<div> This dissertation consists of three essays examining the nature of pricing in the pharmaceutical industry and the behavior of physicians prescribing drugs. I use a combination of structural modeling and reduced-form econometric techniques to illuminate how factors such as bargaining, competition, and network membership can affect prices and prescribing behavior. Ultimately, these insights can be used to influence public policy goals such as reducing prescription drug costs for patients or limiting unnecessary prescribing. </div><div> </div><div> In Chapter One, which is joint work with Sebastian Linde and Ralph Siebert, I focus on the determinants and effects of bargaining power on wholesale pharmaceutical drug prices. We estimate a structural bargaining model and find that large differences in bargaining power explain drug price heterogeneities across buyers, drug classes, and time periods. Our results show that transaction-specific determinants between buyers and sellers (such as transaction volume, buyer's loyalty, multiple drug purchases from the same seller, etc.) exert strong effects on buyer bargaining power and drug prices. Our counterfactuals show that group purchasing organizations achieve price reductions that vary across drug classes and that these price reductions primarily depend on buyer price sensitivity.</div><div> </div><div> In the second chapter, joint with G\"unter Hitsch, Sebastian Linde, and Ralph Siebert, I turn to the retail prescription drug market. Here, we show that there is a significant amount of price variation for prescription drugs in the retail pharmaceutical market. Both negotiated prices (price between retail pharmacies and third-party insurers) and out of pocket prices (prices between retail pharmacies and insured patients) for a drug exhibit a high degree of price variation even when controlling for drug manufacturer, geographic location, pharmacy chain, etc. Furthermore, the nature of this price variation changes depending on if a drug is branded or generic. </div><div> </div><div> In the third chapter, joint with Svetlana Beilfuss and Sebastian Linde, I examine the problem of antimicrobial resistance and how physician membership in Accountable Care Organizations (ACOs) can influence antibiotic prescribing behavior. We use a two-part structural model that accounts for selection into treatment (the ACO group), and non-treatment (control group). We then compare physician antibiotic prescribing across these groups with adjustment for volume, patient, physician, and institutional characteristics. We find that ACO affiliation reduces antibiotic prescribing by about 23\% per year. Furthermore, we show that failure to account for selection into treatment results in an understating of the average treatment effect.</div>

Clinical guidelines developed from randomized controlled trials (RCT) recommend standard treatments for physicians to treat their patients. However, RCT are usually conducted among younger or healthier populations. Patients who did not participate in clinical trials, such as the elderly or patients with comorbidities, might not be suitable for the standard treatments; instead non-standard treatments can be an alternative treatment option to provide clinical benefits. Physicians are key stakeholders in determining the use of non-standard treatments in clinical practice. While a number of studies have reported on the use of non-standard treatments, little is known about factors associated with a physician’s decision to use non-standard treatments and which information sources are associated with their use. The objectives of this study were to identify factors associated with a physician’s decision to use non-standard treatments and to investigate which information sources were associated with their use. This study applied Rogers’ theory of diffusion of innovation to posit that a physician’s decisions to use non-standard treatments are a function of 1) the perceived advantages of non-standard treatments (effectiveness and toxicity), 2) the sources of information (scientific sources, professional contacts, patient demands, and commercial sources), 3) physician characteristics (years of practice, specialty, innovativeness, and practice experience), and 4) practice settings (practice location, academic affiliation, types of facility, and practice size). This study implemented a convergent parallel mixed-method approach consisting of interviews and surveys to address the objectives. For this study, a convenience sample of 10 medical oncologists was interviewed in January and February 2017. Interviews were transcribed and coded using a coding system based on the theoretical model of this study. Case reports were created to summarize each interview. The content analysis and multi-case analysis were both conducted to describe variable-level factors and contrast and compare factors within and across groups. Surveys were distributed to 1,500 medical oncologists and hematologists who currently practice in eight states across the Midwest. Dillman’s tailored design method was used to guide survey development and administration. The survey examined oncologists’ use of non-standard treatments in elderly patients with diffuse large B-cell lymphoma (DLBCL). Factorial survey design was applied to construct six hypothetical patient vignettes representing a variety of patient age and comorbidity. The dependent variable was whether a physician recommended a non-standard treatment to each vignette. Independent variables were selected from the theoretical model. The descriptive and frequency statistics were conducted for each survey item. The reliability tests were used to evaluate internal consistency of multi-item measures. Generalized Estimating Equations (GEE) were used to test the influence of factors on a physician’s decisions to use non-standard treatments. Among ten interview participants, three were open and five were intermediate open toward using non-standard treatments. Approximately 41.5% of survey participants recommended non-standard treatments for two or more vignettes. Both interviews and surveys showed that sources of information were key factors affecting oncologists’ use of non-standard treatments. In particular, interviewed oncologists used various information sources to justify their use of non-standard treatments such as early phase clinical trials or colleagues’ suggestions. Survey data showed that oncologists who placed higher importance on scientific sources were less likely to use non-standard treatments although scientific or medical journals were the top sources where they learned about rituximab with non-anthracyclines. In vignettes involving patients with rheumatoid arthritis with chronic neutropenia, those who placed a higher importance of professional sources were more likely to use non-standard treatments. Additionally, interview data showed that oncologists who have a sub-specialty, practice in academic settings, have high patient volume, have positive past experience with non-standard treatment and were aware of colleagues’ use non-standard treatments were relatively more open toward using non-standard treatments. Survey data showed that oncologists who agreed that rituximab with non-anthracyclines has a safer toxicity profile than rituximab with anthracyclines, those who commonly encounter younger patients, and those who had more years since graduating from medical school were more likely to use non-standard treatments. In addition to patient characteristics such as comorbidity, physicians’ characteristics and their sources of information are influential to the decision of using non-standard treatments.

