of close to 10,000 participants that complete an internet-based survey each month. Respondents are reimbursed for completing the questionnaires. An overview of number of subjects per treatment condition, scale per treatment conditions, stakes and risk of Z1 can be seen in Table 3.3. Financial-based incentives were not tied to specific questions in our survey. The Understanding America Study panel is a representative sample, in terms of observable background characteristics of the general American population. The subsample invited for this survey were recruited to represent the general American population and have completed other health-based questionnaires within the Understanding America Study. Furthermore, important demographic and background health information was captured for participants in a series of preceding surveys tied to the Understanding America Study. These characteristics were reported for the purpose of contextualizing captured responses. Furthermore, theoretical relationships have been identified by baseline demographic characteristics and risk attitudes pertaining to financial-based lotteries. Thus, capturing that data will help us contextualize risk attitudes towards health-based decision making. 70 3.3 Results Prevalence of Relative Risk Aversion, Relative Prudence and Demographic and Health Correlates We measured the incidence or relative prudence and relative risk aversion in our sample, and then considered other factors that correlate with these risk attitudes. We measured an individual’s relative risk aversion with the number of certain outcomes the respondent made from the 9 decisions involving a sure treatment outcome and a risky treatment (tasks 1-3, decisions 1-9 in Table 3.1). This was repeated for each of the 3 relative risk aversion choice sets. We measured relative prudence as the number of certain outcomes selected when faced with a prudent question set (tasks 4-5, decisions 1-9 in Table 3.1). The use of count of the number of binary decisions consistent with relative risk aversion and relative prudence as measures of strengths of these attitudes is consistent with the literature. 9,43–46 This assumes that a respondent who makes zero errors would consistently choose the risk averse (prudent) or the certain outcome in all tasks. However, if a respondent tends to commit errors, errors made are more likely to be closer to the treatment or gamble options than the certain outcomes. If errors occur, a respondent would equally choose both treatments or monetary gambles relative to the certain outcomes. Furthermore, this allows for participants that are somewhat risk averse or prudent choose the risk averse or prudent lottery in a greater percentage of incidences and vice versa (i.e., risk seeking or imprudent respondents selecting the uncertain outcomes more frequently.) These assumptions allow us to produce a ranking of subjects in terms of the underlying strength of their preferences. 47,48 Error incidence could also vary across choice tasks depending on size and risk of payoffs. A risk averse subject is more likely to choose the certain outcome when the difference between the utility of the risk averse 71 and risk seeking alternatives is large (and similarly for relative prudence.) Count variables thus provide information about the strength and prevalence of the higher order risk attitudes (how many and how strongly respondents can be classified as risk averse or prudent.) Furthermore, we can rank these people based on these preferences, without the additional complexity involved in the elicitation of cardinal measures per each choice set. Overall prevalence of relative risk aversion and relative prudence by treatment condition Table 3.4 displays results for the entire sample, as well as separately for each treatment of interest (health score, bedridden weeks, mobility, cognition, vision, and cash). In each treatment condition, most decisions are consistent with relative risk aversion and relative prudence. Respondents largely selected the certain outcome, a risk averse option, in the mobility and cognition treatment conditions. However, respondents overwhelmingly selected the risk averse option in the cash condition. Across the health-related treatment conditions (health score, bedridden weeks, mobility, cognition, and vision), respondents appeared prudent, selecting the certain outcome more often during the prudent questions than the risk averse questions. Figure 3.3 provides further details on the distribution of choices of risk averse selections on health score questions. Relative risk aversion and relative prudence are present, regardless of treatment condition. However, a notable number of respondents select responses throughout the decision tasks (i.e. more risk averse (prudent) or risk seeking (imprudent). Figure 3.4 provides an overview of prevalence of risk averse and prudent selections per health score condition task. Among health score and treatment conditions, there is a noticeable number of respondents either selecting all risk averse or no risk averse certain outcomes. Among all health score questions, 7.8% selected all risk seeking or gamble responses regardless of question or task, and 10.7% of 72 respondents selected all risk averse or certain outcome responses, regardless of question or task. However, there are also several respondents that selected varying degrees of certain outcomes (Figures 3.3 and 3.4). Columns 2-6 of Table 3.4 show results for each treatment condition separately. Although answers were relatively consistent despite treatment conditions, we do see slight differences between the conditions. Relative risk aversion is stronger in cash, cognition, and mobility treatment conditions. Relative prudence is stronger in mobility, cognition, and cash treatment conditions. This is indicated by the number of risk averse choices displayed in rows 1-3 and prudent choices displayed in row 4-5. Prudent decision making is also indicated by the average response when the secondary treatment or monetary gamble occurs in the high wealth state. The number of selected certain outcomes is lower for the secondary gamble in the high wealth state, indicating that subjects are more prudent among health score, bedridden weeks, cognition, and vision treatment conditions. This lines up with evidence provided by Eeckhoudt and Schlesigner (2006) and Noussair (2014) suggesting respondents are more likely to take on aggregated risks when the risk occurs in the high wealth or health state. Approximate Mean Certainty Equivalents Table 3.5 contains approximated certainty equivalents, generated using prevalence of risk averse or prudent selections. For each treatment condition and each task, approximated certainty equivalents on average are significantly lower than expected values of the treatment or wealth gambles. The one non-significantly less approximated mean certainty equivalent is that 73 associated with vision, task 1. These results indicate the sample demonstrated relative risk aversion and relative prudence. Mean approximated certainty equivalents further suggest relative prudence observed in health score, bedridden weeks, cognition, and vision treatment conditions. For each of those conditions, the mean approximated certainty equivalent for task 4 is greater than that for task 5. Higher mean approximated certainty equivalent to a gamble suggests a greater value of the gamble, or higher preference for that gamble. Given the higher mean approximate certainty equivalent is for task 4, the task with the second gamble assigned to the higher wealth state, relative to task 5, second gamble assigned to low wealth state, that indicates this population was demonstrating relative prudence in the previously mentioned domains. Using task 2 and task 3 approximated mean certainty equivalents, significance tests were used to identify if respondents demonstrated decision making under CRRA or CARA conditions (Table 3.5). CRRA was observed in bedridden weeks, mobility, and cognition domains using the Wilcoxon sum rank test to measure if task 2 divided by task 1 was equal to the set multiple assigned to test CRRA. 42 CARA was not observed in the mean approximated values within any treatment condition (Table 3.5). Correlation between relative risk aversion and relative prudence, health score and treatment conditions Table 3.6 demonstrates that there is substantial positive correlation among the two measures (relative risk aversion and relative prudence) in our samples. Table 3.7 provides more detail on the correlation between relative risk aversion and relative prudence. Each row here contains the average number of prudent and risk averse choices of individuals, based on the 74 number of safe choices they made in the risk apportionment relative prudence and aversion tasks. Generally, more risk-averse subjects are prudent. The degree of relative prudence seems to be closely tied to the degree of relative risk aversion. This finding is consistent with findings seen in the literature for both health and financial based decision-making. 9,29,49 Published estimates previously also found that risk seekers tended to be more imprudent than risk avoiders. Relative prudence for our sample is strong for all of our respondents, except those that were the most extreme risk seekers. Despite findings here, relative risk aversion and relative prudence may largely be dependent on baseline characteristics that have not yet been identified or captured in this survey. Among the variables and attributes captured in the Understanding America Study, we will later try to elicit correlation and associated higher order raw risk choices contingent on baseline characteristics. Demographic correlates with relative risk aversion and relative prudence Within our analysis, we also considered the influence of demographic characteristics on the prevalence for relative risk aversion and relative prudence among responders to the health score treatment condition. Demographic variables were captured using baseline questions on respondent characteristics. Independent variables were those captured among an Understanding America Study subpopulation, which had responded to a health-related question set in an earlier survey. We also captured several similar independent variables seen previous literature. 9 Baseline values on respondent health status as measured by the visual analog scale, beginning of the question set were also used as critical independent variables for contextualizing our responses. Health status, education, and income have been shown to be critical influencers on baseline risk attributes and risk preference. 50–54 Although variables here are reported in discrete 75 forms, we report ordinary least squares estimates here for ease of interpretation of impact on baseline demographics and characteristics on risk preference. Table 3.8 displays these results. Education and number of household members had a large impact on the prevalence or risk averse answers on task 1, with 9 th , 10 th and college educated or beyond having a positive association with relative risk aversion prevalence. Currently being on sick live also is associated with relative risk aversion prevalence. Respondents having had 7 number of individuals in household had a negative association with relative risk aversion and relative risk prudence prevalence. Education, race, being on sick leave from work, income, and number of household members, had significant associations on relative risk prudence. 9 th and 10 th grade education were significantly associated with lower levels of relative risk prudence. White respondents had higher degrees of relative risk prudence on task 4, as did being on sick leave and having 3 members in the household. Last, being low income ($5,000 - $7,499) or mid-to-high income ($20,000 to $24,999 and $50,000 to $59,999) had negative correlations with relative risk prudence prevalence. Parametric Estimates of Relative Risk Aversion and Relative Prudence Most microeconomic level empirical studies of risk preferences and decision making, rely on a parametric expected utility framework to estimate higher order estimated risk parameters. In this section, we provide our own estimates of the coefficients of relative risk aversion, relative prudence, and relative temperance for the representative respondent in our survey. To complete this task, we utilized constant relative risk aversion, expo-power, and hyperbolic absolute relative risk aversion functions, and assumed expected utility. 76 All 45 decisions per treatment condition (health score, bedridden weeks, cognition, vision, mobility, and cash) were used in nonlinear least squares equations to determine utility functions coefficients. Per each decision questions, answers to the 9 tasks were used to determine the specific switch point per participant. That switch point was assumed to approximate the certainty equivalent for the nonlinear least squares regression, that was constructed from the constant relative risk aversion, expo-power, and hyperbolic absolute risk aversion functions. Mean, median and broader measures of dispersion of individual utility function parameter estimates were captured from each individual estimated utility function. For utility functions that needed an endowed argument for the purpose of calculating higher order parameters, the average approximated certainty equivalent, 0.475. We provide separate estimate results for each of our six treatment conditions. For the CRRA utility function, 𝑢 ( 𝑥 ) = 𝑥 1 − 𝜌 ( 1 − 𝜌 ) − 1 , where the estimates of relative risk aversion, relative prudence, and relative temperance is provided by 𝜌 , 𝜌 + 1, and 𝜌 + 2, respectively. For the expo-power utility function, we assumed 𝑢 ( 𝑥 ) = ( 1 − exp ( − 𝛼 𝑥 1 − 𝜇 ) ) 𝛼 − 1 . The parameter estimates for relative risk aversion equals 𝑅𝑅 ( 𝑥 ) = 𝜇 + 𝛼 ( 1 − 𝜇 ) 𝑥 1 − 𝜇 . Expressions for the relative prudence and relative temperance are quite a bit more complex and are provided in the appendix with further details for elicitation. HARA utility higher order risk parameters were estimated using the following function 𝑈 ( 𝐻 ) = ⁡ 1 − 𝛾 𝛾 ⁡ ( 𝛼𝐻 1 − 𝛾 + 𝜂 ) 𝛾 . 8 In this equation, Phelps and Lakdawalla demonstrate 𝛼𝐻 1 − 𝛾 + 𝜂 can be simplified to a term 𝑍 . 8 Specification in Lakdawalla et al., 2021 demonstrates that 𝑍 can be simplified to 𝑍 = 𝐻 + 𝜂 . From there, relative risk aversion can be estimated as 𝑟 𝐻 ∗ = ( 1 − 𝛾 ) ( 𝐻 𝑍 ), relative risk prudence can be estimated as 𝜋 𝐻 ∗ = ( 2 − 𝛾 ) ( 1 − 𝛾 ) 𝑟 𝐻 ∗ , and relative temperance can be estimated as 𝜏 𝐻 ∗ = ( 3 − 𝛾 ) ( 1 − 𝛾 ) 𝑟 𝐻 ∗ . Both expanded and simplified forms 77 of HARA were estimated. Constraints for utility function coefficients were examined under the assumption that the marginal utility of life-extension is positive. For example, utility measures from health score cannot be listed as negative, due to the aforementioned consequence. Table 3.9 A-D displays estimates for relative risk aversion, relative prudence, and relative temperance per the four utilized utility functions (hyperbolic absolute risk aversion, constant relative risk aversion, power-expo and expo-power functions) for health score. Eleven functions were estimated, contingent on expanded or simplified forms of utility functions, and with different coefficient constraints. Assuming CRRA utility, the coefficient 𝜌 was found not equal to zero (p-value < 0.01). Given that 𝜌 is positive and not equal to zero, we can assume the utility function is concave and not convex or linear. Assuming HARA utility, the coefficient 𝜂 was not found equal to zero (p- value < 0.01). Since 𝜂 is found not equal to zero, we can assume that HARA is a more representative utility function to risk preference over health score that CRRA. Assuming each of the expo-power utility functions with varying constraints, neither 𝛼 or 𝜇 is found equal to zero (p-value < 0.01), which can be interpreted as further confirmation that utility over health score is not linear. Finally, constant absolute risk aversion (CARA) was tested using approximated certainty equivalents, under the following property 𝑢 ( 𝑥 ) ≽ 𝑢 ( 𝑦 ) if and only if 𝑢 ( 𝑥 + 𝑐 ) ≽ 𝑢 ( 𝑦 + 𝑐 ). Wilcoxon sum rank tests indicated approximated certainty equivalents were not invariant to additive constants, suggesting respondents did not display CARA. Assuming CARA is violated, HARA represents the best utility function option for modeling respondent risk preferences over treatment decision making over health score, relative to expo-power or power- expo utility functions which assume CARA. Among the estimate HARA utility functions, the simplified form, as specified in published literature surrounding the implications of this topic, 78 has the lowest RMSE, suggesting it had the greatest fit for our responses. Thus indicating that utility function should be the one use to model higher order risk parameters among our population. Estimates over higher order risk estimates, relative risk aversion, relative prudence, and relative temperance varied depending on percentile (Table 3.10). However, the median was similar to the mean estimate values for each higher order risk estimate. Higher order parameter estimates also varied across responses framed through financial-based lotteries, and tended to demonstrate greater relative risk aversion, relative prudence, and relative temperance. Figure 3.5 provides visual overviews of the distribution of relative risk aversion, relative prudence, and relative temperance. Higher order risk parameters were consistent across treatment options, with parameters associated with cash related decision-making indicating a greater level of relative prudence and relative risk aversion over treatment options pertaining to quality-of-life or a more specific health conditions related to vision, cognition, mobility, or bedridden weeks spent in the next year (Appendix Table 3.1). Higher Order Relative Risk Estimates Relative to Financial-Based Lotteries After estimating higher order relative risk aversion, relative prudence, and relative temperance over health score and five other treatment conditions, including cash, we evaluated the relationship between higher order risk scores of health and cash. Ratios were created for relative risk aversion, relative prudence, and relative temperance between health score and cash. Although relative risk estimates over health score and cash do not have identical distributions, we found there was some measure of stability between the two conditions, when outliers beyond 79 the 5 th and 95 th percentile were removed (See Table 3.11 and Figure 3.6). Sample size decreased from 220 observations to 198 with the exclusion of the 5 th and 95 th percentiles. 3.4 Discussion In this study, relative risk aversion and relative prudence were measured over a US representative population from the Understanding America Study. Relative risk aversion and relative prudence were observed in across treatment conditions, with escalating values indicating higher relative risk aversion and relative prudence is contingent on the treatment condition (e.g. treatment decision impacting typical mobility v treatment decisions impacting vision). Furthermore, relative risk aversion and relative prudence were also found to be dependent on certain demographic characteristics, such as family size, education, income, and employment status in certain instances. One of the strongest associations was seen with individuals that had lower education demonstrating a higher, negative correlation with risk averse and prudent decision making. Prevalence of risk averse and risk prudent decision making over the health score treatment condition suggests that this population of respondents are risk averse. However, despite framed treatment gambles in the gain domain, there were a sizable number of respondents on either end of the risk averse and risk prudent spectrum. Observing a bimodal response to these questions lines up with what is seen in similar studies examining behavior and preference over treatment decision making. 29,55 Furthermore, prevalence of prudent decision- making also is consistent with previous studies finding respondents who are typically risk- averse, are also likely to be prudent. 9,31,43,45,46 80 Relative risk prudence is characterized as a respondent’s preference for a zero-mean risk gamble in a high wealth state. In this survey, respondents demonstrated a higher prevalence of relative prudence relative to relative risk aversion, across all treatment conditions. However, an advantage of this survey design involved two relative risk prudence tasks, one with the second gamble in a high wealth state and the second with a second gamble happening in the low wealth state. Among the health score, bedridden weeks, cognition, and vision conditions, participants demonstrated a higher prevalence of relative prudence in the low wealth gamble than the high wealth gamble. This suggests respondents are demonstrating relative prudence among these conditions in the classical sense, due to their early switch from the treatment gamble to certain outcome in the low wealth state (i.e., respondents preferred the gamble in the high wealth state, suggesting relative prudence). This same phenomenon was not seen among cash or mobility domains, suggesting when faced with decisions that impact mobility or large cash gains, respondents may be slightly imprudent. Another way of interpreting this result is that when dealing with treatment gambles surrounding general health, respondents prefer to take secondary gambles when they are ahead. Whereas among the respondents in this survey that answered the cash framed question, they preferred the secondary gamble when they already were faced with a low option. The implications of which are that respondents could stand to hit a much lower point if the respondent wins the lower amount and is faced with the loss during the zero-mean risk. Whereas in health score domains, once folks win the lower net gained outcome, those respondents are fine with ending the treatment gamble and remaining in the current state. Over cash, prevalence of relative risk aversion and relative prudence is sizable relative to the health-related treatment conditions (health score, mobility, bedridden weeks, vision, and cognition). This is also consistent with what is seen in the literature of risk preference over 81 monetary gambles. 9,38,42 Furthermore, observing consistent respondent preferences over the monetary treatment condition does suggests a degree of instrumental validation of the survey generated here, despite the seemingly random decision selection in some instances during the treatment conditions tasks. Finally, this study has characterized parametric estimates of relative risk aversion, relative risk prudence, and relative risk temperance in a representative U.S. sample. These estimates were in the range of that previously seen in similar studies examining risk preferences over cash-based surveys. 9–11 Estimates of these higher order risk parameters over the cash treatment condition are higher using assuming HARA utility. Taking the ratio of health score higher order relative risk parameters over cash higher order relative risk parameters provides a ratio that can be used to convert financial based higher order risk parameters to usable values for evaluating healthcare-decision making. In the case of this study, that value would be assumed, using a HARA utility function, as 0.98, 0.67, and 0.62 for relative risk aversion, relative prudence, and relative temperance, assuming expected utility and observations below the 5 th and above the 95 th percentile do not reflect the risk preferences of the general public. Relative risk preference elucidated in this study have several implications for economic evaluations of new health technologies, particularly surrounding the generation of the QALY. Quality-adjusted life years can simply be defined as the value of health gain from a treatment and is often quantified as the following: ⁡𝑇𝑄 ; where 𝑇 represents time and 𝑄 represents health state values. Under this model, 𝑇 is usually determined using results from a real-world study or clinical trial and 𝑄 is elicited from a health preference study. This term serves as the backbone to classic incremental cost-effectiveness studies evaluating new health technologies, but also has a few limitations. Many critics have viewed the QALY to value one life over another, to put a 82 higher denomination on a higher health state than a lower wealth state. 56,57 Other critics have argued that the QALY model can be used to restrict budgets using arbitrary willingness-to-pay thresholds, based on dollars per QALY. 58,59 Some critics have also cited the value as being over egalitarian, that a 0.1 QALY improvement among a sick person is equivalent to a 0.1 QALY improvement in a health person and should be valued as such. 60,61 Values elicited in this study can be used to generate a better version of the QALY. Lakdawalla & Phelps developed the innovative generalized, risk-adjusted cost- effectiveness (GRACE) framework to adjust classic economic evaluations for many of the concerns cited above and more. 7,8 With risk preferences elicited here, criticisms above regarding the QALY can be addressed. The GRACE model contains the GRA-QALY, which scales the original QALY by measures of treatment uncertainty with respondent preferences pertaining to risk over health. The GRA-QALY also adjusts for treatment positive skewness or “hope,” such that a treatment with a greater positive skew will be valued more highly. However, the adjustment is contingent on consumer preferences regarding health risk, which has been captured in this study. Considering the criticism that the QALY arbitrarily assigns restrictive health budgets that rely on the willingness-to-pay threshold ($/QALY), willingness-to-pay thresholds are contingent on improvements in health utility (QALY). However, GRACE allows for a the GRACE willingness-to-pay threshold to vary with health as opposed to solely utility. This creates a dynamic and rigorous willingness-to-pay threshold that varies depending on the health state improvements and risk preferences elicited here. These findings also have implications setting an optimal willingness-to-pay threshold, as identified by Phelps & Cinatl (2021). 62 In this paper, Phelps & Cinatl identify 𝐾 𝑀 as a value that can be multiplied by gross domestic product per capita to determine the optimal willingness-to- 83 pay threshold. 