Table 4.
Model | −2 LnLa | LR χ2 Statistic | Variable LR χ2 Statistic | BICa | AUC | IDIb |
---|---|---|---|---|---|---|
Clinical risk factorsc+fasting glucose level (reference model) | 823.82 | 1845.47 | 897.96 | 0.93 | ||
+WBC count | 817.84 | 1851.45 | 5.98d | 900.22 | 0.93 | 0.001 |
+TNF-R2 level | 823.60 | 1845.69 | 0.22 | 905.98 | 0.93 | <−0.001 |
+IL-6 level | 815.13 | 1854.17 | 8.69d | 897.51 | 0.93 | 0.002 |
+hsCRP level | 814.33 | 1854.96 | 9.49d | 896.71 | 0.93 | 0.005 |
+WBC count and TNF-R2, IL-6, and hsCRP levels | 807.50 | 1861.79 | 16.32d | 914.60 | 0.93 | 0.005 |
+E-selectin level | 820.76 | 1848.53 | 3.05 | 903.14 | 0.93 | 0.002 |
+ICAM-1 level | 810.73 | 1858.56 | 13.08d | 893.11 | 0.93 | 0.007 |
+VCAM-1 level | 820.03 | 1849.26 | 3.79 | 902.41 | 0.93 | 0.001 |
+E-selectin, ICAM-1, and VCAM-1 levels | 809.42 | 1859.88 | 14.40d | 908.27 | 0.93 | 0.007 |
| ||||||
Excluding Women With Baseline Fasting Glucose Level ≥126 mg/dL | ||||||
Clinical risk factorsc+fasting glucose level (reference model) | 719.81 | 733.13 | 793.95 | 0.87 | ||
+WBC count | 715.74 | 737.20 | 4.07d | 798.12 | 0.87 | −0.002 |
+TNF-R2 level | 718.93 | 734.01 | 0.88 | 801.31 | 0.87 | −0.000 |
+IL-6 level | 712.04 | 740.91 | 7.78d | 794.42 | 0.88 | 0.004 |
+hsCRP level | 712.36 | 740.58 | 7.45d | 794.74 | 0.88 | 0.009 |
+WBC count and TNF-R2, IL-6, and hsCRP levels | 705.88 | 747.06 | 13.93d | 812.97 | 0.88 | 0.009 |
+E-selectin level | 717.04 | 735.91 | 2.78 | 799.42 | 0.87 | 0.003 |
+ICAM-1 level | 707.30 | 745.65 | 12.52d | 789.68 | 0.87 | 0.015 |
+VCAM-1 level | 717.78 | 735.16 | 2.03 | 800.16 | 0.87 | 0.002 |
+E-selectin, ICAM-1, and VCAM-1 levels | 706.84 | 746.10 | 12.97d | 805.69 | 0.87 | 0.015 |
Abbreviations: AUC, area under the receiver operating characteristic curve; BIC, Bayesian information criteria; IDI, integrated discrimination improvement; LnL, log likelihood; LR, likelihood ratio; WBC, white blood cell. Other abbreviations: See Table 1.
SI conversion factor: To convert glucose to millimoles per liter, multiply by 0.0555.
The smaller the model −2 LnL and BIC, the better the model fit (considering the number of covariates in the BIC).
None of the IDI was statistically significantly greater than 0 (P>.05, 1-sided test). The 1-sided test was used because the key question of interest was whether the novel plasma markers improved model prediction.
Clinical risk factors included the matching factors (age and race/ethnicity) and waist circumference, history of hypertension, history of high cholesterol levels requiring medication, physical activity, smoking (never, current, or past), current alcohol drinking (0 indicates <1 drink/mo; 1, <1 drink/wk; 2, 1–6 drinks/wk; and 3, ≥7 drinks/wk), and family history of diabetes (yes or no).
P<.05 in deviance test (likelihood ratio test) compared with the reference model.