Variation in Physician Practice Styles within and across Emergency Departments

Abstract

Despite the significant responsibility that physicians have in healthcare delivery, we know surprisingly little about why physician practice styles vary within or across institutions. Estimating variation in physician practice styles is complicated by the fact that patients are rarely randomly assigned to physicians. This paper uses the quasi-random assignment of patients to physicians in emergency departments (EDs) to show how physicians vary in their treatment of patients with minor injuries. The results reveal a considerable degree of variation in practice styles within EDs; physicians at the 75th percentile of the spending distribution spend 20% more than physicians at the 25th percentile. Observable physician characteristics do not explain much of the variation across physicians, but there is a significant degree of sorting between physicians and EDs over time, with high-cost physicians sorting into high-cost EDs as they gain experience. The results may shed light on why some EDs remain persistently higher-cost than others.

Fig 1. Distributions of Physician and Physician×Experience Fixed Effects for Log(Costs).

Notes: This figure plots the distributions of physician fixed effects and physician×experience fixed effects. The blue dotted line plots physician fixed effects λ_j from Eq (1) and the red solid line plots physician×experience fixed effects λ_jt from Eq (2).

Fig 2. Distributions of Physician Fixed Effects for Predicted vs. Realized Log(Costs).

Notes: This figure plots the distributions of physician fixed effects for predicted versus realized Log(Total Costs). The blue dotted line plots the Bayesian-shrinkage physician fixed effects for predicted Log(Total Costs), where Log(Total Costs) were predicted using patient age, sex, race, ethnicity, and insurance. The red solid line plots the Bayesian-shrinkage physician fixed effects from Eq (1).

Fig 3. Attrition of High-Cost ED Physicians Over Time.

Notes: This figure plots the probabilities that physicians work in any ED following their first quarter of work, where their earliest start date occurs on or after Q1-2005. The blue solid line plots the probability that physicians who are not high-cost work in any ED, while the red dotted line plots the probability that high-cost physicians work in any ED. High-cost physicians are in the top 10% of the Bayesian-shrinkage physician fixed effects distribution for Log(Costs) estimated from Eq (1).

Fig 4. Sorting Between Physicians and EDs Over Time.

Notes: This figure plots the probabilities that physicians work in high-cost EDs following their first quarter of work, where their earliest start date occurs on or after Q1-2005. The blue solid line plots the probability that physicians who are not high-cost work in a high-cost ED, while the red dotted line plots the probability that high-cost physicians work in high-cost EDs. High-cost physicians are in the top 10% of the Bayesian-shrinkage physician fixed effects distribution for Log(Costs) estimated from Eq (1). High-cost EDs are in the top 25% of the distribution for the average total costs of ED visits, where the average total costs are calculated using the entire sample of patients from Table 2.