Skip to main content
. 2022 Jan 25;139(14):2212–2226. doi: 10.1182/blood.2021012830

Table 4.

Mediation analysis of racial/ethnic disparities in leukemia death

N Unmediated model Mediated model Rescaled coefficients
HR (95% CI) HR (95% CI) %Change* %Change P
Baseline Census tract structural racism Mediation proportion estimates
Race/ethnicity
 NH White 482 1.0 1.0
 NH Black 120 1.61 (1.15, 2.26) 1.04 (0.65, 1.66) 94% 83% .006
 Hispanic 117 1.20 (0.83, 1.72) 0.96 (0.65, 1.43) 120% 100% .006
 Other 79 1.27 (0.83, 1.92) 1.24 (0.82, 1.89) 9% None
Race/ethnicity Tumor biology
 NH White 477 1.0 1.0
 NH Black 115 1.51 (1.07, 2.12) 1.42 (1.01, 2.00) 17% 9% .15
 Hispanic 114 1.23 (0.86, 1.77) 1.27 (0.88, 1.83) −15% None
 Other 72 1.28 (0.84, 1.96) 1.25 (0.82, 1.92) 10% None
Race/ethnicity Health care access
 NH White 487 1.0 1.0
 NH Black 125 1.63 (1.17, 2.26) 1.66 (1.18, 2.32) −5% None
 Hispanic 114 1.27 (0.89, 1.81) 1.21 (0.84, 1.75) 21% 22% .09
 Other 82 1.33 (0.89, 2.00) 1.33 (0.88, 1.99) 2% None
Race/ethnicity Treatment
 NH White 479 1.0 1.0
 NH Black 118 1.67 (1.19, 2.34) 1.34 (0.95, 1.90) 49% 47% <.0001
 Hispanic 113 1.25 (0.86, 1.79) 1.07 (0.74, 1.54) 73% 72% .002
 Other 78 1.33 (0.88, 2.02) 1.32 (0.86, 2.03) 4% None
Race/ethnicity Comorbidity
 NH White 356 1.0 1.0
 NH Black 107 1.53 (1.08, 2.17) 1.57 (1.10, 2.23) None None
 Hispanic 85 1.11 (0.74, 1.67) 1.07 (0.71, 1.62) 35% 49% .11
 Other 51 0.84 (0.49, 1.45) 0.86 (0.50, 1.49) 13% None
Race/ethnicity ICU admission at induction (N = 573)
 NH White 315 1.0 1.0
 NH Black 85 1.34 (0.85, 2.09) 1.22 (0.78, 1.92) 34% 26% .01
 Hispanic 88 1.23 (0.77, 1.97) 1.20 (0.75, 1.92) 13% 24% .04
 Other 54 1.30 (0.76, 2.21) 1.18 (0.69, 2.02) 39% 48% .001

Mediation proportion estimates <10% are suppressed (reported as “None”), and P > .20 is suppressed. All models are adjusted for age, sex, and diagnosing hospital.

*

%Change in the excess hazard ratio for the disparity in leukemia death before and after controlling for potential mediation.

Proportionate reduction in the rescaled coefficients before vs after controlling for potential mediation.

P value for the difference in rescaled coefficients before vs after controlling for potential mediation.