𝐾 𝑀 is found to be equal to 1 𝜋 ∗ − 2 𝑟 ∗ . Under that assumption, 𝐾 𝑀 is estimated over the 5 th percentile of relative prudence and relative risk aversion estimates to the 95 th percentile of higher order estimates (Appendix Table 3.2). Mean and median estimates of relative risk aversion and relative prudence suggest the optimal willingness-to-pay threshold is either $83,611/QALY and $88,256/QALY. These estimates are similar to previously published estimates that found the optimal willingness-to-pay threshold close to $100,000/QALY. 63,64 There are several limitations to this study to be considered when interpreting these results. First, UAS is also predominately deployed among folks trading off use of an internet, enabled laptop for participation in longitudinal surveys. Because of this, this sample may not completely represent the general US public. However, this survey sample has been used in multiple publications and studies to generate population estimates. 65,66 Second, treatment decisions for health conditions are abstract decisions in an online survey setting. When faced with a severe health condition, debating between an okay certain outcome or a treatment gamble with an improved, but poor outcome or a great outcome may not fully reflect treatment decisions made in a real-world setting. Furthermore, although this sample is representative of the general U.S. public, severe health conditions may not necessarily be reflected in this sample, further alienating this survey from treatment decision making in a real- world setting. Despite this, using ordinal values helps ground the respondent to anchor based off the necessary certain outcome to avoid the gamble. This survey format also allowed for ease of estimation of parametric estimates. Third, unlike studies completed previously in this field, no compensation was tied to outcomes of gambles within this survey. However, this is largely due to an inability to assign requisite cash amounts to certain treatment outcomes, and potentially complicating respondent 84 decision making by cross contaminating health-related decision making with cash-based compensation. Fourth, to remove life expectancy risk due to illness as a factor from the individual decision-making in this survey, respondents were instructed to consider that their lives continued for 30 more years beyond the first year of treatment. For some subjects in this survey, that may have represented a life expectancy well-beyond normal, effectively minimizing the impact of a year of treatment. However, implanting a fix on life expectancy beyond the treatment was necessary to ensure we captured relevant factors impacting treatment decision-making, leaving little misinterpretation by the respondent. Fifth, the tasks in this study can be cognitively burdensome for subjects. Project treatment decision-making on imaginary health scenarios and attempts to identify a switch point at which a respondent would prefer a certain outcome over a treatment gamble is difficult in a lab setting with a research assistant. While detailed instructions were added to make this difficult, this does not take away from the focus necessary to complete this survey. For that reason, there exists a possibility of respondents growing wary of questions or responding without critically thinking through answers. However, 62.4% of respondents reported that the task was easy or not difficult. This suggests that most respondents were able to complete this survey as instructed. Sixth, this study primarily focuses on risk over treatment outcomes in a single dimension, that is whether to improve quality-of-life, regular activities, vision, cognition, or mobility over the course of a single year. None of the scenarios presented to respondents contained multivariate decisions, something that is likely a hallmark in most healthcare decision-making (i.e. a therapy that improves life expectancy with a lower quality-of-life vs. a therapy that 85 improves quality-of-life with a lower life expectancy). 5,28,67,68 Granted this inclusion would provide greater generalizability of this survey to real healthcare treatment options, it would also limit the ability to include respondent answers for calculating individual utility functions and higher-order parameter estimates. An area of future research should emphasize and focus on translating answers to multivariate problems into individual utility functions. 3.5 Conclusion This study characterizes relative risk aversion and relative prudence over general health score, a variety of escalating treatment conditions (vision, cognition, mobility, and bedridden weeks), and cash. By including multiple treatment conditions, cash, and a general form of health, this study provides estimates on how higher order risk parameters change over severity of condition, correlation between relative risk aversion and prudent decision-making over treatment conditions, and a new estimate to scale previously published higher order risk parameter estimates on financial-based gambles or lotteries to health-care decision-making. These findings can be critical in future evaluation of health technologies and healthcare decision-making. 86 3.6 References 1. Wennberg J, Gittelsohn A. Small Area Variations in Health Care Delivery. Science. 1973;182(4117):1102-1108. doi:10.1126/science.182.4117.1102 2. Phelps C. Demand for health insurance: a theoretical and empirical investigation. Published online 1973. 3. Lakdawalla DN, Phelps CE. Health technology assessment with risk aversion in health. Journal of Health Economics. 2020;72:102346. doi:10.1016/j.jhealeco.2020.102346 4. Dubois RW, Westrich K. As Value Assessment Frameworks Evolve, Are They Finally Ready for Prime Time? Value in Health. 2019;22(9):977-980. doi:10.1016/j.jval.2019.06.002 5. Sanders GD, Neumann PJ, Basu A, et al. Recommendations for Conduct, Methodological Practices, and Reporting of Cost-effectiveness Analyses: Second Panel on Cost- Effectiveness in Health and Medicine. JAMA. 2016;316(10):1093-1103. doi:10.1001/jama.2016.12195 6. Neumann PJ, Kim DD, Trikalinos TA, et al. Future Directions for Cost-effectiveness Analyses in Health and Medicine. Med Decis Making. 2018;38(7):767-777. doi:10.1177/0272989X18798833 7. Lakdawalla DN, Phelps CE. Health Technology Assessment With Diminishing Returns to Health: The Generalized Risk-Adjusted Cost-Effectiveness (GRACE) Approach. Value in Health. 2021;24(2):244-249. doi:10.1016/j.jval.2020.10.003 8. Lakdawalla DN, Phelps CE. A guide to extending and implementing generalized risk- adjusted cost-effectiveness (GRACE). Eur J Health Econ. Published online September 8, 2021. doi:10.1007/s10198-021-01367-0 9. Noussair CN, Trautmann ST, van de Kuilen G. Higher Order Risk Attitudes, Demographics, and Financial Decisions. The Review of Economic Studies. 2014;81(1):325-355. doi:10.1093/restud/rdt032 10. Guiso L, Paiella M. Risk Aversion, Wealth, and Background Risk. Journal of the European Economic Association. 2008;6(6):1109-1150. doi:10.1162/JEEA.2008.6.6.1109 11. Harrison GW, Lau MI, Rutström EE. Estimating Risk Attitudes in Denmark: A Field Experiment*. The Scandinavian Journal of Economics. 2007;109(2):341-368. doi:10.1111/j.1467-9442.2007.00496.x 12. Von Gaudecker HM, Van Soest A, Wengstrom E. Heterogeneity in risky choice behavior in a broad population. American Economic Review. 2011;101(2):664-694. 87 13. Gollier C. Pricing the future: The economics of discounting and sustainable development. Unpublished manuscript, to appear with Princeton University Press, Princeton, NJ, USA. Published online 2011. 14. Bramoullé Y, Treich N. Can uncertainty alleviate the commons problem? Journal of the European Economic Association. 2009;7(5):1042-1067. 15. Alm J. Uncertain tax policies, individual behavior, and welfare. The American Economic Review. 1988;78(1):237-245. 16. Snow A, Warren RS. Ambiguity about audit probability, tax compliance, and taxpayer welfare. Economic Inquiry. 2005;43(4):865-871. 17. Garber AM, Phelps CE. Economic foundations of cost-effectiveness analysis. Journal of Health Economics. 1997;16(1):1-31. doi:10.1016/S0167-6296(96)00506-1 18. Linley WG, Hughes DA. Societal views on NICE, cancer drugs fund and value-based pricing criteria for prioritising medicines: a cross-sectional survey of 4118 adults in Great Britain. Health Econ. 2013;22(8):948-964. doi:10.1002/hec.2872 19. Green C, Gerard K. Exploring the social value of health-care interventions: a stated preference discrete choice experiment. Health Econ. 2009;18(8):951-976. doi:10.1002/hec.1414 20. Nord E, Richardson J, Street A, Kuhse H, Singer P. Who cares about cost? Does economic analysis impose or reflect social values? Health Policy. 1995;34(2):79-94. doi:10.1016/0168-8510(95)00751-D 21. Paulden M. Recent amendments to NICE’s value-based assessment of health technologies: implicitly inequitable? Expert Review of Pharmacoeconomics & Outcomes Research. 2017;17(3):239-242. doi:10.1080/14737167.2017.1330152 22. Attema AE, Brouwer WBF, l’Haridon O, Pinto JL. An elicitation of utility for quality of life under prospect theory. Journal of Health Economics. 2016;48:121-134. doi:10.1016/j.jhealeco.2016.04.002 23. Schosser S, Trarbach JN, Vogt B. How does the perception of pain determine the selection between different treatments?: Experimental evidence for convex utility functions over pain duration and concave utility functions over pain intensity. Journal of Economic Behavior & Organization. 2016;131:174-182. doi:10.1016/j.jebo.2015.08.009 24. Rouyard T, Attema A, Baskerville R, Leal J, Gray A. Risk attitudes of people with ‘manageable’chronic disease: an analysis under prospect theory. Social Science & Medicine. 2018;214:144-153. 88 25. Courbage C, Rey B. Prudence and optimal prevention for health risks. Health Economics. 2006;15(12):1323-1327. 26. Pauker SG. Moments when utilities are functional. Medical Decision Making. 2014;34(1):4-7. 27. Bleichrodt H, Quiggin J. Life-cycle preferences over consumption and health: when is cost- effectiveness analysis equivalent to cost–benefit analysis? Journal of health economics. 1999;18(6):681-708. 28. Eeckhoudt L, Rey B, Schlesinger H. A good sign for multivariate risk taking. Management Science. 2007;53(1):117-124. 29. Attema AE, l’Haridon O, van de Kuilen G. Measuring multivariate risk preferences in the health domain. Journal of Health Economics. 2019;64:15-24. doi:10.1016/j.jhealeco.2018.12.004 30. Attema AE, Frasch JJ, L’Haridon O. Multivariate risk preferences in the quality-adjusted life year model. Health Economics. 2022;31(2):382-398. doi:10.1002/hec.4456 31. Doctor JN, Huesch MD, Meeker D. Rethinking the value of survival: clinical trials should measure patient preferences for survival on entry to trials. Journal of Clinical Epidemiology. 2016;77:137-138. doi:10.1016/j.jclinepi.2016.03.030 32. Mayrhofer T, Krieger M. Patient Preferences and Treatment Thresholds under Diagnostic Risk: An Economic Laboratory Experiment. In: ZBW - Deutsche Zentralbibliothek für Wirtschaftswissenschaften, Leibniz-Informationszentrum Wirtschaft; 2012. Accessed March 6, 2022. https://www.econstor.eu/handle/10419/62033 33. Kimball MS. Precautionary Saving in the Small and in the Large. Published online 1989. 34. Kimball MS. Standard Risk Aversion. Econometrica. 1993;61(3):589-611. doi:10.2307/2951719 35. Eeckhoudt L, Schlesinger H. Putting risk in its proper place. American Economic Review. 2006;96(1):280-289. 36. Rothschild M, Stiglitz JE. Increasing risk: I. A definition. Journal of Economic theory. 1970;2(3):225-243. 37. Holt CA, Laury SK. Risk aversion and incentive effects. American economic review. 2002;92(5):1644-1655. 38. Saha A. Expo‐power utility: A ‘flexible’form for absolute and relative risk aversion. American Journal of Agricultural Economics. 1993;75(4):905-913. 89 39. Eeckhoudt L, Etner J, Schroyen F. The values of relative risk aversion and prudence: A context-free interpretation. Mathematical Social Sciences. 2009;58(1):1-7. 40. Tversky A, Kahneman D. The framing of decisions and the psychology of choice. Science. 1981;211(4481):453-458. doi:10.1126/science.7455683 41. Tversky A, Kahneman D. Advances in prospect theory: Cumulative representation of uncertainty. J Risk Uncertainty. 1992;5(4):297-323. doi:10.1007/BF00122574 42. Miyamoto JM, Eraker SA. Parametric models of the utility of survival duration: Tests of axioms in a generic utility framework. Organizational Behavior and Human Decision Processes. 1989;44(2):166-202. doi:10.1016/0749-5978(89)90024-1 43. Deck C, Schlesinger H. Exploring Higher Order Risk Effects. The Review of Economic Studies. 2010;77(4):1403-1420. doi:10.1111/j.1467-937X.2010.00605.x 44. Deck C, Schlesinger H. Consistency of Higher Order Risk Preferences. Econometrica. 2014;82(5):1913-1943. doi:10.3982/ECTA11396 45. Ebert S, Wiesen D. An Experimental Methodology Testing for Prudence and Third-Order Preferences. Bonn Econ Discussion Papers; 2009. Accessed March 5, 2022. https://www.econstor.eu/handle/10419/37027 46. Ebert S, Wiesen D. Joint Measurement of Risk Aversion, Prudence and Temperance. Bonn Econ Discussion Papers; 2010. Accessed March 5, 2022. https://www.econstor.eu/handle/10419/71891 47. Hey JD, Orme C. Investigating Generalizations of Expected Utility Theory Using Experimental Data. Econometrica. 1994;62(6):1291-1326. doi:10.2307/2951750 48. Schmidt U, Neugebauer T. Testing Expected Utility in the Presence of Errors. The Economic Journal. 2007;117(518):470-485. doi:10.1111/j.1468-0297.2007.02026.x 49. Crainich D, Eeckhoudt L, Trannoy A. Even (Mixed) Risk Lovers Are Prudent. American Economic Review. 2013;103(4):1529-1535. doi:10.1257/aer.103.4.1529 50. Zeckhauser R. Medical insurance: A case study of the tradeoff between risk spreading and appropriate incentives. Journal of Economic Theory. 1970;2(1):10-26. doi:10.1016/0022- 0531(70)90010-4 51. Utility Functions That Depend on Health Status: Estimates and Economic Implications on JSTOR. Accessed March 5, 2022. https://www.jstor.org/stable/2006671?seq=1#metadata_info_tab_contents 52. Guiso L, Jappelli T, Terlizzese D. Income Risk, Borrowing Constraints, and Portfolio Choice. The American Economic Review. 1996;86(1):158-172. 90 53. Lusardi A. On the Importance of the Precautionary Saving Motive. The American Economic Review. 1998;88(2):449-453. 54. Lusardi A. Household Saving Behavior: The Role of Financial Literacy, Information, and Financial Education Programs. National Bureau of Economic Research; 2008. doi:10.3386/w13824 55. Gonzalez R, Wu G. On the Shape of the Probability Weighting Function. Cognitive Psychology. 1999;38(1):129-166. doi:10.1006/cogp.1998.0710 56. Schlander M. Measures of efficiency in healthcare: QALMs about QALYs? Z Evid Fortbild Qual Gesundhwes. 2010;104(3):214-226. doi:10.1016/j.zefq.2010.03.012 57. Garau M, Shah KK, Mason AR, Wang Q, Towse A, Drummond MF. Using QALYs in cancer: a review of the methodological limitations. Pharmacoeconomics. 2011;29(8):673- 685. doi:10.2165/11588250-000000000-00000 58. Rawles J. Castigating QALYs. J Med Ethics. 1989;15(3):143-147. doi:10.1136/jme.15.3.143 59. Dougherty CJ. Quality-adjusted life years and the ethical values of health care. Am J Phys Med Rehabil. 1994;73(1):61-65. 60. Hirskyj P. QALY: an ethical issue that dare not speak its name. Nurs Ethics. 2007;14(1):72- 82. doi:10.1177/0969733007071360 61. Whitehead SJ, Ali S. Health outcomes in economic evaluation: the QALY and utilities. British Medical Bulletin. 2010;96(1):5-21. doi:10.1093/bmb/ldq033 62. Phelps CE, Cinatl C. Estimating optimal willingness to pay thresholds for cost- effectiveness analysis: A generalized method. Health Economics. 2021;30(7):1697-1702. doi:https://doi.org/10.1002/hec.4268 63. Vanness DJ, Lomas J, Ahn H. A Health Opportunity Cost Threshold for Cost-Effectiveness Analysis in the United States. Ann Intern Med. Published online November 3, 2020. doi:10.7326/M20-1392 64. Ryen L, Svensson M. The Willingness to Pay for a Quality Adjusted Life Year: A Review of the Empirical Literature. Health Econ. 2015;24(10):1289-1301. doi:10.1002/hec.3085 65. Allen JD, Abuelezam NN, Rose R, Fontenot HB. Factors associated with the intention to obtain a COVID-19 vaccine among a racially/ethnically diverse sample of women in the USA. Translational Behavioral Medicine. 2021;11(3):785-792. doi:10.1093/tbm/ibab014 91 66. Zamarro G, Prados MJ. Gender differences in couples’ division of childcare, work and mental health during COVID-19. Rev Econ Household. 2021;19(1):11-40. doi:10.1007/s11150-020-09534-7 67. Keeney RL, Raiffa H, Meyer RF. Decisions with Multiple Objectives: Preferences and Value Trade-Offs. Cambridge university press; 1993. 68. Ebert S, van de Kuilen G. Measuring multivariate risk preferences. Available at SSRN 2637964. Published online 2015. 92 Chapter 3 Tables Table 3.1 List of Choice Tasks Notes: x_y represents a 50/50 gamble. Selecting the gamble suggests relative risk aversion and relative prudence. The gamble expected value is 46.5, 53.5, 54.5, 51.5, and 51.5 in order. Task 2 is designed to test if respondents respond in line with CRRA, and Task 3 is designed to test if respondents follow CARA. Risk Aversion or Prudence? Task Decision Certain Outcome Gamble 1 23 22_71 2 28 22_71 3 33 22_71 4 38 22_71 5 43 22_71 6 48 22_71 7 53 22_71 8 58 22_71 9 63 22_71 1 26 25_82 2 32 25_82 3 38 25_82 4 44 25_82 5 50 25_82 6 56 25_82 7 62 25_82 8 68 25_82 9 74 25_82 1 31 30_79 2 36 30_79 3 41 30_79 4 46 30_79 5 51 30_79 6 56 30_79 7 61 30_79 8 66 30_79 9 71 30_79 1 32 31_72 + (5_-5) 2 36 31_72 + (5_-5) 3 40 31_72 + (5_-5) 4 44 31_72 + (5_-5) 5 48 31_72 + (5_-5) 6 52 31_72 + (5_-5) 7 56 31_72 + (5_-5) 8 60 31_72 + (5_-5) 9 64 31_72 + (5_-5) 1 30 74_29 + (5_-5) 2 35 74_29 + (5_-5) 3 39 74_29 + (5_-5) 4 43 74_29 + (5_-5) 5 48 74_29 + (5_-5) 6 52 74_29 + (5_-5) 7 57 74_29 + (5_-5) 8 61 74_29 + (5_-5) 9 66 74_29 + (5_-5) 1 2 3 4 5 Risk Aversion Risk Aversion Risk Aversion Risk Prudence Risk Prudence 93 Table 3.2 Descriptive Statistics by Treatment Condition Total Bedridden Weeks Mobility Cognition Vision Cash Observations 1,202 238 237 240 239 247 Age (Years) Mean 56.2 54.5 58.3 56.3 55.9 56.2 Min 21 21 21 25 22 22 Max 90 90 90 86 87 89 SD 15.4 15.9 14.7 14.5 16.3 15.3 Median 58 56 61 57 60 58 Age Category Less than 30 11% 12% 5% 15% 10% 13% More than 30 89% 88% 95% 85% 90% 87% Gender Female 52% 51% 45% 52% 55% 55% Race White 77% 84% 64% 81% 72% 85% Black or African American 14% 13% 18% 10% 18% 8% Asian 7% 2% 11% 7% 8% 4% Pacific Islander or Native American 2% 1% 5% 3% 2% 2% Other 0% 0% 1% 0% 0% 0% Ethnicity Non-Hispanic 83% 80% 84% 87% 82% 81% Hispanic 17% 20% 16% 13% 18% 19% Visual Analog Scale Health 90-100 27% 27% 21% 28% 29% 29% 80-89 25% 23% 29% 21% 26% 28% 70-79 18% 20% 12% 21% 16% 17% 60-69 9% 9% 9% 7% 9% 9% 50-59 10% 10% 11% 11% 8% 9% 0-49 12% 10% 17% 12% 13% 8% 94 Table 3.3 Descriptive Statistics by Treatment Condition Treatment Conditions Sample Size Stakes Min & Max Relative Prudence Zero-Mean Risk Health Score 1,202 Health Score 0 - 100 5 Bedridden Weeks 238 Proportion of Year Bedridden 0 - 100 5% Mobility 237 Feet in 6 Minute Walking Test 100 - 5500 275 ft Cognition 240 Time to Complete Trail Making Test 50 - 275 13.75 seconds Vision 239 Snellen Eye Test 20/200 - 20/10 10 vision points Cash 247 Financial Based Lottery 0 - $100,000 $5,000 95 Table 3.4 Prevalence of Relative Risk Aversion and Relative Prudence Relative Risk Aversion Prevalence Threshold Health Score Bedridden Weeks Mobility Cognition Vision Cash Task 1 >4 4.53 *,+ 4.4 4.90 + 4.69 + 4.98 6.28 ***,+ Task 2 >4 4.71 + 4.17 4.65 + 4.56 *,+ 5.11 + 5.88 ***,+ Task 3 >4 4.81 *,+ 4.35 + 4.48 + 4.62 + 5.29 + 5.99 ***,+ Task 4 >4, high second gamble 4.57 + 4.34 5.24 ***,+ 4.89 *,+ 5.36 + 6.39 ***,+ Task 5 >4, low second gamble 5.02 ***,+ 4.47 + 5.09 ***,+ 5.22 **,+ 5.44 *,+ 6.00 ***,+ Notes: Prevalence here is reported as mean. Each value is mean number of risk averse or prudent choices selected per nine decisions per task. Wilcoxon sign rank test was conducted to test if mean prevalence is significantly different from the number of certain outcomes greater than the expected value of the gamble, 4, and from random choice, 4.5. Significance for number of certain outcome threshold is reported as + at the 1% level. Significance for different than random choice is reported as * at the 10% level, ** at the 5% level, and *** at the 1% level 96 Table 3.5 Mean Approximated Certainty Equivalents Expected Values Health Score Bedridden Weeks Mobility Cognition Vision Cash Task 1 (Baseline RRA) 46.5 43.03 * 43.94 42.00 * 43.11 * 41.84 35.84 * Task 2 (CRRA RRA) 53.5 48.89 *,+ 52.17 *,+ 50.03 *,+ 50.74 * 47.81 * 43.47 *,+ Task 3 (CARA RRA) 54.5 49.63 *,+ 51.99 * 51.77 * 51.35 * 48.36 *,+ 45.23 * Task 4 51.5 47.77 * 48.91 * 46.02 * 47.29 * 45.66 * 42.02 * Task 5 51.5 45.59 * 48.18 * 46.07 * 45.70 * 44.83 * 42.66 * Notes: Significance attached to number signifies if mean approximated certainty equivalents are significantly different from the expected value of the gamble associated with the decision tasks. Significance in the bottom two rows corresponds with if Task 2 or Task 3 values satisfy CRRA or CARA. T-tests were used to calculate significance. Task 4 and Task 5 correspond to relative risk prudence questions. Significance is reported as * the 5% level among testing if mean approximation was different from expected value of gamble and reported as + for significance testing of CRRA and CARA. RRA = Relative Risk Aversion 97 Table 3.6 Correlation Between Relative Risk Aversion and Relative Prudence A Correlations Between Tasks (Health Score Condition) Task 1 RRA Task 2 RRA Task 3 RRA Task 4 RRP Task 2 RRA 0.64 * Task 3 RRA 0.66 * 0.73 * Task 4 RRP 0.54 * 0.64 * 0.63 * Task 5 RRP 0.51 * 0.52 * 0.52 * 0.61 * B RRA Correlations (Task 1) Bedridden Weeks Mobility Cognition Vision Cash Health Score 0.47 * 0.42 * 0.41 * 0.31 * 0.30 * C RRP Correlations (Task 4) Bedridden Weeks Mobility Cognition Vision Cash Health Score 0.52 * 0.48 * 0.42 * 0.36 * 0.22 * Notes: Spearman rank correlation coefficients reported. * Indicates significance at the 1% level. Tasks 1, 2, and 3 represent relative risk aversion apportionment tasks. Tasks 4 and 5 represent relative risk prudence apportionment tasks. RRA = Relative Risk Aversion, RRP = Relative Risk Prudence 98 Table 3.7 Relative Prudence by Risk Averse Choices among Health Score Responses Number of Risk Averse Choices n Mean Different from Random Choice Different from next lower category 0 n = 217 1.91 p = 0 imprudent 1 n = 64 1.56 p = 0 p = 0.00 2 n = 60 3.48 p = 0 p = 0.00 3 n = 164 4.42 p = 0 p = 0.00 4 n = 123 4.26 p = .58 p = 0.00 5 n = 143 5.41 p = 0 p = 0.00 6 n = 69 6.25 p = 0 p = 0.01 7 n = 37 6.80 p = 0 p = 0.32 8 n = 15 4.86 p = .11 p = 0.40 9 n = 251 6.61 p = 0 p = 0.00 Notes: Wilcoxon sum rank test was used to test if prevalence of prudent responses per number of risk averse choices answered in the first health score question is equivalent to 4.5, which would indicate random selections, where * indicates significance at 1%. Wilcoxon sum rank tests were also used to test if prevalence of prudent selections were similar per prevalence of risk averse choices, where a indicates significance at 1%. 99 Table 3.8 Identifying Demographic Correlates with Relative Risk Aversion and Relative Prudence Health Score Task 1 Health Score Task 4 Coefficient P-Value Coefficient P-Value Age -0.01 0.59 0.01 0.62 Gender Female Ref. Male 0.19 0.63 -0.17 0.68 Born in US -0.28 0.73 -0.18 0.83 Marital Status Married Ref. Divorced -0.48 0.62 -0.87 0.35 Widowed -0.42 0.68 -0.59 0.56 Never Married -1.10 0.27 -1.21 0.21 Live with Partner 0.36 0.47 -0.52 0.27 Education 9th Grade Education 1.62*** 0.00 -6.61*** 0.00 10th Grade Education 1.76* 0.06 -4.51** 0.01 11th Grade Education 1.80 0.36 0.36 0.85 12th Grade - No Diploma 1.94 0.77 -1.03 0.61 High School Graduate or GED 1.31 0.11 -1.46 0.28 Some College 1.32 0.11 -1.65 0.22 Associates Degree 1.39** 0.03 -1.56 0.27 Associates Degree + 1.42 0.14 -0.58 0.68 Bachelor's Degree 1.32** 0.04 -1.18 0.38 Master's Degree 1.32** 0.04 -1.15 0.38 Professional School 2.01** 0.03 -1.47 0.54 Doctorate Degree 1.77 0.30 -0.71 0.70 Race and Ethnicity Hispanic or Latino 0.11 0.49 0.30 0.66 White 0.34 0.34 0.60 0.48 Black -0.26 0.74 -0.84 0.34 Native American 0.40 0.60 0.05 0.95 Asian -0.04 0.53 0.77 0.45 Pacific Islander -0.15 0.53 -0.01 0.99 100 Work Status Working 0.15 0.81 0.55 0.36 On Sick Leave 2.71* 0.09 2.48 0.20 Retired 0.51 0.43 0.37 0.55 Disabled -0.40 0.52 0.21 0.72 Household Income $5,000 - $7,499 -1.38 0.43 -4.33*** 0.00 $7,500 to $9,999 0.75 0.60 -1.59 0.20 $10,000 to $12,499 -0.87 0.47 -0.93 0.41 $12,500 to $14,999 -0.81 0.49 -0.22 0.83 $15,000 to $19,999 -0.68 0.50 -0.44 0.66 $20,000 to $24,999 -0.70 0.52 -1.81* 0.07 $25,000 to $2,999 0.49 0.64 -1.66 0.11 $30,000 to $34,999 0.50 0.64 -0.24 0.81 $35,000 to $39,999 0.18 0.88 -0.99 0.41 $40,000 to $49,999 1.38 0.23 -0.67 0.53 $50,000 to $59,999 -0.39 0.73 -1.94* 0.06 $60,000 to $74,999 0.63 0.57 -0.93 0.39 $75,000 to $99,999 -0.93 0.40 -1.36 0.18 $100,000 to $149,999 0.25 0.82 -0.36 0.73 $150,000 or more -0.52 0.69 -1.09 0.38 Number in Household 1 -0.23 0.62 0.46 0.34 2 -0.43 0.50 0.13 0.84 3 0.21 0.82 1.62 0.07 4 0.02 0.99 0.60 0.63 5 1.17 0.31 0.74 0.58 7 -2.53* 0.05 -3.06** 0.02 Treatment Group Mobility Ref. Cash 0.32 0.56 0.28 0.60 Cognition -0.10 0.85 0.20 0.72 Bedridden Weeks -0.38 0.49 0.00 1.00 Vision 0.16 0.78 0.56 0.32 Section Order -0.08 0.84 -0.12 0.75 VAS Category 90-100 Ref. 101 80-89 -0.51 0.34 -0.03 0.96 70-79 0.40 0.45 0.45 0.41 60-69 -0.43 0.53 -0.29 0.69 50-59 0.11 0.88 0.38 0.59 0-49 0.09 0.89 -0.33 0.61 Notes: Ordinary least squares regression were ran with demographic attributes of our respondents and robust standard errors, for the purpose of identifying demographic correlates with prevalence of relative risk aversion and relative risk prudence. */**/*** indicates significance of coefficient or demographic variables at 10%, 5%, and 1% significance. 102 Table 3.9 A Parametric Estimates of Relative Risk Aversion, Relative Risk Prudence, and Relative Risk Temperance under Expected Utility, Constant Relative Risk Aversion and Hyperbolic Absolute Risk Aversion (Simplified) CRRA HARA (simplified) Utility Function 𝑥 1 − 𝜌 1 − 𝜌 1 − 𝛾 𝛾 ⁡ ( 𝑥 + 𝜂 ) 𝛾 Constraints − 1 < 𝜌 < 1 0 < 𝛾 < 1 0 < 𝜂 Coefficient Estimation, 𝜌 = 0.23 𝛾 = ⁡0.29 Mean, 95% CI 0.19 – 0.29 0.26 – 0.31 𝜂 =447,776 -10,035 – 905,587 Coefficient = 0, p-value 𝜌 ; ⁡<0.01 𝜂 ; ⁡<0.01 𝛾 ; ⁡<0.01 RMSE 0.094 0.122 Higher Order Estimates, Mean (Median), 95% CI Relative Risk Aversion 0.23 (0.56) 0.22 (0.21) 0.19 – 0.29 0.21 – 0.23 Relative Prudence 1.23 (1.56) 0.69 (0.80) 1.19 – 1.28 0.66 – 0.72 Relative Temperance 2.23 (2.56) 1.16 (1.20) 2.19 – 2.28 1.10 – 1.21 Notes: x = 0.475 for calculating higher order risk estimates. 0.475 was selected as the average linear approximated certainty equivalent across all respondents, and thus was used as the argument for calculating higher order risk estimates. Estimates are reported as Mean (Median). Wilcoxon sign-rank tests were conducted to test if coefficients were greater than 0. P-values < 0.01 are reported as 0.01. 𝜔 𝐻 = 0 . 24 ⁡ ( 95% ⁡ 𝐶𝐼 ⁡0 . 23 − 0 . 