Working Papers

What Passed Is Past? The Role of Recent Adverse Events in Physician Treatment Decisions [UNDER REVIEW]

In many areas of medical care, physician treatment decisions are made under substantial uncertainty. In the context of obstetrics, this uncertainty may predispose physicians to use decision-making heuristics when choosing between a cesarean or a vaginal mode of delivery. In this study, I examine whether obstetricians overreact to a prior patient’s adverse obstetric events when making subsequent delivery-mode choices, and if so, its effect on patient welfare. Using electronic health record data from a large academic hospital, I find that experiencing adverse obstetric events in one delivery-mode makes the physician more likely to switch to the other – and likely inappropriate–delivery-mode on the next patient, regardless of patient indication. This physician switching response to prior adverse events also results in worse patient outcomes and greater resource use for the subsequent patient. Informed by a simple model of Bayesian updating, I formally test and reject the hypothesis that observed physician switching behavior is consistent with Bayesian learning, and conclude that it is likely an overreaction to salient, negative events. These results highlight a clinical decision that is susceptible to cognitive bias, which may inform future efforts to improve the quality of obstetric decision-making through increased awareness or policy interventions.

MEDIA MENTIONS: New York Times, LA Times

Strained and Constrained: How ICU Capacity Affects Physician Decisions (with David Howard and Thomas Valley)

Limited ICU capacity has been said to exacerbate several consequences of COVID-19, such as mortality, misallocation of resources by physicians, and harmful spillovers on nonCOVID patients. However, whether greater ICU capacity would have avoided these pitfalls is not clear. The goal of this paper is to shed light on how ICU capacity affects physician decision-making and patient welfare when the healthcare system is strained. We use two sources of variation in ICU capacity to estimate its causal effect on physician thresholds for admitting patients to the ICU: i) ICU “expansions”, resulting from increases in hospital number of ICU beds, and ii) ICU “strain”, resulting from random fluctuations in ICU bed availability. Our analysis uses 100% inpatient EHR data from two hospitals (150K encounters. 2015-18) both of which expanded their ICUs separately. Importantly, we use lab test results to assign patients an objective and validated measure of ICU-need, called the eSOFA score. We motivate our empirical analyses using a model of physicians’ dynamic admission decisions when there is uncertainty about a patient’s ICU-need. Results show that increases in capacity cause physicians to lower their thresholds for ICU admission, with mixed effects on patient welfare. Patients are 1.8 pp more likely to be admitted to the ICU post-expansion (even when ICU strain is at pre-expansion levels), but there is significant heterogeneity in likelihood of admission and in-hospital mortality by ICU strain and ICU-need. Expanding ICU capacity does not always help the patients who need it most, and at times harms patients who need it least. Finally, the greatest benefits of ICU expansions accrue to patients in the general wards by allowing physicians to better allocate non-ICU resources, especially when the hospital is operating at capacity.

Who Cares? Heterogeneity in Physician Response to Quality Report Cards

Physician quality “report cards” are popular measures used by states and insurers to, amongst other things, incentivize physicians to improve care quality by publicly evaluating them on specific clinical outcomes. However, research suggests that report cards are less effective than expected, inducing only small improvements in care quality. The goal of this study is to examine whether this lackluster effect occurs due to heterogeneity in physician response to report cards, which would help inform the design of future quality improvement interventions. I examine this question by exploiting the release of the ProPublica Surgeon Report Card, which was released only in 2015 and graded all surgeons who performed eight elective Medicare surgeries on outcomes such as in-hospital mortality and readmission rates. I estimate the within-physician effects of the report card on physician behavior for Medicare and non-Medicare patients with a combined difference-in-difference and regression-discontinuity research design, using inpatient administrative data from the state of Florida (~3Mil encounters, 2012-17). I then identify physicians who are more or less likely to respond to the report card (“learners” and “non-learners”, respectively) using each physician’s past history of learning from their own and others’ successes and failures in the eight key surgeries. Results show that, on average, the report card improved physicians’ performance on key outcomes for the eight Medicare surgeries compared to other elective surgeries, and had positive spillovers on non-Medicare surgeries. This effect persisted several years after the release of the report card, suggesting that even solitary information shocks can have lasting effects on physician behavior. Examining heterogeneous responses, I find that physicians’ past learning propensities positively predicted the extent to which they responded to report card information, though this effect was moderated by several market-level factors.

Published Papers

Druss, B. G., Singh, M., et al. (2018). Peer-led self-management of general medical conditions for patients with serious mental illnesses: A randomized trial. Psychiatric Services, 69(5), 529-535.

Rudolf, V. H., & Singh, M. (2013). Disentangling climate change effects on species interactions: effects of temperature, phenological shifts, and body size. Oecologia, 173(3), 1043-1052.