Table 3.

Sensitivity Analyses of Hospital-Level Variation in ICU Use in Specific Diagnoses and Procedures

	Pneumonia	Congestive Heart Failure	Surgery for Colorectal Cancer	Acute Myocardial Infarction	Excluding Acute Care Transfers	Specification Test†
Random effects variance*
Hospital-level variance (95% CI)	0.27 (0.20, 0.38)	0.64 (0.48, 0.87)	0.29 (0.19, 0.44)	1.24 (0.89, 1.75)	0.57 (0.43, 0.77)	0.03 (0.02, 0.05)
Intraclass correlations** %
Hospital (95% CI)	7.7 (5.4, 10.0)	16.4 (12.2, 20.6)	8.0 (4.9, 11.2)	27.5 (20.7, 34.2)	14.9 (11.1, 19.6)	0.9 (0.5, 1.3)
Residual	92.3	83.6	92	72.5	85.1	99.1
Hospitals	90	90	89	88	90	42
Observations	36,525	76,439	10,232	22,086	981,341	581,503

Notes: All analyses derive from Model 3 refit in after restriction to complete cases; critical illness diagnosis codes eliminated from adjustment variables for pneumonia and surgery models due to model convergence.

^†Specification test, where outcome replaced patient-level marital status (married—yes/no), fit only among Maryland state records.

^*The residual (patient) variance in a mixed logistic regression model is fixed at π^(2/3).

^**ICC calculated as proportion of hospital variance divided by total variance, where total variance is the sum of hospital variance and residual variance π^(2/3).

ICU, intensive care unit.