27 ) for HARA (simplified). CRRA = Constant Relative Risk Aversion; HARA = Hyperbolic Absolute Risk Aversion; CI = Confidence Interval; RMSE = Root Mean Squared Error 103 Table 3.9 B Parametric Estimates of Relative Risk Aversion, Relative Risk Prudence, and Relative Risk Temperance under Expected Utility, Hyperbolic Absolute Risk Aversion (Expanded) HARA (expanded) HARA (expanded) HARA (expanded) Utility Function 1 − 𝛾 𝛾 ⁡ ( 𝛼𝑥 1 − 𝛾 + 𝜂 ) 𝛾 1 − 𝛾 𝛾 ⁡ ( 𝛼𝑥 1 − 𝛾 + 𝜂 ) 𝛾 1 − 𝛾 𝛾 ⁡ ( 𝛼𝑥 1 − 𝛾 + 𝜂 ) 𝛾 Constraints 𝛼 = 1 0 < 𝛾 < 1 0 < 𝛾 < 1 1 = 𝛼 0 < 𝛼 0 < 𝜂 0 < 𝜂 Coefficient Estimation, 𝛾 = ⁡0.99 𝛾 = ⁡0.48 𝛾 = ⁡0.47 Mean, 95% CI 0.96 – 1.00 0.47 – 0.50 0.45 – 0.49 𝜂 = 11,010,491 𝜂 = 0.19 𝜂 = 0.12 8,958,158 – 13,100,000 0.16 – 0.22 0.09 – 0.15 𝛼 =1 𝛼 =1 𝛼 =1.23 1.15 – 1.33 Coefficient = 0, p-value 𝛾 ; <0.01 𝛾 ; <0.01 𝛾 ; <0.01 𝜂 ; <0.01 𝜂 ; <0.01 𝜂 ; <0.01 RMSE 0.126 >100,000* 0.588 Higher Order Estimates, Mean (Median), 95% CI Relative Risk Aversion 0.03 (0.00) 0.45 (0.40) 0.50 (0.39) 0.03 – 0.04 0.44 – 0.46 0.48 – 0.52 Relative Prudence 0.41 (0.00) 1.35 (1.39) 1.45 (1.37) 0.36 – 0.46 1.34 – 1.36 1.43 – 1.48 Relative Temperance 0.78 (0.00) 2.25 (2.36) 2.40 (2.35) 0.69 – 0.88 2.23 – 2.27 2.37 – 2.43 Notes: x = 0.475 for calculating higher order risk estimates. 0.475 was selected as the average linear approximated certainty equivalent across all respondents, and thus was used as the argument for calculating higher order risk estimates. Estimates are reported as Mean (Median). Wilcoxon sign-rank tests were conducted to test if coefficients were greater than 0. P-values < 0.01 are reported as 0.01. *Median RMSE for HARA expanded with two constraints and 𝛼 = 1 is 0.36. Outliers are greatly skewing this estimate. HARA = Hyperbolic Absolute Risk Aversion; CI = Confidence Interval; RMSE = Root Mean Squared Error 104 Table 3.9 C Parametric Estimates of Relative Risk Aversion, Relative Risk Prudence, and Relative Risk Temperance under Expected Utility, Power Expo (Holt & Laury) Power-Expo Power-Expo Power-Expo Power-Expo Utility Function 1 − e x p ⁡ ( − 𝛼 𝑥 1 − 𝜇 ) 𝛼 1 − e x p ⁡ ( − 𝛼 𝑥 1 − 𝜇 ) 𝛼 1 − e x p ⁡ ( − 𝛼 𝑥 1 − 𝜇 ) 𝛼 1 − e x p ⁡ ( − 𝛼 𝑥 1 − 𝜇 ) 𝛼 Constraints no constraints 𝛼 > 𝜇 𝛼 > 0 − 1 < 𝛼 < 1 𝛼 ≠ 0 − 1 < 𝜇 < 1 Coefficient Estimation, 𝛼 = ⁡85,378 𝛼 = ⁡-0.10 𝛼 = ⁡73,272 𝛼 = ⁡0.31 Mean, 95% CI 45,012 – 125,744 -0.14 - -0.06 40,087 – 106,458 0.26 – 0.37 𝜇 = -5.35 𝜇 = 0.11 𝜇 = -2.29 𝜇 = ⁡0.16 -6.81 - -3.89 0.07 – 0.15 -2.51 - -2.05 0.11 – 0.21 Coefficient = 0, p-value 𝛼 ; ⁡<0.01 𝛼 = 0 . 29 𝛼 ; ⁡<0.01 𝛼 ; ⁡<0.01 𝜇 ; ⁡<0.01 𝜇 ; ⁡<0.01 𝜇 ; ⁡<0.01 𝜇 ; ⁡<0.01 RMSE 0.414 0.463 0.528 0.283 Higher Order Estimates, Mean (Median), 95% CI Relative Risk Aversion 56.63 (0.26) 0.06 (0.03) 49.28 (0.21) 0.23 (0.74) 24.36 – 88.91 0.02 – 0.10 23.23 – 75.32 0.18 – 0.29 Relative Prudence -73.69 (-2.91) 0.79 (0.67) -58.60 (-2.91) 0.83 (1.44) -106.70 - -40.68 0.70 – 0.88 -85.31 - -31.89 0.54 – 1.12 Relative Temperance 65.19 (0.03) 1.05 (0.39) 56.83 (1.44) 1.14 (1.85) 25.53 – 104.85 0.97 – 1.12 24.81 – 88.85 0.91 – 1.37 Notes: x = 0.475 for calculating higher order risk estimates. 0.475 was selected as the average linear approximated certainty equivalent across all respondents, and thus was used as the argument for calculating higher order risk estimates. Estimates are reported as Mean (Median). Wilcoxon sign-rank tests were conducted to test if coefficients were greater than 0. P-values < 0.01 are reported as 0.01. 𝜔 𝐻 = 0 . 24 ⁡ ( 95% ⁡ 𝐶𝐼 ⁡0 . 23 − 0 . 27 ) for HARA (simplified) and 𝜔 𝐻 = 0 . 45 ⁡ ( 95% ⁡𝐶𝐼 ⁡0 . 43 − 0 . 47 ) for HARA (expanded). CI = Confidence Interval; RMSE = Root Mean Squared Error 105 Table 3.9 D Parametric Estimates of Relative Risk Aversion, Relative Risk Prudence, and Relative Risk Temperance under Expected Utility, Expo-Power (Saha) Expo-Power Expo-Power Utility Function Θ − { 𝛽 𝑥 𝛼 } Θ − { 𝛽 𝑥 𝛼 } Constraints 𝑛𝑜 ⁡ 𝑐 𝑜𝑛𝑠 𝑡 𝑟 𝑎 𝑖 𝑛𝑡 𝑠 𝛽 < 𝛼 𝛼 ≠ 0 Coefficient Estimation, 𝛽 = ⁡-104,902.5 𝛽 = ⁡-0.14 Mean, 95% CI -126,355 - -83,449 -0.20 - -0.09 𝛼 = -0.32 𝜇 = -0.72 -0.56 - -0.07 -0.83 - -0.60 Coefficient = 0, p-value 𝛽 ; < 0 .01 𝛽 ; < 0 .01 𝛼 ; < 0 .01 𝛼 ; < 0 .01 RMSE 0.080 0.169 Higher Order Estimates, Mean (Median), 95% CI Relative Risk Aversion 117.1 (0.80) -1.25 (1.00) -27.30 – 261.41 -2.04 - -0.46 Relative Prudence 117.5 (1.80) 0.63 (2.00) -26.7 – 261.68 -0.38 – 1.64 Relative Temperance 117.9 (2.80) 1.92 (3.00) -26.1 – 261.9 0.91 – 2.93 Notes: x = 0.475 for calculating higher order risk estimates. 0.475 was selected as the average linear approximated certainty equivalent across all respondents, and thus was used as the argument for calculating higher order risk estimates. Estimates are reported as Mean (Median). Wilcoxon sign-rank tests were conducted to test if coefficients were greater than 0. P-values < 0.01 are reported as 0.01. CRRA = Constant Relative Risk Aversion; HARA = Hyperbolic Absolute Risk Aversion; CI = Confidence Interval; RMSE = Root Mean Squared Error 106 Table 3.10 Distribution of Higher Order Risk Estimates over Health Score and Cash – Relative Risk Aversion, Relative Prudence, and Relative Temperance Percentile Treatment Condition Higher Order Estimates 5th 10th 25th Median 75th 90th 95th Mean Health Score Relative Risk Aversion 0.01 0.04 0.07 0.21 0.30 0.40 0.47 0.22 Relative Prudence 0.03 0.08 0.14 0.80 1.20 1.26 1.28 0.69 Relative Temperance 0.05 0.12 0.22 1.20 2.20 2.26 2.27 1.16 Cash Relative Risk Aversion 0.03 0.20 0.22 0.24 0.24 0.24 0.25 0.22 Relative Prudence 0.07 1.20 1.22 1.24 1.24 1.24 1.25 1.16 Relative Temperance 0.10 2.20 2.22 2.24 2.24 2.24 2.25 2.10 Notes: Higher order risk estimates for relative risk aversion, relative risk prudence, and relative risk temperance were made using the simplified form of the hyperbolic, absolute risk aversion function. Percentile estimates here are provided for both health score and cash treatment conditions. Bolded values are median and mean estimates of higher order risk parameters. 107 Table 3.11 Ratio of Higher Order Risk Estimates over Health Score relative to Cash With Outliers Without Outliers Higher Order Estimates Ratios Mean (Median), 95% CI Relative Risk Aversion 1.35 (1.03) 0.98 (1.03) 0.77 – 1.93 0.89 – 1.06 Relative Prudence 56.16 (0.64) 0.67 (1.01) 3.94 – 108.38 0.55 – 0.80 Relative Temperance 48.06 (0.53) 0.62 (0.53) 2.80 – 93.13 0.50 – 0.74 Notes: Ratios are calculated as higher order parameter estimate for health score divided by higher order parameter estimate for financial-based lottery. “With Outliers” includes every observation. “Without Outliers” excludes observations below the 5 th percentile and above the 95 th percentile. 108 Chapter 3 Figures Figure 3.1 Aversion Task Notes: During the relative risk aversion task, a respondent must select between a gamble with either X or Y outcomes or a guaranteed outcome, as defined by CO (Certain Outcome). Figure 3.2 Relative Prudence Task Notes: During the relative prudence task, a respondent must select between a staggered gamble, with X and Y outcomes, and a zero-mean risk (Z) gamble tied to the Y outcome. During these questions, X and Y vary in terms of a high or low health or wealth stat 109 Figure 3.3 Histogram of Risk Averse and Prudent Selections across Health Score Questions Task 1 Responses Task 2 Responses Task 3 Responses Task 4 Responses Task 5 Responses Notes: Histograms display density of individuals’ risk averse and prudent selections, with 0 being no certain outcomes selected and 9 being all certain outcomes selected. 110 Figure 3.4 Cumulative Distribution Functions (CDFs) of Responses on a Representative Risk Averse and Risk Prudent Questions Task 1. For each treatment condition besides health score Task 4. For each treatment condition besides health score Notes: CDFs demonstrate prevalence of risk averse (certain outcome) question selections per treatment conditions, with 0 being no certain outcomes selected and 9 being all certain outcomes selected. 111 Figure 3.4 Distribution of Health Score Higher Order Parameter Estimates for Relative Risk Aversion, Relative Risk Prudence, and Relative Risk Temperance A. B. C. Notes: The above figures contain the distribution of individually estimated relative risk aversion, relative prudence, and relative temperance. A corresponds to relative risk aversion, B corresponds to relative prudence, and C corresponds to relative temperance. HA_RA = Hyperbolic Absolute Risk Aversion, Estimated Relative Risk Aversion; HA_RT = Hyperbolic Absolute Risk Aversion, Estimated Relative Risk Temperance; HA_RP = Hyperbolic Absolute Risk Aversion, Estimated Relative Risk Prudence 112 Figure 3.5 Histogram of Relative Risk Aversion over Health Score and over Cash A. With End Estimates B. Without End Estimates Notes: Histograms of higher order estimates of relative risk aversion among health score and treatment conditions are plotted here with A) end estimates falling below the 5 th percentile and above the 95 th percentile and B) without end estimates falling below the 5 th percentile and 95 th percentile. 113 3.7 Appendix Appendix Note 3.1 Instructions for Understanding Study Participants In this section, we will present you with hypothetical scenarios regarding your health. Since these scenarios can be complex, please read each question carefully. Throughout the survey, we will reference your health as a numbered score. Your health score will range from 0 to 100. A health score of 0 is equivalent to death while a health score of 100 is perfect health. On a scale of 0-100, how would you rate your health today? Please click anywhere on the slider below or use the textbox to enter your answer. <Next Page> For reference, the scale below indicates a few conditions and how someone may view their corresponding health scores. Please answer the next question assuming the following about your health: A few years ago, your health severely deteriorated due to an unknown cause. Doctors could not 114 identify the cause and could not do anything to improve your health. Today, your health score is a 20. Prior to your health’s deterioration, your health score was 100. How would you view a health score of 20 on a scale from 0 to 100? <Next Page> Good news! Recent medical breakthroughs have allowed doctors to discover the cause of your health deterioration. Not only that, there are now two medical treatments available. Both treatments guarantee that the condition will be completely cured in one year, returning your health score to 100, where you can expect to live for 30 more years at this health level. However, the treatments will have varying effects during the coming year of treatment. For the next set of questions, you will be shown the potential outcomes resulting from your two medical treatment options. Treatment A will have a "certain outcome," in other words, a guaranteed health score that will result if you select the treatment A option. Treatment B will have varying outcomes, dependent on if you are a Type I patient or a Type II 115 patient. You have an equal chance of being a Type I patient or a Type II patient, but you will not know which patient type you are until after you have completed your treatment. The goal of each task in this section is to find the switch point at which you would consider a certain outcome over the treatment with two random outcomes occurring at a 50% likelihood. See the picture below for example. To find this switch point, we will ask you to select between one option (on the left side) or another option (on the right side). <Next Page> For the next set of questions, please select if you would rather have Treatment A or Treatment B. Treatment A will result in one certain outcome while Treatment B will result in varying 116 outcomes depending on if you are a Type I or Type II patient. Assume for each of the following treatment decisions (each row), that is the one decision you will make regarding your received treatment for the next year. Each decision is mutually exclusive, meaning that one decision has no impact on subsequent or previous decisions. For example, in row 1, after your health condition of 20 health points, you have the option between Treatment A that gives you a certain outcome of 23 points or Treatment B that will either give you 22 points or 71 points at a 50/50 chance. You would select the bubble for Treatment A if you prefer the certain outcome or the bubble for Treatment B if you prefer the treatment where you may have a health of 22 or 71 after treatment at a 50/50 chance. For the next decision (row 2), again you would start from your beginning 20 health points. You now have the option between a "Treatment A" which now gives a certain outcome of 28 health points or "Treatment B" which still can give you 22 points or 71 at a 50/50 chance. As a reminder, you have a health score of 20 prior to receiving either treatment. The scale of health scores is included below for reference. See scale above for scale Please indicate the treatment you would select in each scenario given the outcome results below. 117 <Next Page> Subsequent tasks 2 and 3 had the following question prompts with a similar figure and decision display as above. Values are included in Table 1. Please indicate the treatment you would select in each scenario given the outcome results below. As a reminder, you have a health score of 20 prior to receiving either treatment. Each treatment decision you make has no impact on the other treatment outcomes. For example, your decision for row 1 has no impact on your outcome for row 2, and so on and so forth. However, you must make a decision for each row. <Next Page> Prudent questions had the following instructions and question prompts. Now assume the following: 118 Treatment A will result in one certain outcome while Treatment B will result in varying outcomes depending on if you are a Type I or Type II patient. Type II patients must now also take a pill that will improve or worsen their health score by 5 points, within the next year. Whereas, Type I patients do not have to take the pill. You have an equal chance of being a Type I patient or a Type II patient, but you will not know which patient type you are until after you have completed your treatment. For Type II patients, the pill has an equal chance of improving or worsening your health score, chances being 50%. Again, both treatments guarantee that the condition will be completely cured in one year, returning your health score to 100, where you can expect to live for 30 more years at this health level. As a reminder, you have a health score of 20 prior to receiving either treatment. The scale of health scores is included below for reference. See scale above for scale 119 Each treatment decision you make has no impact on the other treatment outcomes. For example, your decision for row 1 has no impact on your outcome for row 2, and so on and so forth. However, you must make a decision for each row. Please indicate the treatment you would select in each scenario given the outcome results below. <Next Page> Subsequent tasks 5 had the following question prompts with a similar figure and decision display as above. Values are included in Table 1. Please indicate the treatment you would select in each scenario given the outcome results below. 120 As a reminder, you have a health score of 20 prior to receiving either treatment. Each treatment decision you make has no impact on the other treatment outcomes. For example, your decision for row 1 has no impact on your outcome for row 2, and so on and so forth. However, you must make a decision for each row. Note: Questions were scaled and tailored to the treatment condition of interest. 121 Appendix Table 3.1 Hyperbolic Absolute Risk Aversion Estimates Across Treatment Conditions Health Bedridden Weeks Mobility Cognition Vision Cash Higher Order Estimates, Mean (Median) Relative Risk Aversion 0.22 (0.21) 0.21 (0.23) 0.22 (0.23) 0.20 (0.23) 0.19 (0.22) 0.22 (0.24) Relative Prudence 0.69 (0.8) 1.13 (1.23) 1.14 (1.23) 1.09 (1.23) 1.05 (1.22) 1.16 (1.24) Relative Temperance 1.16 (1.2) 2.06 (2.23) 2.05 (2.23) 1.97 (2.23) 1.91 (2.22) 2.10 (2.24) 122 Appendix Table 3.2 Impact of Higher Order Estimates on Willingness-to-Pay Thresholds Median Mean Estimates of K/M 1.32 1.39 WTP @ GDP per Capita $83,611 $88,256 Notes: Phelps & Cinatl in 2020, derived the following equation to estimate optimal willingness- to-pay thresholds (WTP), based on higher order risk preferences. Using the risk preferences percentiles detailed above and the following equation detailed in Phelps & Cinatl (2020), 𝐾 𝑀 = ⁡ 1 𝜋 ∗ − 2 𝑟 ∗ ; optimal WTP thresholds were estimated using relative risk aversion and relative prudence estimated above multiplied by United States’ gross domestic product (GDP) per capita ($63,544). For reference, mean 𝜔 𝑐 = 0 . 72 , ⁡95% Confidence Interval 0 . 70 − 0 . 75 , and median was found to be 0 . 76. 123 Appendix Figure 3.1 Distribution of Hyperbolic Absolute Risk Aversion Coefficient Estimates Notes: The above figure provides a histogram of gamma estimates, per respondent. 124 Chapter 4: Cabazitaxel vs. an Alternative Androgen-Signaling Target Inhibitor in Metastatic, Castrate-Resistant Prostate Cancer Patients Previously Treated with Docetaxel and an Alternative Androgen-Signaling Target Inhibitor: Applications of the Novel Generalized Risk-Adjusted Cost-Effectiveness (GRACE) Framework Samuel A. Crawford, Darius N. Lakdawalla, Jason N. Doctor, Mitchell E. Gross, & William V. Padula Abstract Objectives: Metastatic, castrate-resistant prostate cancer is one of the leading causes of death among men that have cancer. Treatment sequencing is an important factor in determining effective treatment. Given the severity of mCRPC and the treatment tradeoff between tolerable therapies and extending life expectancy, modeling treatments for mCRPC are a great case study for an early application of the generalized risk-adjusted cost-effectiveness analysis. In this study, we aimed to identify cost-effectiveness using the classic incremental cost-effectiveness ratio and generalized risk-adjust cost-effectiveness framework (GRACE). Methods: We analyzed cost-utility of an alternative androgen-signaling target inhibitor (ASTi; abiraterone or enzalutamide) relative to cabazitaxel in patients treated with docetaxel and an alternative ASTi (enzalutamide or abiraterone), among metastatic castrate-resistant prostate cancer patients. We used a Markov model, with a 5-year time horizon and 3-week cycles from a US health sector perspective to model this decision problem. Costs, quality-adjusted life years (QALYs), and the incremental GRACE ratio (IGRACER) were used to determine cost- 125 effectiveness. Model uncertainty was evaluated using univariate and probabilistic sensitivity analyses. Results: Both classic, incremental cost-effectiveness ratio and GRACE frameworks at a willingness-to-pay threshold of $100,000/QALY demonstrate cabazitaxel is not a cost-effective alternative to an alternative ASTi in a modeled patient population. However, GRACE demonstrates cabazitaxel can be cost-effective at an assumed willingness-to-pay threshold of $200,000/QALY adjusted for disease severity and risk preference. Incremental costs of cabazitaxel vs. an alternative ASTi was $100,251 and incremental QALYs were 0.16, suggesting cabazitaxel is not cost-effective relative to an ASTi at a willingness-to-pay threshold of $100,000/QALY. IGRACER validated cabazitaxel is not a cost-effective alternative to an alternative ASTi at an adjusted willingness-to-pay thresholds, derived from the classic $100,000/QALY, but is cost-effective at a willingness-to-pay threshold of $200,000/QALY. Findings were robust to sensitivity and scenario analyses. Conclusions: Cabazitaxel is not a cost-effective treatment alternative to an alternative ASTi among patients with mCRPC, under the classic ICER framework and GRACE at a willingness- to-pay threshold of $100,000. Cabazitaxel extends life-expectancy while greatly increasing costs. Patients and decision-makers should now consider the cost of cabazitaxel and limited incremental survival benefit in patients that were previously treated with an ASTi when deciding between cabazitaxel and ASTi, when previously treated with an alternative ASTi and docetaxel. Keywords: Metastatic Castration-Resistant Prostate Cancer, Generalized Risk-Adjusted Cost- Effectiveness, GRACE, Cancer, Oncology 126 4.1 Introduction Prostate cancer (PC) is one of the leading causes of death among U.S. males. 1 Despite significant advancements in diagnostics and treatment, over 190,000 men develop PC, and 33,000 die each year. 1 Of all male cancers, 21% are PC, which creates a high financial toll and economic impact on United States (US) healthcare. Total costs of PC diagnosis and treatment exceeds $10 billion per year. 2 Individuals with nonmetastatic castration-resistant PC average $35,000 per year, while patients with metastatic castration-resistant PC (mCRPC) face an astounding $155,000 per year per course of treatment. 2,3 MCRPC patients are usually treated initially with docetaxel. Patients may be treated with a hormonal therapy such as, androgen-signaling targeted inhibitor (ASTi) of either abiraterone or enzalutamide, either before or after docetaxel. 4 However, if the disease progresses, decision makers are tasked with prescribing an alternative ASTi (abiraterone or enzalutamide) or another chemotherapeutic agent. However, despite having several potentially effective treatment options, cross-resistance may limit the effect of different treatment sequences. For example, patients may not have a subsequent response to abiraterone or enzalutamide after already having received an ASTi (abiraterone or enzalutamide). 5–8 The CARD clinical trial results, as presented by de Wit et al. (2019) suggests cabazitaxel is effective relative to an alternative ASTi (abiraterone or enzalutamide) in patients with mCRPC, post-docetaxel, post-ASTi (abiraterone or enzalutamide). 4 However effectiveness of cabazitaxel, relative to an alternative ASTi, following docetaxel and an alternative ASTi, still needs to be validated in a real-world setting. Cabazitaxel is an effective alternative to abiraterone and enzalutamide in patients with mCRPC. Despite clinical trial evidence that cabazitaxel is efficacious for mCRPC patients, the field could benefit from a more generalizable, comprehensive economic evaluation of these 127 findings. Furthermore, the higher cost of cabazitaxel can be questioned given the limited marginal clinical outcomes (several months of progression-free and overall survival) it provides over abiraterone or enzalutamide in the treatment of post-docetaxel, post alternative ASTi mCRPC. Finally, patients may favor side-effects of ASTi (abiraterone or enzalutamide) over cabazitaxel. Given the severity associated with mCRPC patient outcomes and the quality-of-life and life expectancy trade-off associated with either receiving cabazitaxel or an alternative ASTi, there exists some limitations with evaluating these treatments using classical cost-effectiveness. The generalized risk-adjusted cost-effectiveness (GRACE) framework developed by Lakdawalla & Phelps (2021) is an alternative of the classical incremental cost-effectiveness approach that allows for dynamic optimal willingness to pay thresholds, inclusion of uncertain treatment outcomes, and greater value associated with an intervention for a severe disease. 9–11 This study represents one of the first applications the GRACE framework for economic evaluation of a medical intervention. Studies evaluating the cost-effectiveness of treatment patterns for mCRPC patients have focused exclusively on the classic incremental cost-effectiveness ratio for evaluating treatments. 12,13 In light of these studies, we expand upon those findings using updated cost estimates from recently published studies and a new database analysis. We also further corroborate findings that suggest cabazitaxel is an alternative-ASTi in this patient population using the novel GRACE framework. 128 4.2 Methods and Study Design Overview of Study Design We developed a clinically valid, Markov model to measure the cost-effectiveness of cabazitaxel and an alternative ASTi for post-docetaxel, post-alternative ASTi, mCRPC patients from the U.S. healthcare sector perspective, in accordance with methods prescribed by the U.S. Panel on Cost-effectiveness in Health and Medicine. 14 Model and parameters were validated with the aid of a clinical expert. Parameters were populated using real-world evidence, a population-based survey, and a targeted literature review. Patients traveled through three model stages, progression-free disease, progressed disease, and death. Cycle lengths were three weeks and the modeled time horizon was five years. Common adverse events observed in the CARD clinical trial were also included in the progressed disease arm. 15 Results were discounted at 3% per year modeled. Cabazitaxel was hypothesized as a cost-effective treatment alternative of mCRPC at willingness-to-pay thresholds of $100,000 per quality-adjusted life year (QALY) and $200,000 per QALY, the demonstrated willingness-to-pay threshold for American medical interventions and around the US gross domestic product per capita multiplied by 3. 16–19 Modeled results were interpreted based on the incremental cost-effectiveness ratio (ICER), net monetary benefit (NMB) at prescribed thresholds, and Generalized Risk-Adjusted Cost-Effectiveness (GRACE) findings. Using the classical ICER model, an intervention was considered cost-effective if it satisfied equation 4.1. 129 Classical cost-effectiveness model, condition at which new technologies are adopted ∆ ⁡ 𝐶𝑜𝑠𝑡 ( Δ 𝑆 ) 𝑄 + 𝑆 ( Δ Q ) ⁡ ≤ 𝐾 Eq. 4.1 • 𝑠 represents baseline life expectancy • ∆ 𝑠 represents the change in life expectancy • 𝑄 is baseline QoL • ∆ 𝑄 ⁡is the change in QoL • Δ 𝐶 represents the incremental cost of the new technology • 𝐾 is the wiliness-to-pay threshold Model Structure and comparators A three-state Markov model (Progression Free Disease, Progressed Disease, and Death) was created with 3-week cycles over a 5-year time horizon. Patients could move freely between each health state, until reaching the terminal node death, or the model time horizon expiring. The Markov model was constructed to compare abiraterone or enzalutamide (alternative ASTi’s) relative to cabazitaxel among metastatic, castration-resistant prostate cancer patients (Figure 4.1). Patient simulations began with presentation of mCRPC, and failed treatment of docetaxel or an alternative ASTi. After presentation, patients were treated with cabazitaxel or abiraterone (if previously treated with enzalutamide) or enzalutamide (if previously treated with abiraterone). 130 Model Parameters Probabilities Following treatment, patients were assumed to exist in the progression free disease state. Patients than progressed to the progressed disease state or death, contingent on transition probabilities translated from rates captured in the CARD clinical trial using the DEALE method (Table 4.1). 