This dissertation contains three chapters that examine how physicians make decisions and how those decisions impact patient healthcare utilization and health outcomes.

The first chapter is motivated by this observation that patient referral rates vary dramatically across primary care physicians, but we know relatively little about the drivers of this variation and how this impacts patient health outcomes. In this chapter, I study how physicians with different referring styles impact patient outcomes. To separate the causal effect of physician referring style from patient related factors, I focus on a sample of Medicare patients who switch to a new primary care physician after their original primary care physician exits. I use event study analyses to compare outcomes of patients who switch to new primary care physicians with different referring styles. I find that around 35 − 38 percent of the variation in referrals across primary care physicians is due to physician referring style, with the remainder due to patient factors. Moreover, I find that high-referring physicians are associated with higher healthcare utilization and poor health outcomes. To understand the mechanisms underlying these results, I characterize high-referring physicians along various dimensions. I find that high-referring physicians work in smaller practices, see more patients, and refer their patients to a large number of providers.

In the second chapter I examine the role of market mechanisms through which referring physicians learn about specialist quality. In particular, I examine how patient adverse events affect referrals from referring physicians to cardiac surgeons. I use Medicare data to identify pairs of referring physicians and cardiac specialists who have a patient adverse event within seven days of a major surgical procedure to examine how these events affect referrals. I construct counterfactuals for affected pairs using pairs that experience patient adverse event but five quarters in the future. I find that there is a significant decrease in referrals from the referring physician to the specialist after patient death. Referring physicians with below median number of referrals in the pre-event period and those that work in different practices as specialists respond more. I also examine if the physicians respond differently depending on the patient’s race, but I do not find evidence suggesting that referring physicians respond differently depending on the race of the patient who died.

The third chapter examines the effect of Prescription Drug Monitoring Programs (PDMPs) on opioid use among reproductive age women and on infant health outcomes. PDMPs are state databases that track prescribing and dispensing of controlled substances. To encourage database use, several states have adopted mandatory use policy that requires prescribers to consult these databases before prescribing. I rely on the restricted-use Vital Statistics Natality files for years 2003-2018 to examine the effect of these polices on birth weight outcomes. I use Treatment Episode Data Set-Admissions (TEDS-A) for years 2003-2017 to measure opioid misuse. I estimate the effect of PDMP and mandatory use policy using a difference-in-differences framework - comparing differences in outcomes across states before and after policy adoption. PDMPs are not associated with a significant improvement in infant health outcomes. However, mandatory use of PDMPs is associated with a 0.99 percent decline in the incidence of low birth weight. PDMPs and mandatory use policy have, at best, a very small impact on population-level infant health outcomes. Moreover, there is suggestive evidence that these policies reduce prescription opioid use among reproductive age women, but leads to an increase in heroin misuse.