Table 4. Multi-locus interactions by the multifactor dimensionality reduction approach.
| Best candidate model* | Testing accuracy (%) | p value (sign test) | CVC |
|---|---|---|---|
| −219T>G |
43.61 |
0.999 |
5/10 |
| IVS8–53T>C and IVS5+30C>T |
52.5 |
0.377 |
7/10 |
| IVS2+35A>G, IVS5+30C>T, and −219T>G |
43.59 |
1.0 |
4/10 |
| IVS2+35A>G, IVS5+30C>T, −219T>G, and 290C>T |
47.53 |
0.945 |
9/10 |
| A488A, IVS5+38, IVS5+30C>T, −427T>C, and −219T>G |
49.16 |
0.623 |
10/10 |
| IVS2+35A>G, M98K, IVS5+38T>G, IVS8–53T>C, IVS5+30C>T, and −491A>T | 55.14 | 0.0107 | 10/10 |
Interaction models with one to six factors generated by the MDR program were shown in the table. The six-factor model with highest CVC (10/10) and maximum testing accuracy (55.14%) was selected as the best model in this study. The asterisk indicates that as the sample numbers in cases and controls are dissimilar in this study, the T value, which is the threshold ratio used to distinguish high-risk genotype combinations from low risk genotype combinations, was set to 0.88 (number of cases/number of controls, i.e., 176/200) while running the MDR program. CVC: cross-validation consistency.