Table 4.

Race and Socioeconomic Status: Evaluation of QOCeol Items Not Performed and Ratings of Only Items Performed^a

		Patients			Family members
Outcome (QOCeol)	Predictors	b	Ratiob	p	b	Ratiob	p
Number of aspects not performedc	Evaluator race/ethnicityd			<0.001			0.001
	African American	−0.330	0.719	<0.001	−0.529	0.589	<0.001
	Other minority	−0.249	0.780	0.045	−0.300	0.741	0.138
	Evaluator educatione	0.119	1.126	<0.001	0.119	1.126	0.021
	Evaluator incomef	0.131	1.140	<0.001
Average rating of performanceg	Evaluator race/ethnicityh			>0.500			0.006
	African American	−0.057		0.827	−0.868		0.013
	Other minority	−0.420		0.189	−1.068		0.021
	Evaluator educationi	0.055		0.478	−0.244		0.045
	Evaluator incomej	0.069		0.220

^aAssociations were based on cross-classified (evaluator by trainee) hierarchical regression models and with level-1 slopes fixed over cross-classified level-2 units.

^bEvent-rate ratio.

^cModeled as an over-dispersed Poisson outcome, using a nonlinear log link and penalized quasi-likelihood estimation.

^dThere were no confounders of race in either the patient or family model. Because the Poisson model offered no test for the overall significance of race, the reported test was based on a cross-classified linear model. The reported tests for significance of the separate dummy indicators for race were based on Wald's tests from the Poisson model, and were similar to those from the linear model.

^eThere were no confounders of education in the patient model; the family model included adjustment for evaluator race.

^fThere were no confounders of income.

^gAverage was based on aspects the trainee was reported to have performed; modeled as a continuous outcome, using an identity link and full maximum likelihood estimation.

^hThe patient model included adjustment for patient gender, cancer diagnosis, the SF-1 health rating, inpatient/outpatient, trainee age, and recruitment site. The family model included adjustment for recruitment site and inpatient/outpatient.

ⁱThe patient model included adjustment for patient gender, diagnosis of cancer and/or lung disease, inpatient/outpatient, recruitment site, and trainee age. The family model included adjustment for evaluator's race and recruitment site.

^jThe model included adjustment for patient gender and education, cancer diagnosis, inpatient/outpatient, and recruitment site.

QOCeol, end-of-life care communication score from the Quality of Communication questionnaire.