Table 3. Internal validation performance for the two models: the four clinical traits alone or in combination with the eight SNPs.

Model	Included individuals					AUC mean (s.d.) for 1,000 splits on the individuals held out because of their redundant properties
	Same splits as used for feature selection				AUC mean (s.d.) for 1,000 splits
	AUC	Accuracy	Specificity	Sensitivity
Clinical traits alone	0.810	0.735	0.629	0.811	0.807 (0.0054)	0.992 (0.00008927)
Clinical traits+eight SNPs	0.921	0.838	0.790	0.872	0.919 (0.0046)	0.917 (0.001072)

Abbreviations: AUC, area under the curve; SNP, single-nucleotide polymorphism.

The first four columns show internal validation and performance measures for the cross-validation splits used for feature selection and the 268 individuals remaining after excluding similar patients (as reported throughout the paper). The next column shows internal validation again based on the 268 individuals, but for 1,000 different cross-validation splits. The last column shows the AUC for the 189 individuals initially held out because of their redundant properties in terms of clinical traits. In this last column, the models are trained on the 268 included individuals but evaluated on the 189 held out individuals.