Table 4.

Cox proportional hazard regression estimating the relationship of log of endothelin-1 level at the baseline exam with clinical outcomes (n=1,361)

Hazard Ratio per log increase in endothelin-1	level	(95% CI)	p-value
Hazard of heart failure or cardiovascular death
Unadjusted	0.09	(0.01, 0.73)	0.03
Limited model*	0.06	(0.00, 1.03)	0.05
Full Model†	0.07	(0.00, 1.30)	0.08
Hazard of heart failure
Unadjusted	0.04	(0.00, 1.20)	0.06
Limited model*	0.10	(0.01, 1.90)	0.13
Full Model†	0.14	(0.01, 2.91)	0.20
Hazard cardiovascular death
Unadjusted	0.02	(0.00, 2.09)	0.10
Limited model*	0.06	(0.00, 3.01)	0.16
Full Model†	0.07	(0.00, 3.21)	0.17
Hazard of non-cardiovascular death
Unadjusted	0.37	(0.02, 6.54)	0.50
Limited model*	0.60	(0.07, 5.09)	0.64
Full Model†	0.76	(0.08, 6.83)	0.80

Definition of abbreviations: CI-confidence interval

Because initial models suggested a non-proportional hazard, all models accounted included a term accounting for the possibility of a time-varying relationship between endothelin-1 and the hazard of heart failure or death. Three influential endothelin-1 outliers were excluded from the primary analysis. Inclusion of these outlier strengthened the association with mortality

^*Limited model: age, sex, race/ethnicity, height and weight, study site

^†Full model: Limited + education, smoking status, pack-years, hypertension, systolic blood pressure, diabetes, and cholesterol