Table 3.

Multivariate linear regression analysis of the associations between serum 25(OH)D category (predictor variable) and fasting glucose, fasting insulin, HOMA-IR, HOMA-β and QUICKI (outcomes); adolescents aged 10–19 years, Korea National Health and Nutrition Examination Survey (KNHANES) 2009–2010

	Model 1			Model 2			Model 3			Model 4
Dependent variable	β	se	P value	β	se	P value	β	se	P value	β	se	P value
Total participants (n 1466)
Fasting glucose	−0·473	0·245	0·054	−0·462	0·245	0·061	−0·484	0·246	0·050	−0·463	0·245	0·060
Fasting insulin*	−0·064	0·014	<0·001	−0·063	0·014	<0·001	−0·067	0·014	<0·001	−0·063	0·013	<0·001
HOMA-IR*	−0·069	0·015	<0·001	−0·068	0·014	<0·001	−0·072	0·014	<0·001	−0·068	0·014	<0·001
QUICKI*	0·010	0·002	<0·001	0·010	0·002	<0·001	0·010	0·002	<0·001	0·010	0·002	<0·001
HOMA-β*	−0·048	0·015	0·002	−0·047	0·015	0·002	−0·050	0·015	0·001	−0·048	0·015	0·002
HOMA-β†	−0·006	0·012	0·629	−0·005	0·012	0·666	−0·005	0·012	0·674	−0·005	0·012	0·673
Lean participants (n 1158)
Fasting insulin*	−0·045	0·013	<0·001	−0·044	0·013	0·001	−0·047	0·013	0·001	−0·045	0·013	0·001
HOMA-IR*	−0·048	0·014	0·001	−0·047	0·014	0·001	−0·049	0·014	0·001	−0·048	0·014	0·001
QUICKI*	0·007	0·002	0·001	0·007	0·002	0·001	0·007	0·002	0·001	0·007	0·002	0·001
Overweight or obese participants (n 308)
Fasting insulin*	−0·149	0·038	<0·001	−0·155	0·036	<0·001	−0·159	0·036	<0·001	−0·154	0·037	<0·001
HOMA-IR*	−0·163	0·040	<0·001	−0·170	0·038	<0·001	−0·174	0·037	<0·001	−0·168	0·038	<0·001
QUICKI*	0·022	0·005	<0·001	0·023	0·005	<0·001	0·024	0·005	<0·001	0·023	0·005	<0·001

25(OH)D, 25-hydroxyvitamin D; HOMA-IR, homeostatic model assessment–insulin resistance index; HOMA-β, homeostatic model assessment–β-cell function; QUICKI, quantitative insulin-sensitivity check index.

Multivariate linear regression analyses were performed to assess the association between increasing 25(OH)D category (<37·5 nmol/l, 37·5 to <50 nmol/l, ≥50 nmol/l) and the change in fasting glucose, fasting insulin, HOMA-IR, HOMA-β and QUICKI, after adjustment for potential confounders. If there was an interaction between the three categories of serum 25(OH)D and BMI Z-scores (<85th percentile or ≥85th percentile), we split the group into two (lean v. overweight/obese). The modifying effect of overweight status on the relationship between the three categories of serum 25(OH)D and markers of insulin resistance was significant (test for interaction: P=0·046 for fasting insulin, P = 0·030 for HOMA-IR and P = 0·034 for QUICKI). No significant modifying effect of overweight status on the association between the 25(OH)D categories and fasting glucose or HOMA-β was observed. We entered potential confounders as follows: model 1 included age, sex, physical activity (PA) and BMI Z-score; model 2 included age, sex, PA and fat mass (g) Z-score; model 3 included age, sex, PA and fat mass (%) Z-score; and model 4 included age, sex, PA and fat mass index Z-score.

*The analysis was performed using natural log-transformed values for the dependent variables.

†The analysis was performed after additional adjustment of natural log-transformed HOMA-IR.