15,20–22 Due to severity of progressed disease among third line mCRPC patients, it was assumed progressed disease patients did not transition make to progression-free disease state. Progressed disease patients either remained in the progressed disease stage or transitioned to death stage. Patients existing in the progressed disease state were also monitored as having a high prevalent adverse event that was captured during the CARD trial. Costs A retrospective claims analysis was used to aggregate and identify costs per cycle, per event per patient in this model. A targeted literature review was used to populate costs associated with events and stages of this model (Table 4.1). Model assumed a fixed cost of care, per patient per cycle. Variations in patient outcomes were relegated to adverse events associated with treatment captured in the progressed disease state. Given this model was conducted from the healthcare sector perspective, societal costs such as caregiver burden and patient wait time was not included. Costs in the progression free disease stage consisted of costs associated with cabazitaxel, abiraterone or enzalutamide treatment (among patients in the alternative ASTi arm). Among patients in the alternative ASTi arm, expected cost of treatment in the progressed disease arm was the average of enzalutamide and abiraterone-based treatment regimens. Patient costs in the 131 progressed disease arm were identified based off treatment costs associated with onset of specific clinical events. Supportive care was assumed for all patients in the progressed disease state, regardless of adverse event onset. Routine laboratory tests, scans, and follow-up were aggregated each cycle regardless of stage. Prices were converted to 2021 US dollars, assuming the medical consumer index. 23 Costs were discounted at a rate of 3% a year. 24 Health State Utilities Health state utilities were populated using published literature estimates from multiple studies characterize quality-of-life among patients with metastatic forms of prostate cancer and on either a treatment regimen of cabazitaxel or an alternative ASTi (enzalutamide or abiraterone). 25–27 Health state utility estimates were assumed to be lower in patients with progressed forms of the disease and populated using literature estimates of advanced forms of prostate cancer following treatment with abiraterone, enzalutamide or cabazitaxel. Health state utilities were generated in the reference studies using time trade-off and the EuroQol 5- Dimensions. Each utility value was scaled to cycle length. In the event a patient died mid-year, utilities were scaled to the length of time that the patient lived. For example, if a patient utility for the cabazitaxel arm is 0.6 and that patient only lived for 6 months, then they were assumed to have lived 0.3 QALY. Classic Model Main Outcome Measures Model outcomes were measured in life-years, QALYs, mortality over the course of the model, and total costs. QALYs were scaled life-expectancies to reported health utilities for each health state. ICERs were calculated as additional cost per QALY saved or per life-year saved. 132 Findings were examined using a reference of a lower bound of $100,000/QALY to an upper bound of $200,000/QALY. 18 Sensitivity Analyses Univariate, two-way, and probabilistic sensitivity analyses were conducted to evaluate parameter uncertainty within the model. With univariate sensitivity analyses, parameter values were varied either by published upper or lower bound estimates of values or by ±20% of the original value. Two-way sensitivity analyses were conducted for two parameters that had the largest impact on the model. For probabilistic sensitivity analyses, 10,000 Monte Carlo simulations were conducted, assuming the range used for univariate sensitivity analyses and using beta distributions for transition probabilities, health state utilities, and log-normal distributions for costs. Sensitivity of ICERs and NMBs were examined at willingness-to-pay thresholds of $200,000 per QALY. A cost-effectiveness acceptability curve was generated from Monte Carlo simulations, scaled to varying willingness-to-pay thresholds from $0 to $500,000 per QALY to determine at which point different therapeutic options and sequalae are most cost- effective. Scenario Analyses Multiple scenario analyses were conducted to further validate the robustness of our findings and add to the interpretation of our findings. First, total cancer-related medical costs were abstracted from a claims analysis and populated in this model. Second, costs like those used in previous literature were abstracted and included in this model. Third, a second 3-stage Markov model was added to represent patients that may have been treated with the alternative therapy 133 (cabazitaxel if in the ASTi arm or an ASTi if in the cabazitaxel arm) once progression occurred, as opposed to assuming patients entirely ceased treatment. The number of patients progressing to a subsequent therapy corresponded with published proportions in the CARD clinical trial. Fourth, costs were scaled to 4 cycles for ASTi and 7 cycles for cabazitaxel, following previously published models and the CARD clinical trial. Finally, expected value of perfect information analysis was conducted to determine a hypothetical threshold that would be necessary to identify the patient population best suited for cabazitaxel or an alternative ASTi treatment, among those previously treated with an ASTi and docetaxel. Overview of GRACE Study Design Classical cost-effectiveness is based on the assumptions of 1) risk neutral decision- makers and 2) constant returns to health in the creation of utility. In addition to the prototypical modeling approach, a GRACE model was developed for post-docetaxel, post-alternative ASTi mCRPC patients as a new application of this novel modeling approach (See Appendix Note 1). The GRACE modeling method builds off the classical model but allows for diminishing returns to health-related QoL. 9 The implications of this modeling technique allow for optimal cost- effectiveness thresholds to rise for more severe diseases and fall for milder ones. This allows for novel modeling for mCRPC patients where outlook may be poor and treatment outcome can have uncertainty. Results from our survey will be used to populate novel parameters necessary for GRACE modelling. The GRACE framework consists of a two-period model or multiple period model. A 5- year Markov model, with 3-week cycles will necessitate the multi-period model. Parameters necessary to calculate and necessary for executing GRACE will be presented here in their two- 134 period form to introduce the necessary pieces. However, necessary iterations for the multi-period model will be mentioned following this GRACE parameter introduction. GRACE framework overview ∆ ⁡ 𝐶𝑜𝑠𝑡 ( Δ 𝑆 ) 𝛿 ⁡ + 𝑆 ( Δ Q ) 𝜖 ⁡ ≤ 𝐾 𝜔 𝐻 𝑅 Eq. 4.2 • 𝛿 is the QoL units a patient or decision maker would give up in exchange for 1 more life-year • 𝜖 reflects the change in value associated with uncertain treatment outcomes, otherwise known as relative risk preference • 𝜔 𝐻 describes how utility changes with health related QoL • 𝑅 adjusts the cost-effectiveness threshold for disease severity GRACE Model Outcomes Outcomes will be measured in life-years, risk-adjusted QALYs, and costs will be captured in 2021 USD. Key GRACE outcomes will include the incremental-GRACE-ratio (IGRACER), GRACE-willingness-to-pay thresholds at $100,000 per QALY and $200,000 per QALY, total value of medical intervention, and univariate sensitivity analyses to gauge which parameters have the largest impact on GRACE outcomes. Assumptions Several assumptions were critical to the execution of this model. First, patients who were treated with an alternative ASTi (abiraterone or enzalutamide) after docetaxel and an ASTi (enzalutamide or abiraterone) were assumed to have been treated with abiraterone or enzalutamide at the same proportions. Second, this model was assumed to have a simple 3-stage 135 design, with patients either being progression free, having progressed disease, or dying. Clinical pathways and patient journeys of this patient population are much more complicated. Third, patients in this model are assumed to cease treatment after their third line of therapy. Fourth, costs associated with progressed disease are assumed to be zero, given patients ceasing treatment. Fifth, patient outcomes are assumed to be like that seen in the CARD randomized clinical trial. Sixth, costs abstracted from an Optum Clinformatics database analysis are representative of all medical costs experienced by mCRPC patients treated with cabazitaxel or an alternative ASTi, following docetaxel or ASTi treatment. 4.3 Results Survival and Effectiveness Estimates In our model, 99% of patients reach the death state at the end of 5 years, regardless of comparator arm. The estimated 1-year overall survival rates were 49.6% for patients treated with an alternative-androgen signaling targeted inhibitor and 58.8% for patients treated with cabazitaxel. The estimated 1-year progression-free survival rates were 10.2% for an alternative- androgen signaling targeted inhibitor or enzalutamide or abiraterone and 23.8% for patients treated with cabazitaxel. The 5-year mortality hazard ratio of cabazitaxel is 0.492 relative to alternative-androgen signaling targeted inhibitor (Table 4.2). Median cycles per cabazitaxel and ASTi arms were 8 and 4, respectively. Median overall survival for simulated patients was 15.2 months for those treated with cabazitaxel and 12.6 for those treated with an alternative ASTi. Finally, median progression free survival was 5.8 months and 3.7 months for those treated with cabazitaxel and an alternative ASTi, respectively. 136 ICER Results In base-case analyses, average discounted time horizon cost of treatments was $166,554 among metastatic, castrate-resistant prostate cancer patients treated with cabazitaxel and $66,302 per patient treated with an alternative androgen-signaling targeted inhibitor, enzalutamide or abiraterone. These estimates align with previously published healthcare resource utilization costs characterizations of this patient population. 28 Total expected effectiveness seen among the cabazitaxel arm was 0.92 QALYs, and 0.76 QALYs among the alternative androgen-signaling targeted inhibitor patients (enzalutamide or abiraterone, Table 3). Incremental cost-effectiveness ratio is $646,578 per QALY. Among post-docetaxel, post-alternative androgen-signaling targeted inhibitor patients, cabazitaxel dominated an alternative androgen-signaling targeted inhibitor (enzalutamide or abiraterone). NMBs are -$84,747 and -$69,242 respectively at willingness to pay thresholds of $100,000 and $200,000 (Table 4.3). ICER Univariate, Two-Way, and Probabilistic Sensitivity Analysis The base case ICER model was sensitive to some parameters. Lower associated cabazitaxel cost was associated with an cabazitaxel having a greater net monetary benefit than an alternative ASTi. Beyond cabazitaxel cost, base case results were robust to all other parameters tested with univariate sensitivity analysis (Figure 4.2). Univariate sensitivity analyses also indicated that the transition probabilities moving patients to the death arm from progressed disease; from progression to progression-free, and to death from progression-free for both cabazitaxel and an alternative ASTi had large impacts on base case NMB (Figure 4.2). 137 One way sensitivity analysis demonstrated cabazitaxel, abiraterone, and enzalutamide 21- day scaled standard costs had the largest impact on base case results. To evaluate the relationship of these parameters on net monetary benefit, 2-way sensitivity analysis was conducted on a scale of $1 to $100,000. At lower an equal cost, cabazitaxel had a positive NMB (Table 4.5). However, at higher equal costs, cabazitaxel had a negative NMB, suggesting the improvement in life expectancy relative to an alternative ASTi may not be cost-effective at higher associated total medical expense. Despite the impact of certain parameters on base-case model cost-effectiveness, results shown here were robust to multivariate probabilistic sensitivity analysis. Probabilistic sensitivity analysis was conducted using 10,000 Monte Carlo simulations. 11% of simulations fell below the $50,000 willingness-to-pay threshold, 15% of simulations fell below the $100,000 willingness-to-pay threshold, and 23% fell below the $200,000 willingness-to-pay threshold (Figure 4.3). Simulations were then mapped on a scale of $0/QALY to $1,000,000/QALY. At lower willingness-to-pay thresholds, alternative ASTi are clearly more likely to be a cost- effective relative to cabazitaxel among mCRPC patients, following post-docetaxel, post- alternative ASTi treatment. As the willingness-to-pay threshold increases, cabazitaxel becomes more likely to be cost-effective (Figure 4.4). At a willingness-to-pay threshold of $538,000/QALY, cabazitaxel becomes more likely to be cost-effective than an alternative ASTi (Figure 4.4). Scenario Analyses ICERs evaluated for a range of scenarios all exceeded $100,000/QALY (See Table 4.6). Model outcomes using costs used in previous literature estimates found the ICER to be 138 $978,072/QALY. When those costs were scaled to 4 ASTi cycles and 7 cabazitaxel cycles, ICER was $155,900/QALY. Total medical costs associated with cancer treatment, as evaluated from Optum Clinformatics DataMart database was found to generate an ICER of $1,009,245/QALY. Assuming subsequent therapy of the other comparator following disease progression, among a proportion of patients’ correspondent to the CARD clinical trial, ICER was found to be $383,139/QALY. Considering varying outcomes found during PSA and scenario analyses, EVPI analyses were conducted to determine the value of elucidating accurate information for gauging this decision problem. At a willingness-to-pay threshold of $100,000/QALY, EVPI per simulated patient was found to be $4,461. Scaled to the potential prevalence of mCRPC patients in the United States, the total EVPI was found to be $191,680,555 3 . Greatest value of EVPI was found to near the base case ICER, around $650,000/QALY (See Figure 4.5). Generalized Risk-Adjusted Cost-Effectiveness In base-case generalized risk-adjusted cost-effectiveness analysis, generalized risk- adjusted cost-effectiveness willingness-to-pay thresholds increase to $32,653 and $65,306, from baseline willingness-to-pay thresholds of $100,000 and $200,000, respectively (Table 4.6). IGRACER for cabazitaxel relative to an alternative ASTi is $64,851/GRA-QALY, suggesting cabazitaxel is not cost-effective at a GRACE-adjusted willingness-to-pay threshold derived from $100,000/QALY; but is cost-effective at a GRACE-adjusted willingness-to-pay threshold derived from $200,000/QALY (Table 4.5). GRACE also indicated at a willingness-to-pay 3 Assuming modeled prevalence of mCRPC is accurate, around 42,790 patients (See Scher, 2015). 139 thresholds $100,000 and $200,000, cabazitaxel has a net monetary benefit of -$512 and $14, respectively. GRACE outcomes 𝐾 𝐺 𝑅 𝐴𝐶𝐸 at willingness-to-pay thresholds of $100,000/QALY and $200,000/QALY; GRACE NMB at willingness-to-pay thresholds of $100,000/QALY and $200,000/QALY; and IGRACER were estimated over a range of percentile higher order risk estimates, generated using the hyperbolic absolute risk aversion utility function. GRACE NMB at willingness-to-pay thresholds of $100,000/QALY and $200,000/QALY did not change notably over a variety of higher order risk parameter estimates (Table 4.7). IGRACER and 𝐾 𝐺 𝑅 𝐴𝐶𝐸 at willingness-to-pay thresholds of $100,000/QALY and $200,000/QALY increased as 𝜔 𝐻 , a function of relative risk aversion and relative prudence, increased (Table 4.7). To better evaluate the relationship of higher order risk parameters and GRACE outcomes, the ratio of IGRACER to 𝐾 𝐺 𝑅 𝐴𝐶𝐸 was examined. This ratio also increased with 𝜔 𝐻 , regardless of willingness-to-pay threshold. 4.4 Discussion Cabazitaxel does not represent a cost-effective treatment option relative to alternative- ASTi (abiraterone or enzalutamide) among post-docetaxel, post-alternative-ASTi (enzalutamide or abiraterone) among modeled patient journeys using the classic incremental cost-effectiveness ratio and GRACE framework at a willingness-to-pay threshold of $100,000/QALY. At a willingness-to-pay threshold of $200,000/QALY, IGRACER was found to be cost-effective relative to 𝐾 𝐺 𝑅 𝐴𝐶𝐸 . Using the GRACE-framework, the disease severity and health risk preference-adjusted willingness-to-pay threshold changed from $100,000/QALY to $32,653/GRA-QALY and $200,000/QALY to $65,306/GRA-QALY. The decrease in 140 willingness-to-pay under GRACE are largely in part due to elasticity of utility with respect to health, 𝜔 𝐻 . Although use of the GRACE framework confirmed the classic ICER evaluation, cabazitaxel as it is currently priced is closer to being cost-effective at a willingness-to-pay threshold of $100,000/QALY and is cost-effective at a willingness-to-pay threshold of $200,000/QALY, despite a greatly reduced decision threshold. This may be in part due to the following assumption. If representative patients in this sample reflect that seen in a generalizable survey, we believe they are risk averse, thus valuing interventions for more severe conditions than less severe, assuming a similar net gain in outcome. Given mCRPC is a severe illness, cabazitaxel becomes a more cost-effective alternative to an ASTi than in the classic ICER framework. Interestingly, under the classic ICER approach, cabazitaxel had considerably different estimates of net monetary benefit under the ICER versus the GRACE frameworks, although with similar signs. This is largely due to GRACE frameworks’ NMB calculation being contingent on the difference in cost between the two comparator treatment options, whereas under the classic ICER NMB calculation, that is not the case. The classic ICER NMB may be a more suitable gauge for evaluating GRACE NMB. In this case, cabazitaxel has greater net monetary benefit relative to an alternative ASTi than under the GRACE framework. This may in part be due to risk-adjusted impact of treatment in the GRACE model. Since greater risk of treatment outcome lowers GRACE NMB, we may see a lower NMB in the GRACE framework with the risk of taking cabazitaxel over an alternative ASTi in terms of overall survival. Ratio of IGRACER to 𝐾 𝐺 𝑅 𝐴𝐶𝐸 was evaluated by percentile estimates of HARA-based higher order risk estimates of relative risk preference and by 𝜔 𝐻 . Although the ratio only 141 changed on a magnitude of 0.001, this analysis demonstrated that as 𝜔 𝐻 increases, health technology valued using the GRACE framework, is more likely to be found as cost-effective. However, IGRACER and GRACE NMB are both contingent on 1) survival into the next period as indicated ( Π) and 2) the population incidence of our disease of interest, mCRPC ( Φ). For the context of this decision problem, Π = 0.01 and Φ = 0.00021. Considering both the disease rarity and probability of survival are so low, NMB and IGRACER are largely impacted, such that cabazitaxel is less likely to be evaluated as cost-effective. However, despite this large value adjustment, cabazitaxel still appears more likely to be cost-effective under a GRACE framework than ICER framework, further supporting the underlying aspects at which GRACE may be deemed more favorable than ICER: greater value assigned to health technology that treats sick populations. However, this impact may be muted by rare diseases. Despite the treatment decision between quality-of-life and life expectancy when considering cabazitaxel relative to an alternative-ASTi, cabazitaxel is not found to be a cost- effective alternative under the ICER framework. The life expectancy improvement at the associated treatment cost of cabazitaxel does not justify the current treatment costs. These findings were found to hold when controlling for treatment certainty, a dynamic willingness to pay threshold accounting for disease severity, and patients who are risk-averse over health- related quality of life. As decision-makers make determinations between these two treatment options, evidence here supports the use of cabazitaxel among mCRPC patients. In the context of this decision problem, relative risk prudence also has an impact on our economic evaluation. It has a practical application pertaining to a patient and decision-maker’s willingness to receive a therapy that requires more subsequent medication and treatment that impacts quality of life, despite improving life expectancy. Prudence also has an impact on 142 GRACE calculations and is critical for the calculation of the change in value associated with uncertain treatment outcomes, how utility may change with health-related quality of life and it can adjust the GRACE willingness-to-pay threshold through the disease severity ratio. There may be 12.2 patients with severe forms of mCRPC that may necessitate a treatment decision between an alternative ASTi (abiraterone or enzalutamide) and cabazitaxel per one million patients enrolled in a health plan. 29 With that consideration, the cost-savings to treating a patient with cabazitaxel in order to extend life is validated here using a classic and innovative value assessment framework. Payers should support the use of cabazitaxel for mCRPC patients who have been treated with an ASTi and docetaxel. However, the decision between a chemotherapy-based regimen and a hormonal-based therapy when life expectancy gains are minimal and the number of patients is small should rest on care teams and the patient, considering chemo-based regimens can still have a large negative impact on patient quality-of- life due to time related to therapy and side effects associated with treatment. Assuming a one- million-member health plan, the incremental per member per month cost increase for allowing a patient and their care team to select an cabazitaxel is only $0.004, when considering the costs associated with the alternative therapy. Patient preference and economic burden is still seen as a serious clinical issue for mCRPC patients being treated with an ASTi or cabazitaxel. 30,31 Randomized clinical trials are still the gold standard for evaluating clinical effectiveness, and should be weighted as such when considering clinical outcomes. However, for mCRPC patients and care teams weighing the decision between cabazitaxel and an alternative ASTi, the advantages of one or the other may be case dependent. This dissertation supports the use of cabazitaxel due to improved outcomes from the patient perspective, given lower out-of-pocket costs. However, this specific chapter also 143 acknowledges that there may be other important features to consider when evaluating the two therapies. In light of this, during instances in which there is a national or general public decision planner (e.g. a national health plan or larger payer), the improvement cabazitaxel provides does not justify the current cost of the drug, thus an alternative ASTi may be more appropriate. Given values not supporting cabazitaxel relative to an alternative ASTi, incremental budget savings of cabazitaxel relative to an alternative ASTi decision between these two alternatives will continue to not be a simple one. These findings are in line with similar studies, even with the inclusion of the innovative GRACE evaluation. 12,28,32 Having a clinically relevant economic evaluation with a new execution of GRACE also provides an example case-study for future applications of the novel GRACE framework. Base case analyses demonstrated costs per cabazitaxel arm equivalent to twice as much as that seen in previous studies. 13 To further elucidate which parameter are driving that difference, some scenario analyses were conducted using published literature estimates for treatments. Using those values and attempted scaling of those values, base case ICER values increased, suggesting the primary driver of ICER difference from the previously published findings was that coming from specific costs specifications. One potential parameter category that may be impacting the found cost differences from the previously published estimates may be transition probabilities. Published literature used a mapped Weibull survival curve from the CARD clinical trial to determine transition probabilities depending on the modeled patient cycle (See solid line on right side figure of Appendix Figure 4.3). Whereas transition probabilities for this model were calculated using the DEALE method, which assumes a constant transition probability derived from the median survival, as determined for this study. While the clinical effectiveness measures demonstrate similar patient outcomes to 144 that of the CARD clinical trial, suggesting some clinical validity. However, the DEALE method for calculating transition probability may underestimate the number of patients on the right tail of the curve for progression free survival, where the greatest cost burden from ASTi or cabazitaxel therapy occurs. Future research should attempt to better align transition probability estimates with that of other published economic evaluations. In light of that, core findings still stand, which find cabazitaxel is not a cost-effective alternative to an ASTi for patients facing this modeled decision problem. ICER differences seen in these analyses as driven by medical costs associated with either comparator arm and a wide variation in probabilistic sensitivity analyses Monte Carlo simulations indicates value can be abstracted from better identifying which patients stand to benefit from cabazitaxel among these patients. 