Table 3.

The best models to predict CAD risk by MDR analyses.

No. of risk factors	Best interaction models	CVCa	Testing accuracy (%)b	Permutation test (P-value)c
1	History of hyperlipidemia	74/100	0.6097	0.0102
2	History of hyperlipidemia, history of T2DM	86/100	0.6228	0.0023
3	History of hyperlipidemia, history of T2DM, smoking status	96/100	0.6442	<0.0001
4	History of hyperlipidemia, history of T2DM, smoking status, SNP rs1136410 (AA + AG/GG)	100/100	0.6773	<0.0001
5	History of hyperlipidemia, history of T2DM, smoking status, SNP rs1136410 (AA + AG/GG), history of hypertension	100/100	0.6547	<0.0001
6	History of hyperlipidemia, history of T2DM, smoking status, SNP rs1136410 (AA + AG/GG), history of hypertension, alcohol drinking status	100/100	0.6480	<0.0001
7	History of hyperlipidemia, history of T2DM, smoking status, SNP rs1136410 (AA + AG/GG), history of hypertension, alcohol drinking status, BMI (≤ 25/> 25)	92/100	0.6380	0.0002
8	History of hyperlipidemia, history of T2DM, smoking status, SNP rs1136410 (AA + AG/GG), history of hypertension, alcohol drinking status, BMI (≤ 25/> 25), age (≤ 60/> 60)	91/100	0.6356	0.0003
9	Smoking status, alcohol drinking status, history of T2DM, history of hyperlipidemia, history of hypertension, SNP rs1136410 (AA + AG/GG), BMI (> 25/ <25), age (<60/> 60), sex	95/100	0.6384	0.0001

^aCVC means the number of times that a given combination of factors is identified in each testing set (a total of 100 times).

^bTesting accuracy (%) is the percentage of participants for whom a correct prediction is made.

^cThe permutation test was carried out to repeat the MDR analyses 1,000 times and to calculate the CVC and testing accuracy of each n-factor model.

Bold values indicate the models that have the maximal CVC and the optimal testing accuracy as well as the most significant P value for permutation test.

T2DM, type 2 diabetes mellitus; CVC, cross-validation consistency.