Physician-level variations in resource use and outcomes in GIM
| Outcome measures | Absolute difference between highest and lowest physician quartiles | ||||
| All admissions (95% CI) | Propensity score matching (95% CI) | Propensity score matching, restricted sample* (95% CI) | Split sample validation† | Single attending physician‡ | |
| In-hospital mortality (%) | 2.4 (0.7 to 4.1) | 2.4 (0.6 to 4.3) | 1.7 (0.3 to 3.0) | 0.3 (−1.3, 2.0) | 2.5 (1.1, 3.9) |
| Length of stay (mean days per admission) | 1.2 (0.6 to 1.9) | 1.2 (0.5 to 1.9) | 1.2 (0.5 to 2.0) | 0.7 (0.04, 1.34) | 1.0 (0.1, 1.8) |
| 30-day readmission (%) | 3.9 (1.2 to 6.7) | 3.3 (0.7 to 5.9) | 4.1 (2.3 to 6.0) | 0.8 (−2.0, 3.6) | 4.1 (0.8, 7.3) |
| Advanced imaging use (mean tests per admission) | 0.35 (0.15 to 0.55) | 0.32 (0.12 to 0.52) | 0.25 (0.03 to 0.48) | 0.26 (0.06, 0.47) | 0.30 (0.1, 0.5) |
Physicians were categorised into quartiles within each hospital based on each outcome measure. We report the difference between the highest and lowest physician quartiles for each measure among all admissions and after matching. In the matched samples, each admission for physicians in the highest and lowest quartiles was matched 1:1 on age, sex, hospital site, fiscal year of admission and a propensity score that included all baseline characteristics.
*Only admissions involving the 10 most common discharge diagnoses were included to avoid case mix differences that might be related to rare diagnoses.
†The 5-year sample was split and physicians were categorised into quartiles based on outcomes in the first 2 years and here we report outcomes in the same quartiles in the last 3 years.
‡Only admissions in which there was a single attending physician throughout were included.
GIM, general internal medicine.