15% of simulated patients from probabilistic sensitivity analyses were determined to have received valuable care when treated with cabazitaxel. Given this finding, EVPI analyses was used to identify the upper threshold of amount to be spent on better identifying those patients. EVPI analysis demonstrated at a willingness-to-pay threshold of $100,000/QALY, there may be $4,461 per simulated patient to be gained by better identifying who stands to benefit from this therapy. Scaled to the prevalence of mCRPC patients in the United States, total EVPI was found to be $191,680,555, suggested immense value could be abstracted from better identifying who stands to benefit from cabazitaxel or an alternative ASTi therapy among this patient population. This study has several limitations. Results captured in this economic evaluation may not represent specific patient journey and care pathways. However, the range of each should be captured by sensitivity analyses, to better inform the decision problem facing physicians prescribing post-docetaxel therapies. Second, in the event published literature is needed to 145 populate a specific parameter due to data not captured in the preceding aims, there is always a risk of omission of current literature estimates. To avoid this, a comprehensive and transparent literature search will be conducted to confirm recent and strong parameter estimates are found. Third, occasionally parameter estimates do not exist in the published literature. In the event this occurs, we will rely on expert opinion to fill any remaining gaps. Fourth, this model is likely an oversimplification of both the patient journey and clinical pathway of mCRPC patients being treated with a third or later line of therapy. Future models attempting to model this decision problem should consider adding stages to capture the full nuance of the patient journey or perhaps generate a model that captures more events specific to these patients, such as what can be accomplished with a discrete event simulation. Fifth, trial evidence from the CARD clinical trial may report more more favorable clinical outcomes, due to the inclusion criteria necessary An updated analysis of real-world clinical outcomes could help further uncover the value of cabazitaxel relative to an alternative ASTi in this context, using more representative clinical outcomes. Sixth, the generalizability of this this study is limited to clinical outcomes that may have occurred in a clinical trial setting and costs for individuals that may be covered under a major insurer. Patients with severe forms of mCRPC may not fully be represented by this population. Seventh, patients were assumed to have ceased treatment or care after their third line of therapy. Further research should aim to examine how many patients continue care if the third line is unsuccessful. Eighth, medical costs were assumed to be zero following disease progression. A healthcare resource analysis or patient survey could be critical in uncovering the true cost for patients, after there’s evidence that their disease is progressing. Analysis presented here represents one of the first executed models of the novel GRACE framework. GRACE systematically incorporates several factors into an economic evaluation that 146 have previously gone unaccounted for outside of the scope of a scenario or sensitivity analysis, such as treatment certainty, treatment hope, severity of illness, and risk aversion related to health. In this context, we see a higher willingness-to-pay for interventions for mCRPC, a severe form of prostate cancer, and greater net monetary benefit due to therapy. This research can serve as a guide for future executions of GRACE and output dissemination. However, there are aspects of GRACE that are still important to consider that have not been covered. For example, GRACE in this context was calculated using the simple 2- stage model, assuming the first stage was the beginning of the patient journey in our model and the last stage is the end of the time horizon. GRACE has multiple versions that include a stepwise approach to execution per cycle, which again allows for a more dynamically modeled treatment decision. The impact of that step-approach should be evaluated relative to what is seen in a 2-staged model. The full value of implementing GRACE when evaluating an uncertain drug or a drug that has a large positive impact for a small group of patients is not seen here. Outcomes captured in the CARD clinical trial are relatively certain relative to the time horizon, only 4–5-month difference in outcomes relative to 5 years in our model. A model evaluating a more uncertain outcome relative to a very certain one could also help illuminate GRACE dynamics and model mechanics. Finally, risk preferences over health were informed using a generalizable survey from the broader US population. Risk preferences are critical for GRACE execution, however, could vary for the patient population being modeled relative to a general US population. Future research should also examine mCRPC-specific patient risk preferences within the GRACE framework to see how model outcomes change. 147 4.5 Conclusion Cabazitaxel does not represent a cost-effective treatment alternative to alternative-ASTi’s (abiraterone or enzalutamide) among mCRPC patients previously treated with docetaxel and an alternative ASTi (enzalutamide or abiraterone) under an ICER value assessment framework assuming a willingness-to-pay threshold of $100,000/QALY or $200,000/QALY or GRACE framework assuming a typical $100,000/QALY willingness-to-pay threshold. Under a more liberal willingness-to-pay threshold of $200,000/QALY, cabazitaxel is cost-effective. Knowing this, mCRPC patients and treatment decision-makers should carefully consider the limited survival benefit given the cabazitaxel cost. Findings in this study were one of the first applications the GRACE framework, which account for treatment outcome certainty, patient risk-aversion over health-related quality-of-life and updated willingness to pay thresholds given treatment severity. Although cabazitaxel is found in this study as a not cost-effective alternative to abiraterone or enzalutamide in later line mCRPC patients, quality-of-life and life expectancy are still important treatment features for a patient to consider and may warrant the use of cabazitaxel. 148 4.6 References 1. Street W. Cancer Facts & Figures 2020. Published online 1930:76. 2. Yabroff KR, Lund J, Kepka D, Mariotto A. Economic Burden of Cancer in the United States: Estimates, Projections, and Future Research. Cancer Epidemiol Biomarkers Prev. 2011;20(10):2006-2014. doi:10.1158/1055-9965.EPI-11-0650 3. Appukkuttan S, Tangirala K, Babajanyan S, Wen L, Simmons S, Shore N. A Retrospective Claims Analysis of Advanced Prostate Cancer Costs and Resource Use. PharmacoEconomics Open. Published online October 22, 2019. doi:10.1007/s41669-019- 00185-8 4. de Wit R, de Bono J, Sternberg CN, et al. Cabazitaxel versus abiraterone or enzalutamide in metastatic prostate cancer. New England Journal of Medicine. 2019;381(26):2506-2518. 5. Loriot Y, Bianchini D, Ileana E, et al. Antitumour activity of abiraterone acetate against metastatic castration-resistant prostate cancer progressing after docetaxel and enzalutamide (MDV3100). Annals of oncology. 2013;24(7):1807-1812. 6. Maines F, Caffo O, Veccia A, et al. Sequencing new agents after docetaxel in patients with metastatic castration-resistant prostate cancer. Critical reviews in oncology/hematology. 2015;96(3):498-506. 7. Attard G, Borre M, Gurney H, et al. Abiraterone alone or in combination with enzalutamide in metastatic castration-resistant prostate cancer with rising prostate-specific antigen during enzalutamide treatment. Journal of Clinical Oncology. 2018;36(25):2639. 8. Oh WK, Cheng WY, Miao R, et al. Real-world outcomes in patients with metastatic castration-resistant prostate cancer receiving second-line chemotherapy versus an alternative androgen receptor-targeted agent (ARTA) following early progression on a first- line ARTA in a US community oncology setting. In: Urologic Oncology: Seminars and Original Investigations. Vol 36. Elsevier; 2018:500. e1-500. e9. 9. Lakdawalla DN, Phelps CE. Health Technology Assessment With Diminishing Returns to Health: The Generalized Risk-Adjusted Cost-Effectiveness (GRACE) Approach. Value in Health. 2021;24(2):244-249. doi:10.1016/j.jval.2020.10.003 10. Lakdawalla DN, Phelps CE. Health technology assessment with risk aversion in health. Journal of Health Economics. 2020;72:102346. doi:10.1016/j.jhealeco.2020.102346 11. Lakdawalla DN, Phelps CE. Evaluation of Medical Technologies with Uncertain Benefits. National Bureau of Economic Research; 2019. doi:10.3386/w26058 12. Barqawi YK, Borrego ME, Roberts MH, Abraham I. Cost-effectiveness model of abiraterone plus prednisone, cabazitaxel plus prednisone and enzalutamide for visceral 149 metastatic castration resistant prostate cancer therapy after docetaxel therapy resistance. J Med Econ. 2019;22(11):1202-1209. doi:10.1080/13696998.2019.1661581 13. Zhang PF, Xie D, Li Q. Cost-effectiveness analysis of cabazitaxel for metastatic castration resistant prostate cancer after docetaxel and androgen-signaling-targeted inhibitor resistance. BMC Cancer. 2021;21(1):35. doi:10.1186/s12885-020-07754-9 14. Sanders GD, Neumann PJ, Basu A, et al. Recommendations for Conduct, Methodological Practices, and Reporting of Cost-effectiveness Analyses: Second Panel on Cost- Effectiveness in Health and Medicine. JAMA. 2016;316(10):1093-1103. doi:10.1001/jama.2016.12195 15. de Wit R, de Bono J, Sternberg CN, et al. Cabazitaxel versus abiraterone or enzalutamide in metastatic prostate cancer. New England Journal of Medicine. 2019;381(26):2506-2518. 16. Phelps C, Cinatl C. Estimating Optimal Willingness to Pay Thresholds for Cost- Effectiveness Analysis: A Generalized Method. Health Economics. 2021;(Forthcoming). 17. Phelps CE. A New Method to Determine the Optimal Willingness to Pay in Cost- Effectiveness Analysis. Value in Health. 2019;22(7):785-791. doi:10.1016/j.jval.2019.03.003 18. Vanness DJ, Lomas J, Ahn H. A Health Opportunity Cost Threshold for Cost-Effectiveness Analysis in the United States. Ann Intern Med. Published online November 3, 2020. doi:10.7326/M20-1392 19. Neumann PJ, Cohen JT, Weinstein MC. Updating Cost-Effectiveness — The Curious Resilience of the $50,000-per-QALY Threshold. N Engl J Med. 2014;371(9):796-797. doi:10.1056/NEJMp1405158 20. Fleurence RL, Hollenbeak CS. Rates and probabilities in economic modelling: transformation, translation and appropriate application. Pharmacoeconomics. 2007;25(1):3- 6. doi:10.2165/00019053-200725010-00002 21. Miller DK, Homan SM. Determining transition probabilities: confusion and suggestions. Med Decis Making. 1994;14(1):52-58. doi:10.1177/0272989X9401400107 22. Jones E, Epstein D, García-Mochón L. A Procedure for Deriving Formulas to Convert Transition Rates to Probabilities for Multistate Markov Models. Med Decis Making. 2017;37(7):779-789. doi:10.1177/0272989X17696997 23. U.S. Bureau of Labor Statistics. Consumer Price Index for All Urban Consumers: Medical Care in U.S. City Average. FRED, Federal Reserve Bank of St. Louis. Published January 1, 1947. Accessed March 10, 2022. https://fred.stlouisfed.org/series/CPIMEDSL 150 24. Bureau of Labor Statistics Data. Accessed March 10, 2022. https://data.bls.gov/timeseries/CUUR0000SA0L1E?output_view=pct_12mths 25. Matza LS, Cong Z, Chung K, et al. Utilities associated with subcutaneous injections and intravenous infusions for treatment of patients with bone metastases. Patient Prefer Adherence. 2013;7:855-865. doi:10.2147/PPA.S44947 26. Torvinen S, Färkkilä N, Sintonen H, Saarto T, Roine RP, Taari K. Health-related quality of life in prostate cancer. Acta Oncol. 2013;52(6):1094-1101. doi:10.3109/0284186X.2012.760848 27. Fizazi K, Shore N, Tammela TL, et al. Nonmetastatic, Castration-Resistant Prostate Cancer and Survival with Darolutamide. N Engl J Med. 2020;383(11):1040-1049. doi:10.1056/NEJMoa2001342 28. Morgans A, Hutson T, Guan AK, et al. Clinical and cost impact of cabazitaxel versus a second androgen receptor-targeted agent (ARTA) for patients with metastatic castration- resistant prostate cancer (mCRPC) previously treated with docetaxel and the alternative ARTA (abiraterone or enzalutamide). ESMO Virtual Congress. 587P. 29. Scher HI, Solo K, Valant J, Todd MB, Mehra M. Prevalence of Prostate Cancer Clinical States and Mortality in the United States: Estimates Using a Dynamic Progression Model. PLoS One. 2015;10(10):e0139440. doi:10.1371/journal.pone.0139440 30. $150K Prostate Cancer Drug Draws New Attention to “March-In” Rights. Published April 6, 2022. Accessed April 12, 2022. https://www.medpagetoday.com/special- reports/exclusives/98091 31. Caram MEV, Oerline MK, Dusetzina S, et al. Adherence and out-of-pocket costs among Medicare beneficiaries who are prescribed oral targeted therapies for advanced prostate cancer. Cancer. 2020;126(23):5050-5059. doi:10.1002/cncr.33176 32. Zhang PF, Xie D, Li Q. Cost-effectiveness analysis of cabazitaxel for metastatic castration resistant prostate cancer after docetaxel and androgen-signaling-targeted inhibitor resistance. BMC Cancer. 2021;21(1):35. doi:10.1186/s12885-020-07754-9 33. Phelps CE, Lakdawalla DN. A Guide to Extending and Implementing Generalized Risk- Adjusted Cost-Effectiveness (GRACE)*. European Journal of Health Economics. Forthcoming. 34. Thurin NH, Rouyer M, Gross-Goupil M, et al. Epidemiology of metastatic castration- resistant prostate cancer: A first estimate of incidence and prevalence using the French nationwide healthcare database. Cancer Epidemiol. 2020;69:101833. doi:10.1016/j.canep.2020.101833 151 35. Lloyd AJ, Kerr C, Penton J, Knerer G. Health-Related Quality of Life and Health Utilities in Metastatic Castrate-Resistant Prostate Cancer: A Survey Capturing Experiences from a Diverse Sample of UK Patients. Value in Health. 2015;18(8):1152-1157. doi:10.1016/j.jval.2015.08.012 36. Jiang R, Janssen MFB, Pickard AS. US population norms for the EQ-5D-5L and comparison of norms from face-to-face and online samples. Qual Life Res. Published online October 6, 2020:1-14. doi:10.1007/s11136-020-02650-y 37. Siegel DA. Prostate Cancer Incidence and Survival, by Stage and Race/Ethnicity — United States, 2001–2017. MMWR Morb Mortal Wkly Rep. 2020;69. doi:10.15585/mmwr.mm6941a1 38. George DJ, Sartor O, Miller K, et al. Treatment Patterns and Outcomes in Patients With Metastatic Castration-resistant Prostate Cancer in a Real-world Clinical Practice Setting in the United States. Clinical Genitourinary Cancer. 2020;18(4):284-294. doi:10.1016/j.clgc.2019.12.019 39. Chen Q, Ayer T, Nastoupil LJ, Rose AC, Flowers CR. Comparing the cost-effectiveness of rituximab maintenance and radioimmunotherapy consolidation versus observation following first-line therapy in patients with follicular lymphoma. Value Health. 2015;18(2):189-197. doi:10.1016/j.jval.2014.12.017 40. Wu B, Li SS, Song J, Pericone CD, Behl AS, Dawson NA. Total cost of care for castration- resistant prostate cancer in a commercially insured population and a medicare supplemental insured population. J Med Econ. 2020;23(1):54-63. doi:10.1080/13696998.2019.1678171 152 Chapter 4 Tables Table 4.1 Classic Model Parameters Parameter Category Parameter Deterministic Values Lower Bound Upper Bound Distribution Source Transition Probability PFS (ASTi) 0.77 0.72 0.82 Beta Derived using DEALE Method and CARD Trial (15) PFS (Cab) 0.81 0.77 0.85 Beta PD (ASTi) 0.95 0.94 0.96 Beta PD (Cab) 0.96 0.95 0.97 Beta Death (ASTi) 0.05 0.04 0.06 Beta 37 Death (Cab) 0.04 0.03 0.05 Beta 37 PD (ASTi) 0.18 0.14 0.22 Beta 15 PD (Cab) 0.15 0.12 0.18 Beta 15 PD -> Death (ASTi) 0.05 0.04 0.06 Beta 38 PD -> Death (Cab) 0.04 0.03 0.05 Beta 38 Utility Utility - Cabazitaxel (PFS) 0.72 0.69 0.81 Beta 25–27 Utility - Cabazitaxel (PD) 0.6 0.48 0.7 Beta Utility - ASTi (PFS) 0.7 0.64 0.75 Beta Utility - ASTi (PD) 0.6 0.48 0.7 Beta Adverse Event Transition Probabilities Musculoskeletal Pain or Discomfort (Cab) 0.001 0 0.0008 Beta CARD Trial (15) Renal Disorder (Cab) 0.0021 0 0.0017 Beta Anemia (Cab) 0.0052 0.01 0.0041 Beta Leukopenia (Cab) 0.0206 0.02 0.0165 Beta Neutropenia (Cab) 0.0288 0.03 0.0231 Beta Musculoskeletal Pain or Discomfort (ASTi) 0.0032 0 0.0026 Beta Renal Disorder (ASTi) 0.0047 0.01 0.0037 Beta Anemia (ASTi) 0.0028 0 0.0022 Beta Leukopenia (ASTi) 0.0009 0 0.0007 Beta Neutropenia (ASTi) 0.0018 0 0.0015 Beta Costs Cabazitaxel, 3 rd Line Setting $36,408.75 $29,127.00 $43,691 Log-normal Retrospective Claims Analysis Abiraterone, 3 rd Line Setting $42,802.24 $34,241.79 $51,363 Log-normal Retrospective Claims Analysis Enzalutamide, 3 rd Line Setting $43,906.49 $35,125.19 $52,688 Log-normal Retrospective Claims Analysis 153 Musculoskeletal Pain or Discomfort $384.35 $252.56 $378.84 Log-normal 39 Renal Disorder $6,286.43 $4,130.85 $6,196.28 Log-normal 39 Anemia $1,981.81 $1,302.26 $1,953.38 Log-normal 39 Leukopenia $3,230.31 $2,122.66 $3,183.98 Log-normal 39 Neutropenia $3,230.31 $2,122.66 $3,183.98 Log-normal 39 Routine Follow- Up $407.56 $292.16 $438.24 Log-normal 40 Supportive Care following Progressed Disease $1,171.48 $839.79 $1,259.68 Log-normal 40 Notes: Assuming 1 cycle is 3 weeks, each cost is scaled per cycle. Medical costs were abstracted from an analysis of Optum claims identifying medical costs associated with 3 rd line therapy for the 3 rd drugs of interest. Medical costs were the aggregated of inpatient, outpatient, emergency, hospice, pharmacy, and other captured costs within the claims database. PFS = Progression-Free Survival; PD = Progressed Disease; ASTi = Androgen-Signaling Targeted Inhibitor; Cab = Cabazitaxel 154 Table 4.2 Model Clinical Outcomes Model Outcomes Cabazitaxel ASTi Median Cycles 8 4 Median PFS (months) 5.8 3.7 Median OS (months) 15.2 12.6 5-year Survival 1.03% 0.51% 1-year Survival 58.8% 49.6% 5-year Mortality, Hazard Ratio 0.492 Ref 1-year Mortality, Hazard Ratio 0.811 Ref Notes: ASTi refers to mCRPC, post-docetaxel, post-alternative ASTi patients treated with either enzalutamide or abiraterone. ASTi = Alternative Androgen-Signaling Targeted Inhibitor (enzalutamide or abiraterone); PFS = Progression Free Survival; OS = Overall Survival 155 Table 4.3 Expected Base Case Results Discounted Costs Incremental Costs QALYs Incremental QALYs ICERs ($/QALY) NMB ($200k) NMB ($100k) Cabazitaxel $166,554 $100,251 0.92 0.16 $646,578/QALY -$69,242 -$84,747 ASTi $66,302 Ref 0.76 Ref Ref Ref Ref Notes: ASTi refers to mCRPC, post-docetaxel, post-alternative ASTi patients treated with either enzalutamide or abiraterone. QALYs = Quality-Adjusted Life Years (QALYs); NMB = Net Monetary Benefit; mCRPC = metastatic, castration-resistant prostate cancer 156 Table 4.4 Two-Way Sensitivity Analysis Abiraterone or Enzalutamide Cost $1 $20,000 $40,000 $60,000 $80,000 $100,000 Cabazitaxel Cost $1 $12,483 $151,924 $291,371 $430,819 $570,267 $709,715 $20,000 -$214,334 -$74,893 $64,555 $204,003 $343,451 $482,899 $40,000 -$441,162 -$301,721 -$162,273 -$22,825 $116,623 $256,071 $60,000 -$667,990 -$528,549 -$389,101 -$249,653 -$110,205 $29,243 $80,000 -$894,817 -$755,376 -$615,929 -$476,481 -$337,033 -$197,585 $100,000 -$1,121,645 -$982,204 -$842,756 -$703,308 -$563,861 -$424,413 Notes: Costs associated post docetaxel, post abiraterone or enzalutamide cabazitaxel or alternative ASTi were varied from $1 - $100,000 per 21 days. Green cells indicate instances in which cabazitaxel had a positive net monetary benefit, and red cells indicate instances in which cabazitaxel did not have a positive net monetary benefit. Patients were assumed to have an equal likelihood of taking either ASTi (abiraterone or enzalutamide) in the alternative ASTi arm. 157 Table 4.5 Base Case Scenario Analyses Results ICER NMB Base Case $646,578 -$84,747 Zhang, 2021 Costs $978,072 -$136,144 Zhang Costs Scaled to 4 ASTi Cycles and 7 Cabazitaxel Costs $155,900 -$8,667 Optum CDM Cancer Costs $1,009,245 -$140,978 Subsequent Therapy following PD $383,139 -$58,469 Notes: Base Case ICER and NMB are presented here as a result of four sensitivity analyses based on published literature. ICER = Incremental Cost-Effectiveness Ratio; GRACE = NMB = Net Monetary Benefit; Alternative ASTi = Alternative Androgen-Signaling Targeted Inhibitor (enzalutamide or abiraterone) 158 Table 4.6 Expected Base Case Results, Classic and GRACE Model ICER Outcomes GRACE Outcomes ICER $646,578 $64,851 IGRACER $100,000 NMB -$74,608 -$512 $100,000 NMB $200,000 NMB $17,337 $14 $200,000 NMB K $100,000 $32,653 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 @ $100k K $200,000 $65,306 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 @ $200k Notes: Key results are displayed in the middle two columns, allowing for direct comparison. ICER base case results and reference willingness-to-pay thresholds are delineated on the left. GRACE base results and scaled willingness-to-pay thresholds are delineated on the right. K’s in orange represent willingness-to-pay thresholds below ICER or IGRACER. K’s in green represent willingness-to-pay threshold above ICER. ICER = Incremental Cost-Effectiveness Ratio; GRACE = Generalized Risk-Adjusted Cost Effectiveness Ratio; NMB = Net Monetary Benefit; K = Willingness to Pay Threshold; KGRACE = GRACE-adjusted willingness to pay threshold; Alternative ASTi = Alternative Androgen-Signaling Targeted Inhibitor (enzalutamide or abiraterone) 159 Table 4.7 Hyperbolic Absolute Risk Aversion Higher Order Estimates Impact on GRACE Output Percentile Hyperbolic Absolute Risk Estimates over Health 5th 10th 25th Median 75th 90th 95th Mean Higher Order Relative Risk Parameter Estimates over Health Aversion 0.006 0.038 0.071 0.206 0.296 0.399 0.473 0.221 Prudence 0.031 0.079 0.144 0.798 1.197 1.261 1.277 0.688 Temperance 0.046 0.119 0.218 1.196 2.197 2.261 2.271 1.155 ⁡ ⁡ ⁡ ⁡ 𝜔 𝐻 0.018 0.003 0.001 0.385 0.605 0.464 0.331 0.246 GRACE Output 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 100k $2,311 $432 $166 $52,975 $88,038 $68,009 $48,701 $32,653 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 200K $4,622 $865 $331 $105,951 $176,076 $136,019 $97,402 $65,306 NMB 100K -$512 -$512 -$512 -$512 -$512 -$512 -$512 -$512 NMB 200K $14 $14 $14 $14 $14 $14 $14 $14 IGRACER $4,592 $859 $329 $105,132 $174,366 $134,876 $96,674 $64,851 IGRACER and 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 Ratio 𝐼 𝐺 𝑅𝐴 𝐶 𝐸𝑅 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 ⁡ @ ⁡ 100 𝑘 1.9869 1.9869 1.9869 1.9845 1.9806 1.9832 1.9851 1.986 𝐼 𝐺 𝑅𝐴 𝐶 𝐸𝑅 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 @ ⁡ 2 0 0 𝑘 0.9935 0.9935 0.9935 0.9923 0.9903 0.9916 0.9925 0.993 Notes: Key GRACE outputs are presented for varying percentile and mean estimates of higher order risk parameters estimated using the hyperbolic absolute risk aversion function. 𝜔 𝐻 = Elasticity of utility with respect to QoL (H) at 𝐻 0 ; GRACE = Generalized Risk-Adjusted Cost- Effectiveness; 𝐾 𝐺 𝑅 𝐴𝐶𝐸 = GRACE-adjusted willingness-to-pay threshold; NMB = Net Monetary Benefit; IGRACER = Incremental Generalized, Risk-Adjusted Cost-Effectiveness Ratio 160 Chapter 4 Figures Figure 4.1 Markov Model Notes: A Markov model simulating treatment of post-alternative ASTi, post-docetaxel metastatic, castrate-resistant prostate cancer patients treated with either cabazitaxel or an alternative ASTi (abiraterone or enzalutamide). 161 Figure 4.2 Univariate Sensitivity Analyses Notes: Tornado diagram displaying outcomes of univariate sensitivity analyses. Each parameter was varied by the reported upper and lower bounds of published estimates or 20% if reported estimates lacked upper and lower bounds. Results are presented here as net monetary benefits, at a willingness to pay threshold of $100,000/QALY. White bars represent upper bound estimates of parameter. Black bars represent lower bound estimates of parameter. Red dashed line represents $0 net monetary benefit. PFS = Progression Free Survival; ASTi = Androgen Signaling Targeted Inhibitor; PD = Progressed Disease; NMB = Net Monetary Benefit; QALY = Quality-Adjusted Life Year 162 Figure 4.3 Incremental Cost-effectiveness Plane. Note: 10,000 simulations were conducted to perform a probabilistic sensitivity analysis of our model. Each dot represents one simulated Monte Carlo trial. The orange dot represents the origin, and the black line represents the willingness to pay threshold. Simulations that fall below that line represent valuable care. Solid line represents a willingness-to-pay decision threshold of $200,000/QALY, dashed line represents a willingness-to-pay threshold of $100,000/QALY, and dotted line represents a willingness-to-pay decision threshold of $50,000/QALY. QALY = Quality-Adjusted Life Year 163 Figure 4.4 Cost-Effectiveness Acceptability Curve. Notes: Cost-effectiveness acceptability curve contains mapped 10,000 Monte Carlo simulations at willingness to pay thresholds increasing in $2,000 intervals from $0 to $1,000,000. The vertical doted line represents the observed Base Case ICER, at $646,578/QALY. Cab = Cabazitaxel; ASTi = Alternative-ASTi 164 Figure 4.5 Expected Value of Perfect Information at Varying Willingness-to-Pay Thresholds. Notes: Expected Value of Perfect Information (EVPI) is mapped here over varying willingness- to-pay (WTP) thresholds, from $0 to $1,000,000. Peak in EVPI occurs at the previously calculated ICER, around $650,000 QALY. QALY = Quality-Adjusted Life Years 165 4.7 Appendix Appendix Note 4.1 Generalized Risk-Adjusted Cost-Effectiveness Modeling Approach A three-state (Survival, Onset of Adverse Event, and Death) 5-year time horizon Markov again will be used to model patient flow for the GRACE framework (Figure 1). “One-time” parameters will be calculated through the successful implantation of the general population survey in dissertation aim 2. That includes the following parameters: relative risk aversion, related to treatment outcomes variance; and relative risk prudence, related to skewness of treatment outcomes. Severity-adjusted willingness-to-pay will be calculated using the following formula. 𝐾 𝐺 𝑅 𝐴𝐶𝐸 = 𝜔 𝐻 𝜔 𝐶 × 𝑅 × 𝐶 = 𝐾 𝜔 𝐻 𝑅 Eq. 4.3 Whereas, 𝐾 is equivalent to 𝐶 𝜔 𝐶 . 𝐶 𝜔 𝐶 is equal to the classical willingness-to-pay threshold. For the purposes of comparing GRACE to the classical framework, this value will be equivalent to the cost-effectiveness threshold of $100,000/QALY or $200,000/QALY. 16,17 𝑅 will be calculated using higher order relative risk preferences determined in Aim 2, and the severity of untreated disease, as captured in the targeted claims analysis. 𝜔 𝐻 will also be estimated using the higher order risk parameters captured in Aim 2. Each of these parameters, excluding R, will only need to be estimated once and can be used for future GRACE applications. The untreated disease severity parameter can be used for subsequent analyses of mCRPC. In classical cost-effectiveness analysis, interventions are considered valuable if the incremental costs, relative to the incremental benefits are less than the willingness-to-pay 166 threshold. However, GRACE adjusts this framework, such that an intervention is considered valuable if the risk-adjusted cost-effectiveness ratio is less than or equal to the risk-adjusted willingness-to-pay. 𝐾 𝐺 𝑅 𝐴𝐶𝐸 = 𝐾 × 𝜔 𝐻 × 𝑅 ≥ Δ 𝐶𝑜𝑠𝑡 Δ ( Ris k ⁡ A d ju s t ed ⁡ 𝑄 𝐴𝐿 𝑌 𝑠 ) Eq. 4.4 To complete this equation for determining risk-adjusted value, the following parameters were calculated, which will be specific to each future GRACE calculation. Cost were calculated in a similar fashion to the classical model. Δ ( Risk ⁡A d j usted ⁡𝑄 𝐴 𝐿𝑌 𝑠 ) will be the next parameters we need to estimate, that are treatment and disease-specific. Formula for calculating Δ ( Risk ⁡A d j usted ⁡𝑄 𝐴 𝐿𝑌 𝑠 ) is as follows. ( Δ 𝑆 ) 𝛿 ⁡ + 𝑆 ( Δ Q ) 𝜖 Eq. 4.5 ( Δ 𝑆 ) 𝛿 ⁡ + 𝑆 ( Δ Q ) 𝜖 and 𝑅 will be calculated using the following parameters. First, incremental gain in QALYs were calculated compared with the current standard of care, as defined as 𝜇 𝐵 . This value is commonly estimated and is equivalent to the difference between the average QoL outcomes between the new treatment (cabazitaxel) and its comparator (an alternative ASTi, enzalutamide or abiraterone). 