There is substantial variation in rates of hospital admissions of Medicare patients among ED physicians at the same hospital that is unrelated to observable patient characteristics, a study found.
Researchers used Medicare fee-for-service claims for a 20% random sample of beneficiaries from Jan. 1, 2012, through Sept. 30, 2015 (when ICD-10 was introduced). They included visits to nonfederal hospital EDs in all 50 states and Washington, D.C., for beneficiaries of any age who were continuously enrolled in Medicare Parts A and B and didn't have end-stage renal disease. The sample was limited to visits for medical (non-surgical) issues, and researchers included the 37 most frequently seen diagnoses. For each visit, they determined whether the patient was discharged from the ED, admitted to the hospital, or admitted to observation status. All transfers to another hospital and visits under observation status (whether in the ED or hospital unit) were considered to be clinically equivalent to an admission. The researchers examined physician-level variation in admission rates using a mixed-effect linear regression model predicting each patient demographic and clinical characteristic as a function of physician random effects and hospital fixed effects. In addition, they examined consistency in physicians' admission rates across clinical conditions. Results were published in the February Health Affairs.
The study sample included 5,778,218 visits with 45,491 physicians at 3,480 EDs. The mean patient age at the time of ED visit was 72.5 years, and patients were predominantly women (58.4%) and White (76.7%). Emergency medicine physicians handled 81.2% of visits, compared with 12.4% for family/internal medicine clinicians and 6.5% for physicians of other specialties. The majority of physicians were men (75.3%). Admission rates varied by diagnosis, from a high of 94.9% for septicemia to a low of 8.3% for “other back problems.” The mean rate of admissions among physicians was 38.9%, but adjusted admission rates varied substantially within hospitals, ranging from 32.2% to 45.6% for physicians at the 10th and 90th percentiles, respectively, of the distribution predicted by the estimated physician-level variance (absolute difference, 13.4 percentage points; 95% CI, 13.31 to 13.44 percentage points).
Across clinical conditions, there was moderate to high correlation in admission rates at the physician level, ranging in magnitude from 0.59 to 0.96 (for example, the correlation between admission tendency was 0.73 for gastrointestinal vs. cardiovascular conditions and 0.81 for pulmonary vs. gastrointestinal conditions), meaning that physicians generally had consistently higher or lower tendencies to admit (relative to other physicians in the same hospital) across conditions. Their rates of admissions for one particular clinical condition were also predictive of their admission rates for other conditions, “suggesting that variation in admission rates reflects physician-level tendencies that are consistent across different types of clinical conditions,” the authors wrote.
Among other limitations, the study could not determine whether the variation in admission rates was associated with differences in patient outcomes, the authors noted. They added that they were unable to ascertain all potentially relevant predictors of admission (e.g., vital signs, presenting symptoms, availability of home support) from administrative claims data.
“The wide variation in ED physicians' admission rates seen in our study suggests that physician decision making contributes considerably to whether a patient in the ED is admitted and might therefore be a fruitful target for interventions” to reduce unnecessary admissions while ensuring patients who could benefit from hospitalization are admitted, the authors concluded.