Table 3.

Performances of binary classifiers of smoking status

Data	Accuracy statistics	AHRR model (reference)	Candidate CpG LASSO model	Maas model	Agnostic LASSO model
Ever/never smokers
Training data	Accuracy (95% CI)	0.721 (0.693–0.747)	0.771 (0.744–0.795)	0.752 (0.725–0.777)	0.792 (0.766–0.816)
	NIR (P: Acc > NIR)	0.658 (6.3 × 10−6)	0.658 (7.3 × 10−16)	0.658 (2.1 × 10−11)	0.658 (< 2.2 × 10−16)
	Kappa	0.444	0.527	0.480	0.577
	Sensitivity	0.661	0.743	0.745	0.744
	Specificity	0.835	0.824	0.764	0.885
	PPV	0.885	0.890	0.858	0.925
	NPV	0.562	0.625	0.610	0.658
External validation data	Accuracy (95% CI)	0.815 (0.784–0.842)	0.837 (0.808–0.863)	0.822 (0.791–0.849)	0.822 (0.791–0.849)
	NIR (P: Acc > NIR)	0.637 (< 2.2 × 10−16)	0.637 (< 2.2 × 10−16)	0.637 (< 2.2 × 10−16)	0.637 (< 2.2 × 10−16)
	Kappa	0.624	0.661	0.627	0.633
	Sensitivity	0.766	0.818	0.814	0.792
	Specificity	0.900	0.869	0.835	0.873
	PPV	0.931	0.917	0.896	0.917
	NPV	0.686	0.731	0.719	0.705
Current/former smokers
Training data	Accuracy (95% CI)	0.707 (0.671–0.740)	0.512 (0.474–0.550)	0.700 (0.664–0.733)	0.757 (0.723–0.788)
	NIR (P: Acc > NIR)	0.522 (< 2.2 × 10−16)	0.522 (0.715)	0.522 (< 2.2 × 10−16)	0.522 (< 2.2 × 10−16)
	Kappa	0.416	0.025	0.403	0.516
	Sensitivity	0.658	0.504	0.625	0.701
	Specificity	0.761	0.521	0.781	0.817
	PPV	0.750	0.535	0.758	0.808
	NPV	0.670	0.490	0.656	0.715
External validation data	Accuracy (95% CI)	0.646 (0.600–0.689)	0.541 (0.494–0.587)	0.619 (0.573–0.664)	0.674 (0.629–0.717)
	NIR (P: Acc > NIR)	0.576 (1.3 × 10−3)	0.576 (0.940)	0.576 (0.03)	0.576 (9.9 × 10−6)
	Kappa	0.318	0.093	0.251	0.373
	Sensitivity	0.825	0.603	0.706	0.861
	Specificity	0.513	0.494	0.555	0.536
	PPV	0.556	0.468	0.539	0.578
	NPV	0.799	0.628	0.719	0.839