𝜇 𝐵 was estimated using values from the Monte Carlo simulation. Second, we calculated the variance of both new treatment and comparator outcomes, as captured in claims data, published clinical studies, or from results of Monte Carlo simulations of our classical model. For GRACE we need the difference in variances, as described by Δ 𝜎 𝐻 2 and the variances within each arm. Third, we captured the skewness of patient outcomes in each arm. To do this, we must estimate Pearson’s skewness parameter, 𝛾 1 , for both cabazitaxel and alternative ASTi arm’s. 16 GRACE will need the difference in skewness between each comparator arm, as well as the 167 skewness within each arm. These values can also be abstracted from the original cost- effectiveness model PSA. Fourth, we estimated relative QoL loss from mCRPC before and after treatment with both cabazitaxel and alternative-ASTi in post-docetaxel, post-alternative ASTi mCRPC patients. Loss in untreated states is defined as percentage losses in QoL suffered in untreated states (before cabazitaxel or alternative-ASTi treatment). Losses in treated states are defined percentage loss in QoL after treatment by cabazitaxel or alternative-ASTi. We determined loss in treated states estimates for each comparator in our study. Finally, we calculated ex ante QoL level, which is defined as the QoL prior to onset of mCRPC. In this case, this will be the QoL in patients treated are being treated by docetaxel before prescription of one of our comparators. Fifth, we calculated the increase in survival probability for post-docetaxel, post- alternative ASTi mCRPC patients compared to an alternative ASTi, described in GRACE as 𝜇 𝑝 . Approaches to this calculation may vary depending on long-term survival gains contingent on QoL distributions. Sixth, we determined the probability of survival in the first-year post-comparator (cabazitaxel or alternative ASTi) prescription for our target patient population. This is defined in GRACE as 𝑝 1 . Finally, we calculated the population incidence of post-docetaxel, post-alternative ASTi mCRPC, which can be estimated using epidemiological burden of illness studies or estimated using claims data or a national cancer registry such as Center for Disease Control’s Surveillance, Epidemiology, and End Results Program. Each of these parameters were readily available from our targeted claims analysis, published cost-effectiveness studies or value assessments, classical ICER analysis included in 168 this proposal, or published epidemiological studies. 𝜖 can then be estimated using the following Taylor Series expansion. ϵ ≈ 1 + [ 1 𝜇 𝐵 ] [ − 1 2 𝑟 𝐻 ∗ ( 1 μ H ) Δ σ H 2 + 1 6 𝜋 𝐻 ∗ 𝑟 𝐻 ∗ ( 1 μ H ) Δ [ 𝛾 1 𝜎 𝐻 3 ] + ⋯ ] Eq. 4.6 𝛿 is the marginal rate of substitution between QoL units and 1-year of life expectancy. This can be calculated using the efficient estimation method, mentioned in appendix note in Lakdawalla & Phelps. 33 Δ 𝑆 is the incremental change in life expectancy between cabazitaxel and an alternative- ASTi, 𝑆 is baseline life expectancy, and Δ Q is the change in QoL between cabazitaxel and an alternative-ASTi. Each of these variables combined with the above equations will yield risk adjusted QALYs of cabazitaxel relative to an alternative ASTi in post-docetaxel, post-alternative ASTi mCRPC patients. Since we evaluated treatments for mCPRC over a 5-year time horizon, in 3-week cycles, some parameters will need to be estimated over each individual cycle. Those include the following: mean QoL benefit; difference in the variance of QoL benefit; difference in skewness of QoL benefit; percentage QoL loss from untreated disease; percentage QoL loss from permanent disability per period; percentage QoL loss from treated disease; indexing ratio of expected utility from QoL in period 𝑛 to the utility from QoL in period 𝑛 − 1; indexing changes in uncertainty in treatment outcomes in period 𝑛 to period 𝑛 − 1; disease severity, indexed per period; incremental cost per period; and the average increase in probability of surviving period 𝑛 − 1 to period 𝑛 . A one period discount factor also needs to be incorporated per cycle. 169 Appendix Table 4.1 Generalized Risk-Adjusted Cost-Effectiveness Modeling, Literature Dependent Values Parameter Category Symbol Parameter Deterministic Values Source Literature- Based GRACE Parameters 𝜔 𝐻 Elasticity of utility itself with respect to health (H) 0.24 Parameters derived in Aim 2 𝑟 𝐻 ∗ Relative risk aversion 0.22 Parameters derived in Aim 2 𝜋 𝐻 ∗ Relative risk prudence 0.69 Parameters derived in Aim 2 𝜏 𝐻 ∗ Relative risk temperance 1.16 Parameters derived in Aim 2 𝑝 1 The probability of surviving to period 1, described in GRACE (level of survival probability instead of change) 0.0103 15 𝜙 Weighted sum of acute illness state (S) and other for the "well" state (W), population incidence of the disease of interest 0.00021 34 𝐻 1 𝑠 Period 1 level of untreated health 0.63 35 𝐵 Stochastic QoL benefit arising from the intervention 0.16 15 𝐻 1 𝑊 Health when no acute illness occurs 0.81 36 𝜙 Conditional probability of acute illness in period 1, given survival 0.00021 34 𝐻 1 𝑠 Health-related quality of life in the untreated sick state 0.625 35 𝐻 0 Health-related quality of life in the baseline state 1 36 𝐶 Income available for consumption after medical spending is removed $65,297.00 The World Bank Assumed GRACE Parameters 𝜔 𝐶 Elasticity of utility with respect to consumption ( C ) 0.30 16 𝑑 Health reduced by permanent disabilities 0.00 Assumed 0 𝜄 ∗ Untreated illness severity, the average QoL that occurs between baseline period and untreated illness states, relative to 𝐻 0 0.984 Calculated using model values Notes: This table contains value estimates and sources for literature-dependent values for GRACE execution. QoL = Quality of Life; GRACE = Generalized Risk-Adjusted Cost- Effectiveness 170 Appendix Table 4.2 Generalized Risk-Adjusted Cost-Effectiveness, Key Calculations Variable Description Formula Estimates 𝜔 𝐻 Elasticity of utility itself with respect to health (H) 𝑊 ′ ( 𝐻 𝑜 ) 𝐻 𝑜 𝑊 ( 𝐻 𝑜 ) 0.24 R Severity-of-illness adjustment factor { 1 + 𝑟 𝐻 ∗ 𝜄 ∗ + ⁡ 1 2 𝑟 𝐻 ∗ 𝜋 𝐻 ∗ 𝜄 ∗ 2 + ⁡ 1 6 𝑟 𝐻 ∗ 𝜋 𝐻 ∗ 𝜏 𝐻 ∗ 𝜄 ∗ 2 … ] 1.36 𝛿 Marginal rate of substitution between LE and QoL 𝜌 𝐻 0 𝜔 𝐻 𝑅 3.06 𝜖 Adjustment for changes in treatment outcomes uncertainty 1 + [ 1 𝜇 𝐵 ] [ − 1 2 𝑟 𝐻 ∗ ( 1 𝜇 𝐻 ) Δ 𝜎 𝐻 2 + 1 6 𝑟 𝐻 ∗ 𝜋 𝐻 ∗ ( 1 𝜇 𝐻 ) Δ [ 𝛾 1 𝜎 𝐻 3 ] + ⋯ ⁡ ] 1.01 K Traditional cost- effectiveness threshold 𝑐 𝑤 𝑐 $200,000 or $100,000 𝜌 Ratio of expected utility from QoL in period 1 to the utility of QoL in the base period Φ ρ s + ( 1 − 𝜙 ) 𝜌 𝑤 1 𝜌 𝑠 Ration of expected utility from QoL in period 1 for the acute illness state 1 − 𝜔 𝐻 [ 𝑡 ∗ + 1 2 𝑟 𝐻 ∗ [ 𝜎 𝜏 2 𝐻 0 2 ] − ⁡ 1 6 𝑟 𝐻 ∗ 𝜋 𝐻 ∗ [ 𝜎 3 𝐻 0 3 ] 𝛾 𝜏 + ⋯ ] 0.764 𝜌 𝑤 Ration of expected utility from QoL in period 1 for the well state 1 − 𝜔 𝐻 𝑑 ∗ 1 Notes: This table highlights GRACE key inputs, formulas, and values used for this model 171 Appendix Figure 4.1 Impact of HARA Estimates on GRACE Notes: The above figure displays the relationship between 𝐾 𝐺 𝑅 𝐴𝐶𝐸 and IGRACER, as a plot of 𝐼 𝐺 𝑅 𝐴𝐶𝐸 𝑅 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 by varying percentiles of HARA estimates, generated from the third chapter of this dissertation. HARA = Hyperbolic Absolute Risk Aversion; GRACE = Generalized Risk- Adjusted Cost-Effectiveness; 𝐾 𝐺 𝑅 𝐴𝐶𝐸 = GRACE-adjusted willingness-to-pay ratio of $100,000/QALY; IGRACER = Incremental-GRACE-Ratio 172 Appendix Figure 4.2 Impact of 𝝎 𝑯 on GRACE Output Notes: The above figure displays the relationship between 𝐾 𝐺 𝑅 𝐴𝐶𝐸 and IGRACER, as a plot of 𝐼 𝐺 𝑅 𝐴𝐶𝐸 𝑅 𝐾 𝐺 𝑅 𝐴 𝐶 𝐸 by varying 𝜔 𝐻 , as a function of the inverse of 1 𝜋 ∗ − 2 𝑟 ∗ . HARA = Hyperbolic Absolute Risk Aversion; GRACE = Generalized Risk-Adjusted Cost-Effectiveness; 𝐾 𝐺𝑅 𝐴 𝐶𝐸 = GRACE- adjusted willingness-to-pay ratio of $100,000/QALY; IGRACER = Incremental-GRACE-Ratio 173 Appendix Figure 4.3 Visual Impact of DEALE Method vs. Estimating Weibull Survival Curves Notes: Differences in costs abstracted in both economic evaluations may be a function of the transition probability abstraction method. By using the DEALE method, patients may exist in the progression free survival stage for longer than those using the transition probability extraction method seen in Zhang et al., which could translate to a difference in costs since patients are treated with our comparators of interest during PFS. The black lines indicate how DEALE abstracted transition probabilities may be a linear representation of PFS, instead of the more dynamic Weibull function seen on the right-hand side. PFS = Progression Free Survival. 174 Chapter 5: Conclusions and Policy Implications Samuel A. Crawford 5.1 Conclusions In this dissertation we examine novel treatment sequencing for the care of patients with metastatic, castrate-resistant prostate cancer (mCRPC) and conduct a novel value assessment using the Generalized Risk-Adjusted Cost-Effectiveness (GRACE) framework using values derived from the aims of this dissertation. In Chapter 2, we evaluate the economic burden associated with treatment options in second- and third-line care for patients with mCRPC, treated with an androgen-signaling targeted inhibitor (ASTi), enzalutamide or abiraterone, docetaxel, and cabazitaxel. We find costs are high with associated intravenous therapeutic options in either a second- or third-line setting. However, the out-of-pocket burden for those therapies is considerably less than for the oral agents, enzalutamide or abiraterone, likely due to differences in coverage for intravenous and oral therapies. In Chapter 3, we characterize higher order estimates of relative risk preferences over decision making for a treatment condition for general health and specific conditions using the hyperbolic absolute risk aversion. We find relative risk aversion, relative prudence, and relative temperance are lower among treatment decision making over general health relative to a financial-based lottery. Estimates over the financial-based lottery are slightly lower than previously published risk estimates. 1,2 The ratio generated from health relative risk estimates and cash relative risk estimates from this aim can be used in future research to convert other published relative risk estimates over monetary gains to those over health. 175 Finally, in chapter 4, we examine the value of cabazitaxel relative to an ASTi among patients treated with an alternative ASTi and docetaxel using the classical incremental cost- effectiveness ratio (ICER) and novel GRACE value assessment frameworks. We find using ICER that cabazitaxel is not a cost-effective alternative to an ASTi among mCRPC patients previously treated with docetaxel an alternative ASTi at willingness-to-pay thresholds of $100,000/QALY and $200,000/QALY. At a GRACE willingness-to-pay ( 𝐾 𝐺 𝑅 𝐴𝐶𝐸 ) threshold derived from $100,000/QALY, cabazitaxel was also found not to be a cost-effective alternative to an ASTi. However, at a 𝐾 𝐺 𝑅 𝐴𝐶𝐸 derived from $200,000, cabazitaxel was found to be cost- effective alternative to an ASTi. These findings support previously published literature, but also demonstrate the flexibility afforded by GRACE through the rigorous inclusion of parameters previously unconsidered in ICER modeling, beyond specific sensitivity analyses. 3,4 5.2 Aims and Hypotheses Revisited In the first chapter, we present the three primary aims of this dissertation and accompanying hypotheses. Here we discuss each objective and evaluate our findings with respect to those objectives. Our first aim was “to determine the medical costs and healthcare resource utilization associated with different post-docetaxel treatment alternatives among patients with progressed prostate cancer.” This was addressed in the second chapter of this dissertation. Using Optum Optum Clinformatics™ Data Mart database, we find the total per-member per month (PPPM) cost burden was $35,747, $56,854, and $53,165 for patients treated with ASTi-ASTi, ASTi-Doc, or ASTi-Cab in a second-line setting and $211,248 and $185,188 among patients treated with ASTi-Doc-ASTi and ASTi-Doc-Cab, in third-line setting. Cancer-related cost-burden reflects 176 these trends across each cohort, albeit at lower levels. Despite how total and cancer-related medical costs among patients treated with an intravenous-based therapy relative to an oral therapy, patient-borne costs were lower for patients in intravenous-based therapy cohorts. This finding is likely due to the difference in prescription and medical benefits. As we also hypothesized, patients treated with cabazitaxel had higher medical and pharmacy expenses than patients treated with other therapies. Although most of that cost was seen by the payer as opposed to the patient. Our second aim was to “develop and deploy a survey instrument to calculate precise estimates of higher order risk attributes with respect to health-state utility,” and then to “analyze the responses captured in a US representative sample to calculate relative risk aversion, relative risk prudence and relative risk temperance.” Several qualitative pilot surveys and online pilot surveys were administered to best develop our investigational survey to elicit estimates of higher order risk parameters. These pilots were based on similar previous studies, aimed at identifying risk behavior over financial-based lotteries or other health scenarios. 1,2,5–8 After successful development, our survey was administered among a representative sample using the Understanding America Study sample. We found respondents demonstrated lower relative risk aversion, relative risk prudence, and relative risk temperance relative over questions pertaining to health conditions and general health than those pertaining to cash. However, we did find an estimate that potentially can convert previously established financial-based higher order risk estimates to estimates over health. Our last aim was to “conduct a classical and novel economic evaluation, comparing cabazitaxel to an alternative ASTi in patient previously treated with docetaxel and an alternative ASTi, using the GRACE modelling framework to account for higher order relative risk 177 preferences over general health values, value of information, value of hope, and value of uncertainty.” We found that cabazitaxel was not a cost-effective alternative to an ASTi among the modeled population using ICER at a willingness-to-pay threshold of $100,000/quality- adjusted life year (QALY) or $200,000/QALY or GRACE at a 𝐾 𝐺 𝑅 𝐴𝐶𝐸 derived from a willingness-to-pay threshold of $100,000/QALY. These findings confirm previous studies executed on this topic but reject our hypothesis that cabazitaxel may be a cost-effective alternative to an ASTi. 3,4 This largely due in part to the higher costs associated with cabazitaxel therapy relative to an ASTi, for little gain in QALY. However, we did find incremental GRACE ratio (IGRACER) was cost-effective at a 𝐾 𝐺 𝑅 𝐴𝐶𝐸 derived from $200,000/QALY. This finding suggests GRACE’s ability to account for treatment uncertainty, patient risk preference, value of hope, and value of insurance and disease severity does add another dimension that may find therapies for severe diseases to be cost-effective when they may not have been found cost- effective assuming the classic framework. With the rigorous inclusion of elements of value as presented by the value flower, a therapy that was not cost-effective under the original ICER model at a willingness-to-pay threshold of $200,000/QALY, is now seen as cost-effective at the 𝐾 𝐺 𝑅 𝐴𝐶𝐸 derived from that same willingness-to-pay threshold. This indicates that through the inclusion of the novel value elements listed here, there is value seen with cabazitaxel that was not previously captured. However, research still needs to be done to examine where that value is coming from during execution. Now that GRACE has been executed, we can confidently say that those novel elements of value do have an impact on gauging the incremental value of a new medical technology. Under this objective, we also hypothesized that 𝐾 𝐺 𝑅 𝐴𝐶𝐸 would be higher among this patient population given disease severity than our ICER willingness-to-pay thresholds. We found 178 that was not the case. However, we did learn that 𝐾 𝐺 𝑅 𝐴𝐶𝐸 cannot be compared in a 1:1 setting to the classic willingness-to-pay threshold. The dynamic of GRACE does not only adjust the willingness-to-pay threshold for person preferences, but also adjusts ICER for those preferences. So the comparison between 𝐾 𝐺 𝑅 𝐴𝐶𝐸 and the ICER willingness-to-pay threshold, should actually be a comparison of the relationship between IGRACER and 𝐾 𝐺 𝑅 𝐴𝐶𝐸 , and ICER and the willingness-to-pay threshold. 5.3 Policy Implications Prostate Cancer (PC) is a leading cause of male cancer deaths in the United States. 9,10 MCRPC is a leading contributor to these deaths, and novel therapies are necessary to ease the pain and suffering of patients and caregivers. 11–13 Among the treatment options that are available, appropriate sequencing also needs to be identified given necessary tradeoffs between one treatment or another. 13 However, common later line options for patients with mCRPC incorporate an important tradeoff that extends beyond survival or cost. Patients selecting an ASTi may be receiving a drug with a more favorable side effect profile but are burdened with a drug they must take daily, has a worse impact on overall survival relative to cabazitaxel in a similar treatment setting, and has less associated of hope of a miracle survival improvement. 14,15 Whereas those taking cabazitaxel in the same setting might see a greater gain in survival, but are burdened with side effects associated with chemotherapy, have greater uncertainty to the treatment outcomes, and must receive intravenous infusions which may represent a burden on the patient’s quality-of-life. 14,15 Some of these treatment attributes can only be modeled using the classic ICER under specific scenario analyses. However, the GRACE framework incorporates many of these attributes, such as treatment uncertainty, value of hope, and a newly characterized 179 attribute, relative risk preferences for treatment outcomes. This dissertation better informs the tradeoffs necessary between therapeutic options for mCRPC, generates relative risk estimates over a general health measure, and applies each of these outcomes along with published literature to inform healthcare decision-makers and patients dealing with this decision problem. Chapter two showed the costs associated with different treatment options among a second line or third line setting for mCRPC patients. Although the cost findings have an implication in contributing to the economic evaluation conducted in Chapter four of this dissertation, another critical finding has an important impact for clinical decision-making. Despite greater overall cost for intravenous treatment options, patient born out-of-pocket (OOP) costs were lower for those intravenous options relative to the oral treatments, abiraterone and enzalutamide. OOP costs have a notable impact on treatment adherence and persistence. 16 Knowing this, the decision between cabazitaxel or abiraterone extends beyond overall survival or patient quality-of-life. If cost precludes a patient from proper adherence, then overall survival or quality-of-life due to therapy starts to have lower importance. Access is still critical for proper treatment and maintenance of mCRPC patient disease progression, and chapter two demonstrates access might be hindered by out-of-pocket costs for oral therapies. Payers should also consider this element in the context of total treatment costs. If total treatment cost covered by the payer is higher for intravenous therapies relative to oral therapies, but out-of-pocket costs are higher for oral therapies relative to intravenous therapies, payers may be artificially selecting patients into intravenous treatment options, which could lead to higher experienced cost by the insurer. These findings could be critical for benefit design for patients dealing with the option of oral or intravenous therapies for cancer treatment. 180 Cost burden examined in chapter two was used to populate an a classic ICER economic evaluation and novel GRACE economic evaluation in chapter four. However, to execute the GRACE economic evaluation, higher order relative risk estimates pertaining to health had to be captured and estimated over a general population survey in chapter three of this dissertation. To anchor responses over general health and four different treatment conditions varying in perceived intensity, estimates were also generated using prototypical cash-based lotteries. These higher order risk estimates over a financial-based lottery allows us to generate the necessary parameters, relative risk aversion and relative prudence, to provide an updated willingness-to-pay threshold for health-based technologies using the formulas derived in Phelps & Cinatl (2021). 17 These estimates came out to $83,611/QALY and $88,256/QALY, assuming the mean and median estimates are representative of the broader population. The findings provide support for previously derived estimates of willingness-to-pay near $100,000k/QALY. 18,19 Another important implication of these higher order risk estimates is contributing to the calculation of GRACE key outputs. Now with a published estimate of higher order relative risk estimates over health, healthcare economic analysts can now incorporate new elements into their economic evaluations. As the classic ICER is improved and new economic evaluation frameworks are developed, rigorously executed examples of those frameworks will be critical for educating the field. Chapter four represents one of the first executions of the GRACE framework. Presented in parallel with the ICER framework gives analysts a benchmark in generating their own GRACE studies. With this economic evaluation, we hope that future analysts can look to this model as a do-it-yourself example for their own studies. Furthermore, we also display a few different 181 variations in GRACE-critical parameters to demonstrate GRACE flexibility and to better inform GRACE interpretation. Finally, our goal as a health economist and analysts is to better inform clinical decision- making for the improvement of patient care given scare resources available for treatment options. Our economic evaluation using both the classical ICER and novel GRACE provides support for the continued use of abiraterone or enzalutamide relative to cabazitaxel, considering cost and clinical outcomes. However, the incorporation of both GRACE and ICER models, and several sensitivity analyses should help illustrate when either treatment is appropriate for mCRPC patients and with what certainty clinicians, patients, and decision-makers can have when deciding between treatment options. Until greater treatment options come online for patients with mCRPC in a later line setting, finding the current best treatment sequence given the requisite tradeoffs is critical for improving patient care, without notable improvements in therapy beyond the currently available options. 182 5.4 References 1. Holt CA, Laury SK. Risk aversion and incentive effects. American economic review. 2002;92(5):1644-1655. 2. Noussair CN, Trautmann ST, van de Kuilen G. Higher Order Risk Attitudes, Demographics, and Financial Decisions. The Review of Economic Studies. 2014;81(1):325- 355. doi:10.1093/restud/rdt032 3. Barqawi YK, Borrego ME, Roberts MH, Abraham I. Cost-effectiveness model of abiraterone plus prednisone, cabazitaxel plus prednisone and enzalutamide for visceral metastatic castration resistant prostate cancer therapy after docetaxel therapy resistance. J Med Econ. 2019;22(11):1202-1209. doi:10.1080/13696998.2019.1661581 4. Zhang PF, Xie D, Li Q. Cost-effectiveness analysis of cabazitaxel for metastatic castration resistant prostate cancer after docetaxel and androgen-signaling-targeted inhibitor resistance. BMC Cancer. 2021;21(1):35. doi:10.1186/s12885-020-07754-9 5. Attema AE, l’Haridon O, van de Kuilen G. Measuring multivariate risk preferences in the health domain. Journal of Health Economics. 2019;64:15-24. doi:10.1016/j.jhealeco.2018.12.004 6. Attema AE, Frasch JJ, L’Haridon O. Multivariate risk preferences in the quality-adjusted life year model. Health Economics. 2022;31(2):382-398. doi:10.1002/hec.4456 7. Attema AE, Krol M, van Exel J, Brouwer WBF. New findings from the time trade-off for income approach to elicit willingness to pay for a quality adjusted life year. Eur J Health Econ. 2018;19(2):277-291. doi:10.1007/s10198-017-0883-9 8. Gonzalez R, Wu G. On the Shape of the Probability Weighting Function. Cognitive Psychology. 1999;38(1):129-166. doi:10.1006/cogp.1998.0710 9. Street W. Cancer Facts & Figures 2018. Published online 1930:76. 10. Yabroff KR, Lund J, Kepka D, Mariotto A. Economic Burden of Cancer in the United States: Estimates, Projections, and Future Research. Cancer Epidemiol Biomarkers Prev. 2011;20(10):2006-2014. doi:10.1158/1055-9965.EPI-11-0650 11. Arap W, Pasqualini R, Costello JF. Prostate Cancer Progression and the Epigenome. New England Journal of Medicine. 2020;383(23):2287-2290. doi:10.1056/NEJMcibr2030475 12. Siegel DA. Prostate Cancer Incidence and Survival, by Stage and Race/Ethnicity — United States, 2001–2017. MMWR Morb Mortal Wkly Rep. 2020;69. doi:10.15585/mmwr.mm6941a1 183 13. George DJ, Sartor O, Miller K, et al. Treatment Patterns and Outcomes in Patients With Metastatic Castration-resistant Prostate Cancer in a Real-world Clinical Practice Setting in the United States. Clinical Genitourinary Cancer. 2020;18(4):284-294. doi:10.1016/j.clgc.2019.12.019 14. de Wit R, de Bono J, Sternberg CN, et al. Cabazitaxel versus abiraterone or enzalutamide in metastatic prostate cancer. New England Journal of Medicine. 2019;381(26):2506-2518. 15. Eliasson L, de Freitas HM, Dearden L, Calimlim B, Lloyd AJ. Patients’ Preferences for the Treatment of Metastatic Castrate-resistant Prostate Cancer: A Discrete Choice Experiment. Clinical Therapeutics. 2017;39(4):723-737. doi:10.1016/j.clinthera.2017.02.009 16. Caram MEV, Oerline MK, Dusetzina S, et al. Adherence and out-of-pocket costs among Medicare beneficiaries who are prescribed oral targeted therapies for advanced prostate cancer. Cancer. 2020;126(23):5050-5059. doi:10.1002/cncr.33176 17. Phelps C, Cinatl C. Estimating Optimal Willingness to Pay Thresholds for Cost- Effectiveness Analysis: A Generalized Method. Health Economics. 2021;(Forthcoming). 18. Vanness DJ, Lomas J, Ahn H. A Health Opportunity Cost Threshold for Cost-Effectiveness Analysis in the United States. Ann Intern Med. Published online November 3, 2020. doi:10.7326/M20-1392 19. Ryen L, Svensson M. The Willingness to Pay for a Quality Adjusted Life Year: A Review of the Empirical Literature. Health Econ. 2015;24(10):1289-1301. doi:10.1002/hec.3085

