Table 1.

This table investigates the performance in terms of AUC of the different modeling approaches on the various subpopulations as defined by the chronic conditions CHF, COPD, and diabetes. Each row represents a separate subpopulation (with ‘N’ indicating the absence of the condition and ‘Y’ the presence). The best AUC for each subpopulation is printed in bold.

			AUC						Number of variables selected
(CHF, COPD, Diabetes)	Sample size train	Sample size validation	VennLasso	VennLasso adaptive	Interaction lasso	Interaction hierLasso	Separate lasso	Expanded lasso	VennLasso	vennLasso adaptive	Separate lasso	Expanded lasso
(N, N, N)	14, 939	14, 693	0.760	0.767	0.769	0.766	0.770	0.701	119	85	47	17
(Y, N, N)	1, 488	1, 543	0.692	0.692	0.687	0.690	0.683	0.665	35	34	15	17
(N, Y, N)	471	518	0.727	0.721	0.667	0.701	0.604	0.687	20	17	97	4
(N, N, Y)	2, 917	3, 022	0.699	0.693	0.690	0.695	0.679	0.649	36	35	15	19
(Y, Y, N)	196	189	0.587	0.593	0.609	0.519	0.583	0.512	33	31	2	12
(Y, N, Y)	720	784	0.752	0.750	0.760	0.740	0.706	0.722	54	54	34	18
(N, Y, Y)	138	131	0.727	0.726	0.688	0.725	0.569	0.510	43	42	2	2
(Y, Y, Y)	120	110	0.619	0.629	0.567	0.568	0.501	0.533	55	60	33	14