Dissertation Title: "Physician Preferences for Medical Innovation"
Health care spending growth challenges the budgets of governments, businesses, and families. In 2015, one in every six dollars spent in the U.S. economy was related to health care – twice the share in 1980. The rapid growth in health care spending leaves less room for other investments, and this pressure will only increase over time if these expenditures continue to grow as projected. Existing research has shown that innovations in medical technology (e.g., new drugs and procedures) drive health care spending growth. This dissertation explores the role physicians play in the integration of new treatments and procedures into the health care system. It examines physician preferences for new technologies, how these preferences change in response to information shocks, and proposes an approach to disentangle patient demand factors from physician-level proclivity for medical innovation.
Chapter one is a descriptive analysis of the utilization of a broad range of innovative medical technologies. It uses Medicare claims data to identify 46 medical technologies that were new or rapidly diffusing over 2005 to 2010, documents variation in utilization in two categories and three sub-categories across provider organizations and estimates correlations in utilization across these categories within provider organizations. There was substantial variation between provider organizations. The relationship in utilization across categories of technologies within provider organizations, however, was modest. These results suggest provider organizations do not broadly influence the utilization of all types of new medical technology. This implies that payment reforms focused on provider organizations will likely have different effects on the utilization of new technology depending on the type of medical innovation.
Chapter two examines how physician preferences for drugs with uncertain benefits and risks change following a medical reversal of a drug already in use. In May 2007, evidence emerged of cardiovascular risk associated with Avandia (rosiglitazone), one of two products in the thiazolidinedione (TZD) class of oral anti-diabetics. In response, the FDA immediately issued a safety alert, and all drugs in the class were required to carry a black box warning on their labels discouraging use for certain patients. This study linked physicians’ responses to this safety alert with their future prescribing of a new class of direct oral anticoagulants (DOACs). Like TZDs, when DOACs were first approved there was not robust evidence that these new drugs were superior to the existing treatment. We first modeled the probability of prescribing drugs in the TZD class (among all diabetes prescriptions) before and after the safety alert using multi-level logistic regression models that included random effects for individual physicians’ levels of prescribing relative to their peers. We next assessed the relationship between the physician-specific effects and use of DOACs. We found no difference in the use of DOACs based on how physicians responded to a safety alert for drugs in the TZD class. These results suggest that the effects of a medical reversal for pharmaceutical products do not spill over across drugs in different therapeutic areas. Additionally, consistent with previous studies, we found that physicians responded to a safety-related information shock for TZDs, but mostly confined their response to the affected drug. If this were to hold more generally, it suggests evidence can change physician behavior, but to do so broadly, nearly everything would have to be studied.
Chapter three develops a framework for assessing the dimensions along which patients sort randomly to physicians. Many factors influence treatment decisions. Patient preferences pose a specific challenge because they are an input into physician decision-making and are also potentially correlated across types of services and treatments, as well as with outcomes, such as total spending. I examine using an instrumental variable (IV) approach defining a measure of physician preference for novel treatments that is plausibly unrelated to patient demand. The set of descriptive analyses and empirical tests presented in this chapter are intended to evaluate whether inclusion criteria used to define a study population selects a sample that is as good as randomly assigned – the key assumption in an IV approach. Once such a sample has been constructed, the relationship between a physician’s practice patterns and the broader trends in the spending of their patients can be examined. I demonstrate how my proposed instrument performs in the context of prescription diabetes medications for patients receiving care from an endocrinologist.