Table 3.

Multivariate linear regression: covariates associated with each outcome variable^a assessed from baseline through 12 months.

Outcome variableb	Independent variable	Likelihood ratioc	Parameter estimate	Std. error	t value	Pr(>t)
CRP	Androidal fat mass	90.69	0.052	0.005	24.50	≤0.0001
	Site		−0.236	0.121	10.79	0.072
Fibrinogen	Androidal fat mass	31.63	0.012	0.004	4.32	≤0.0001
	Site		−1.469	0.227	−6.48	≤0.0001
	TLMP		−0.118	0.038	−3.13	0.0020
	TLMP × Site		0.159	0.051	3.12	0.0021
	Positive Fam Hist		0.204	0.112	1.82	0.070
Insulin	Androidal fat mass	51.16	0.143	0.024	5.95	≤0.0001
	Site		−4.516	0.588	−7.68	≤0.0001
	Time point 12		2.013	0.650	3.1	0.0022
	Glucose		0.095	0.028	3.4	0.0008
HOMA	Androidal fat mass	43.73	0.012	0.002	6.88	≤0.0001
	Site		−0.312	0.039	−7.92	≤0.0001
	Time point 12		0.011	0.004	3.44	0.0007

^aThe number of observations included in the analysis varied depending on the outcome: N = 217 for CRP and Hcy, N = 218 for fibrinogen, insulin, and HOMA. The Hcy model was not statistically significant and thus not shown in this table.

^bEach final longitudinal model included these obligatory variables: treatment (control versus combined treatment with 80 mg or 120 mg of isoflavones), time point (baseline versus 12, 24, or 36 months), treatment by time point interaction, and site (ISU versus UCD), as well as potential covariates that included androidal fat mass (kg) adjusted for height, time since last menstrual period (TLMP), family history of CVD (FamHx) coded as a categorical variable (none versus positive or none versus unknown), and Healthy Eating Index (HEI) score. The fibrinogen model included TLMP by site interaction. The insulin model (but not other models) included glucose as a covariate. This table shows only those covariates that were significant (P ≤ .05) for each outcome variable or had a tendency (P ≤ .10) to be significant.

^cThe likelihood ratio for each model represents the ratio of a model that includes only obligatory variables compared to the final model that includes obligatory variables and covariates. The overall P-value for each model was ≤.0001.