Table 3.

Linear regression of factors independently associated with BMD

Characteristics	Univariate model
	LS (0.01 g/cm2)			TH (0.01 g/cm2)			FN (0.01 g/cm2)			1/3 Radius (0.01 g/cm2)
	β	r	P	β	r	P	β	r	P	β	Partial r	P
Age (10 yr)	−4.19	0.19	<0.01	−3.68	0.18	0.02	−4.97	0.23	<0.01	−5.91	0.48	<0.0001
BMI (5 kg/m2)	4.18	0.37	<0.0001	5.93	0.54	<0.0001	5.94	0.53	<0.0001	1.89	0.29	<0.0001
AA:Hispanic	12.35	0.4	<0.0001	6.08	0.2	<0.01	8.67	0.28	<0.0001	5.42	0.3	<0.0001
Alcohol use	4.85	0.21	<0.01	3.22	0.14	0.051	2.81	0.12	0.1	1.41	0.11	0.15
OC (ng/ml)	−1.15	0.21	<0.01	−1.54	0.3	<0.0001	−1.17	0.22	<0.01	−0.72	0.23	<0.01
BAP (10 U/liter)	−1.14	0.12	0.12	−2.22	0.23	<0.01	−1.57	0.16	0.03	−1.24	0.22	<0.01
NTx (10 nmol/BCE/liter)	−2.19	0.12	0.1	−3.63	0.2	<0.01	−2.60	0.14	0.05	−1.74	0.16	0.02
TNF-α (pg/ml)	−0.10	0.14	0.08	−0.09	0.14	0.08	−0.08	0.12	0.13	−0.06	0.14	0.08
IL-6 (pg/ml)	0.92	0.18	0.03	0.54	0.11	0.17	0.76	0.15	0.06	−0.06	0.02	0.81
HIV status +/−	−3.64	0.13	0.09	−5.02	0.18	0.01	−3.25	0.11	0.12	−0.34	0.02	0.78
	Full model
	LS: R2 = 0.296, P < 0.0001			TH: R2 = 0.345, P < 0.0001			FN: R2 = 0.353, P < 0.0001			1/3 Radius: R2 = 0.343, P < 0.0001
	β	Partial r	P	β	Partial r	P	β	Partial r	P	β	Partial r	P
Age (10 yr)	−3.20	0.16	0.03	−2.90	0.16	0.03	−3.88	0.21	<0.01	−5.72	0.47	<0.0001
BMI (5 kg/m2)	3.04	0.29	<0.0001	5.28	0.49	<0.0001	5.21	0.48	<0.0001	1.19	0.21	<0.01
AA:Hispanic	9.86	0.33	<0.0001	2.37	0.10	0.16	5.84	0.22	<0.01	3.91	0.25	<0.001
Alcohol use	2.64	0.13	0.09	2.71	0.12	0.10
HIV status +/−	−4.46	0.17	0.02	−3.72	0.15	0.04	−2.27	0.09	0.22	−2.14	0.14	0.05

Simple linear regression parameter estimates and correlation coefficients for covariates found to be significant in univariate screening against BMD, ignoring HIV-status, are presented on the top half of the table for each BMD site. The full model at each BMD site started with race/ethnicity (Hispanic/African-American), corrected calcium, hypertension, history of prior fracture, age at menarche, calcium supplements, alcohol, statin or multivitamin use, and age and BMI forced into the model; partial, semipartial correlations and colinearity were examined, and redundant predictors were removed. Remaining predictors entered a stepwise selection model with criteria of P value of 0.20 to enter and 0.05 to stay in the final model. The parameter estimates and partial-r coefficients reported come from the selected reduced model. The partial-r is the correlation of the target regressor with the dependent variable after the influence of the other predictors in the model has been removed from both the dependent variable and the target predictor. AA, African-American.