Skip to main content
. Author manuscript; available in PMC: 2011 Sep 27.
Published in final edited form as: Arch Intern Med. 2010 Sep 27;170(17):1557–1565. doi: 10.1001/archinternmed.2010.312

Table 6.

Comparison of Best Parsimonious Prediction Models

Model 1a
 Model 2b
 Model 3c
 Model 4d
HR P Value HR P Value HR P Value HR P Value
Waist circumference per cm 1.09 <.001 1.06 <.001 1.07 <.001 1.10 <.001
History of hypertension 1.58 <.01 1.51 <.01 1.49 <.001 1.52 <.001
History of treated high cholesterol levels 1.78 <.01 1.88 <.01 1.93 <.001 1.88 <.001
Current smoker 1.74 .03
Past smoker 1.49 <.01 1.51 <.01
Alcohol intake 0.80 <.01 0.82 <.01
Family history of diabetes 1.56 <.001 1.55 <.001 1.55 <.001 1.55 <.001
Log fasting blood glucose level per SD 3.24 <.001 3.17 <.001 3.14 <.001 3.20 <.001
Log WBC count 1.68 .05 1.73 .03
Log hsCRP level 1.19 <.01 1.18 .01
ICAM-1 level 1.003 <.01 1.002 <.001
Model 1a Model 2b Model 3c Model 4d
−2 LnL 834.87 808.71 822.55 851.58
BIC 900.78 891.09 888.45 892.77
AUC 0.93 0.93 0.93 0.93 0.93
IDIe 1.000 [Reference] 0.012 0.007 −0.006
NRIe 1.000 [Reference] −0.005 −0.008 −0.008
SE/SPf 0.83/0.91 0.84/0.89 0.83/0.90 0.82/0.91
PPV/NPVf 0.24/0.99 0.22/0.99 0.23/0.99 0.24/0.99
PPV/NPVg 0.18/0.99 0.17/0.99 0.17/0.99 0.18/0.99
PPV/NPVh 0.28/0.99 0.26/0.99 0.27/0.99 0.30/0.99
Excluding Women With Baseline Fasting Glucose Level ≥126 mg/dLi
AUC 0.87 0.88 0.87 0.87
IDIe 1.000 [Reference] 0.026 0.012 −0.013
NRIe 1.000 [Reference] −0.011 −0.007 −0.023
SE/SPf 0.66/0.91 0.70/0.90 0.68/0.90 0.66/0.91
PPV/NPVf 0.21/0.99 0.20/0.99 0.21/0.99 0.21/0.99
PPV/NPVg 0.16/0.99 0.16/0.99 0.16/0.99 0.16/0.99
PPV/NPVh 0.24/0.99 0.23/0.99 0.23/0.99 0.25/0.98

Abbreviations: AUC, area under the receiver operating characteristic curve; BIC, Bayesian information criteria; HR, hazard ratio; hsCRP, high-sensitivity C-reactive protein; ICAM-1, soluble intercellular adhesion molecule 1; IDI, integrated discrimination improvement; LnL, log likelihood; NPV, negative predictive value; NRI, net reclassification improvement; PPV, positive predictive value; SE, sensitivity; SP, specificity; WBC, white blood cell.

SI conversion factor: To convert glucose to millimoles per liter, multiply by 0.0555.

a

Includes clinical risk factors described in Table 4 and fasting glucose level. This model is different from the reference model specified in Table 4 because this is the best parsimonious model and had only the variables retained from the model selection procedure.

b

Includes clinical risk factors described in Table 4, fasting glucose level, and novel plasma markers retained from the model selection procedure.

c

Risk factors that can be objectively measured included all plasma markers and clinical risk factors described in Table 4 except physical activity, cigarette smoking, and alcohol use.

d

Includes risk factors based on the metabolic syndrome and family history of type 2 diabetes mellitus.

e

None of the IDI or NRI was 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. Risk categories for NRI were low (<5%), medium (5%–15%), and high (>15%) for 6-year type 2 diabetes mellitus risk.

f

Cutoff was 15%.

g

Cutoff was 10%.

h

Cutoff was 20%.

i

Model 2 selected for this group included the following covariates: waist circumference, history of hypertension, history of high cholesterol levels requiring medication, past smoking, current alcohol drinking, family history of diabetes, and fasting blood glucose, hsCRP, tumor necrosis factor receptor 2, and ICAM-1 levels. Model 4 selected for this group included the following covariates: waist circumference, history of hypertension, history of high cholesterol levels requiring medication, family history of diabetes, and fasting blood glucose, hsCRP, and ICAM-1 levels. Models 1 and 4 selected for this group included the same covariates as those in the primary analysis for all women (as shown in this table).