UC Berkeley

Nonprofit teaching hospitals contribute almost half of Health Care and Social Assistance GDP and educate more than 90% of all future physicians. Training physicians involves an important trade-off between the short-term delivery of health services and the long-term benefits of physician training. I leverage unusually detailed electronic health record and audit log data from the emergency department of a large, urban teaching hospital to characterize the static costs of training across a range of granular patient outcomes and process measures. Using panel variation in patient assignment to residents, I find that hospitals must extend length of stay for complex patients by 1% to make a resident 0.053% faster in the future. Over the four-year program, this accrues to a reduction of about 10.3% and implies faster patient throughput. I develop and estimate a dynamic model of physician training and care quality to understand how the emergency department of an academic hospital trades off costs today with the future benefits of more intense physician training. Results inform the policy discourse aimed at improving healthcare efficiency and extend existing models of nonprofit hospitals to account for the teaching objective. I find that commonly-discussed payment reforms for insurers to reduce costs may increase the shadow cost of training. This could have negative effects on the career outcomes of graduating physicians over four times larger than the savings for the teaching hospital, but feasible remedies such as increasing the staffing of attending physicians by 5% lessens the penalty by 65%.

Medicine has a reputation of being a gender-egalitarian profession, but there is also evidence of persistent differences in hours worked as well as procedures and tasks performed. We investigate gender differences in the intensive margin in detail by leveraging a unique dataset that contains granular information based on the Electronic Medical Records and Audit Log at a large teaching hospital. Our primary analysis sample contains 1,620 physicians, of which about 47% are women. In this highly standardized environment, we find that even after controlling for a detailed set of physician attributes, women spend about 10% more time on notes per shift than men. Next, we show that patients quasi-randomly assigned to female physicians upon inpatient hospital admission receive 7.6% fewer orders without any declines in quality of care (readmissions or days in the hospital). Analysis of note text reveals that women include 23% more clinical concepts in their notes. Despite meaningful improvements in clinical efficiency caused by additional note writing effort, physician salary and other measures of career advancement are not correlated with this value-adding task.