本文透過健保資料精神疾病歸人檔，擷取1997～2004年門診中之精神分裂症患者，針對其中701位精神科專科醫師對患者的精神分裂症一、二代藥物之處方行為進行研究。迴歸結果顯示醫師性別對處方選擇沒有特別的影響，同儕效果在相當程度上對醫師產生蠻大的改變。藥價調整前，教學醫院的醫師會受到周遭醫師開藥行為影響，其受影響程度大於非教學醫院。藥價調整後，教學醫院醫師幾乎不受同儕影響，相對的，非教學醫院比起藥價變動前，更易受到同僚行為而改變其行為。不過大體而言當周遭醫師容易開新藥（舊藥）時，醫師本身也會較容易開第二代（第一代）藥物。

<div>This dissertation consists of three chapters. The first chapter, Inertia and Switching in Health Insurance Plans, seeks to examine health insurance choice of families and individuals employed by a large Midwestern public university during the years 2012-2016. A growing number of studies indicate that consumers do not understand the basics of health insurance, make inefficient plan choices, and may hesitate to switch plans even when it is optimal to do so. In this study, I identify what are later defined as unanticipated, exogenous health shocks in the health insurance claims data, in order to examine their effect on families' plan choice and switching behavior. Observing switches into relatively generous plans after a shock is indicative of adverse selection. Adverse retention and inertia, on the other hand, may be present if people remain in the relatively less generous plans after experiencing a shock. The results could help inform the policy-makers about consumer cost-effectiveness in plan choice over time.</div><div> Physicians’ relationships with the pharmaceutical industry have recently come under public scrutiny, particularly in the context of opioid drug prescribing. The second chapter, Pharmaceutical Opioid Marketing and Physician Prescribing Behavior, examines the effect of doctor-industry marketing interactions on subsequent prescribing patterns of opioids using linked Medicare Part D and Open Payments data for the years 2014-2017. Results indicate that both the number and the dollar value of marketing visits increase physicians’ patented opioid claims. Furthermore, direct-to-physician marketing of safer abuse-deterrent formulations of opioids is the primary driver of positive and persistent spillovers on the prescribing of less safe generic opioids - a result that may be driven by insurance coverage policies. These findings suggest that pharmaceutical marketing efforts may have unintended public health implications.</div><div> The third chapter, Accountable Care Organizations and Physician Antibiotic Prescribing Behavior, examines the effects of Accountable Care Organizations (ACOs). Physician accountable care organization affiliation has been found to reduce cost and improve quality across metrics that are directly measured by the ACO shared savings program. However, little is known about potential spillover effects from this program onto non-measured physician behavior such as antibiotic over-prescribing. Using a two-part structural selection model that accounts for selection into treatment (ACO group), and non-treatment (control group), this chapter compares physician/nurse antibiotic prescribing across these groups with adjustment for geographic, physician, patient and institutional characteristics. Heterogeneous treatment responses across specialties are also estimated. The findings indicate that ACO affiliation helps reduce antibiotic prescribing by 23.9 prescriptions (about 19.4 percent) per year. The treatment effects are found to vary with specialty with internal medicine physicians experiencing an average decrease of 19 percent, family and general practice physicians a decrease of 16 percent, and nurse practitioners a reduction of 12.5 percent in their antibiotic prescribing per year. In terms of selection into treatment, the failure to account for selection on physician unobservable characteristics results in an understating of the average treatment effects. In assessing the impact of programs, such as the ACO Shared Savings Program, which act to augment how physicians interact with each other and their patients, it is important to account for spillover effects. As an example of such spillover effect - this study finds that ACO affiliation has had a measurable impact on physician antibiotic prescribing.</div>