Table 4.

Interactive effects between POSTN and SOST genes on BMD variation by MDR and conditional logistic regression analyses

	Either LS or FN	LS	FN
SNP of POSTN	rs9547970	rs9547970	rs9547970
SNP of SOST	rs2301682	rs9899889	rs9899889
		rs865429	rs865429
		rs2301682
MDR
Cross validation consistency	20/20	19/20	20/20
Prediction accuracy	0.57	0.57	0.56
Sign test P-value	<0.0001	0.001	0.0087
Conditional logistic regression analysis
P value	0.001	0.002	0.002

Several output parameters are used to select the best interaction model in MDR. The cross-validation consistency score measures the degree of consistency with which the reported interaction is identified as the most evident model. The testing accuracy score measures the degree to which the interaction accurately predicts case–control status (accuracy score ≥0.55 is suggested as “interesting”). The best model is the one with the maximal cross-validation consistency and minimal prediction error. When cross-validation consistency is higher for one model and prediction error is lower for another model, the model involving the fewest loci/factors is taken as the best. The statistical significance (sign test P value) derived empirically from 1,000 permutations was adjusted for multiple comparisons