Table 3. Risk prediction by continuous environmental risk score (ERS) using single-phenotype approach^a (n = 3847).

Phenotype	Continuous Outcome						Dichotomizedb Outcome
	Model 1c		ERS1d		ERS2e		Model 1c	ERS1d	ERS2e
	R2	PRESSf	R2	PRESSf	R2	PRESSf	AUCg	AUCg	AUCg
Total cholesterol	0.3270	122.46	0.3306	121.88	0.3308	121.85	0.7672 (0.7523, 0.7820)	0.7695 (0.7547, 0.7842)	0.7691 (0.7543, 0.7838)
HDL	0.2636	231.70	0.2677	230.52	0.2665	230.91	0.7193 (0.7024, 0.7362)	0.7217 (0.7050, 0.7385)	0.7208 (0.7040, 0.7376)
LDL	0.1342	539.62	0.1375	537.84	0.1376	537.80	0.7213 (0.7050, 0.7376)	0.7255 (0.7093, 0.7416)	0.7253 (0.7091, 0.7414)
Triglyceride	0.3709	967.24	0.3781	956.76	0.3775	957.70	0.8164 (0.8021, 0.8306)	0.8178 (0.8036, 0.8320)	0.8183 (0.8041, 0.8324)

^aPollutants selected by single-phenotype regression (n = 13, 9, 5 and 27 for total cholesterol, HDL, LDL and triglyceride, respectively) to construct ERS which was computed in the validation data (n = 3847), with adjustment for base covariates and phenotype-specific micronutrients.

^bContinuous phenotypes dichotomized to be high vs. low by thresholds: 200 mg/dL for CHOL, 40 mg/dL (male) or 50 mg/dL (female) for HDL, 130 mg/dL for LDL and 150 mg/dL for TRIG.

^cadjusted for base covariates and phenotype-specific micronutrients.

^dModel 1 plus ERS constructed with coefficient estimates from single-pollutant models as weights.

^eModel 1 plus ERS constructed with coefficient estimates from multi-pollutant models as weights.

^fPredicted residual sums of squares.

^gArea under the receiver operating characteristic (ROC) curve and its 95% confidence interval computed with 2000 stratified bootstrap replicates.