Table 3. Multiple logistic regression analyses of the associations between vegetarian status and selected metabolic parameters as outcome variables using the subjective variable selection procedure based on pathophysiologic knowledge.

				Outcome	variable
		Central						High
Covariate	Obesity	obesity	DM	Hypertension	Hyper TG	Hyper LDL-C	Hypo HDL-C	HOMA-IR*
Vegetarian (yes vs. no)	0.56 (0.38–0.81) (0.002)	0.39 (0.26–0.59) (< 0.001)	0.62 (0.30–1.27) (0.190)	0.91 (0.64–1.31) (0.618)	0.84 (0.55–1.29) (0.432)	0.23 (0.12–0.40) (< 0.001)	1.56 (1.08–2.28) (0.019)	0.71 (0.48–1.06) (0.094)
Nagelkerke R 2	0.202	0.165	0.203	0.219	0.129	0.129	0.274	0.138
HL GOF test p value	0.01	0.295	0.484	0.231	0.726	0.679	0.020	0.060
AUC	0.756	0.735	0.817	0.739	0.691	0.735	0.780	0.700
n	706	706	706	706	706	706	706	670

The statistics listed in each cell are the estimate of adjusted odds ratio, (95% confidence interval of adjusted odds ratio), and (p value of Wald test) respectively.

The units and abbreviations are the same as those specified in Table 1.

^*The patients with diabetics were excluded.

The covariates on the variable list to be selected during variable selection procedure for control of confounding bias were those listed in Table 1. Specifically, the adjusted covariates in each multiple logistic regression model were the same as those in the corresponding multiple linear regression model (BMI, waist, glucose levels, systolic blood pressure, log(TG), LDL-C, HDL-C, and logHOMA-IR) in Table 2 respectively. For example, the adjusted covariates for obesity in Table 3 were the same as those for BMI in Table 2.

HL GOF test: Hosmer-Lemeshow goodness-of-fit test, where p value > 0.05 indicated a good fit of logistic regression model to data.

AUC: Area under receiver operating characteristic curve. AUC of 0.7 and above was considered acceptable discrimination.