Abstract (if available)

Abstract Advanced prostate cancer is a leading killer of American males. Despite the development of new treatments, cost and clinical burden of metastatic, castrate-resistant prostate cancer (mCRPC) is staggering among patients and our health system. Recent work identifies optimal treatment sequences for those with mCRPC. In 2019, investigators leading the CARD clinical trial identified cabazitaxel as a potential treatment alternative to the standard of care, an alternative androgen-signaling targeted inhibitor (ASTi, abiraterone or enzalutamide), among mCRPC patients previously treated with docetaxel and an ASTi, enzalutamide or abiraterone.
This dissertation had three primary research objectives, aimed at better informing an economic analysis of the treatment decision problem facing clinicians, patients, and other healthcare decision-makers. In the first chapter, we provide context regarding this specific decision problem and outline the necessary steps to conduct a thorough economic evaluation, using both the classic incremental cost-effectiveness ratio (ICER) analysis and the novel, generalized risk-adjusted cost-effectiveness (GRACE) framework. In the second chapter, we evaluate the economic burden facing patients treated with the therapies often seen in the decision problem presented here. In the third chapter, we present the results of a representative survey aimed at eliciting higher order risk estimates of patients facing treatment decisions over general health and a variety of treatment conditions pertaining to health and financial-based lotteries. In the fourth chapter, we present classic and novel economic evaluations of cabazitaxel relative to an ASTi among mCRPC patients treated with docetaxel and an alternative ASTi. Finally in the fifth chapter, we present our key findings in the context of hypotheses presented in chapter one and discuss critical policy implications of this research.
In this dissertation, we find that the economic burden for the treatment of mCRPC is substantial for patients undergoing either with an oral or intravenous-based chemotherapy regimen, with the intravenous-based regimen leading to greater overall per-patient per month costs. Despite this, patient out-of-pocket expenses are still greater for oral-based regimens, suggesting the need for a re-evaluation of medical and prescription benefits by insurers. From there, we characterize higher order risk estimates pertaining to health-care decision making. These estimates provide values that can play a critical role in re-appraising an appropriate willingness-to-pay threshold and feed into the novel GRACE framework for economic evaluations of medical technology in the United States. Finally, using values generated in the economic burden analysis, our representative survey, and published literature, we find that cabazitaxel is not a cost-effective alternative to an ASTi, among mCRPC patients treated with docetaxel and an alternative ASTi in most instances. However, the GRACE framework, with the inclusion of new parameters surrounding patient preference and drug effectiveness does provide some instances in which there may be a net benefit from cabazitaxel treatment in our modeled patient population. Publication of these higher order risk estimates and this early execution of GRACE will hopefully inform future executions of this novel cost-effectiveness analysis framework.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays in health economics

PDF

Crohn’s disease: health outcomes and resource utilization in the biologic era

PDF

Essays on the economics of infectious diseases

PDF

Evaluating cancer treatments with health economics methods

PDF

Advances and applications for economic evaluation methods in health technology assessment (HTA)

PDF

Economic aspects of obesity

PDF

Essays on access to prescription drugs

PDF

Documenting the impact of under-treatment for chronic disorders

PDF

Disease modifying treatments for Alzeimer's disease: modeling, value assessment and eligible patients

PDF

The impact of treatment decisions and adherence on outcomes in small hereditary disease populations

PDF

Value in oncology care and opportunities for improvement

PDF

Value in health in the era of vertical integration

PDF

Physician behavior and low-value care

PDF

Assessing value defects in limb preservation care

PDF

Outcomes of antibiotic use among children with acute respiratory tract infections

PDF

The impact of healthcare interventions using electronic health records: an evaluation within an integrated healthcare system

PDF

The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins

PDF

Burden of illness in hemophilia A: taking the patient’s perspective

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

Selected papers on the evaluation of healthcare costs of prematurity and necrotizing enterocolitis using large retrospective databases

Asset Metadata

Creator Crawford, Samuel Austin (author)

Core Title Evaluating treatment options for metastatic, castration-resistant prostate cancer: a comprehensive value assessment

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Health Economics

Degree Conferral Date 2022-08

Publication Date 07/22/2022

Defense Date 06/06/2022

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Abiraterone,Cabazitaxel,cost-effectiveness analysis,Docetaxel,Enzalutamide,Grace,healthcare resource utilization,higher order parameters,modeling utility over health,OAI-PMH Harvest,prostate cancer,relative risk aversion,relative risk preference

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Padula, William (committee chair), Gross, Mitchell (committee member), Lakdawalla, Darius (committee member)

Creator Email crawfosa@usc.edu,samuac94@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111373911

Unique identifier UC111373911

Legacy Identifier etd-CrawfordSa-10904

Document Type Dissertation

Format application/pdf (imt)

Rights Crawford, Samuel Austin

Type texts

Source 20220722-usctheses-batch-960 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu