Table 5.

Comparison of positive and negative predictive value of individual models and ensembles on holdout data (ADNI3).

	Negative Predictive Value		Positive Predictive Value
Model	17% prevalence	32% prevalence	17% prevalence	32% prevalence
Baseline (age + sex + APOE)	0.913 (0.883, 0.944)	0.820 (0.764, 0.878)	0.274 (0.232, 0.324)	0.455 (0.391, 0.535)
PRS	0.919 (0.889, 0.949)	0.831 (0.773, 0.894)	0.286 (0.241, 0.337)	0.471 (0.402, 0.550)
XGBoost	0.950 (0.925, 0.975)	0.893 (0.840, 0.946)	0.341 (0.290, 0.402)	0.524 (0.456, 0.608)
NODEnn	0.937 (0.902, 0.969)	0.866 (0.802, 0.929)	0.188 (0.159, 0.222)	0.319 (0.278, 0.370)
PRS + XBGoost	0.933 (0.905, 0.960)	0.861 (0.808, 0.915)	0.354 (0.301, 0.415)	0.550 (0.479, 0.641)
XGBoost + NODEnn	0.951 (0.924, 0.975)	0.892 (0.840, 0.945)	0.347 (0.296, 0.405)	0.532 (0.462, 0.616)
PRS + NODEnn	0.935 (0.905, 0.964)	0.863 (0.803, 0.920)	0.263 (0.225, 0.312)	0.427 (0.370, 0.493)
PRS + XGBoost + NODEnn	0.935 (0.907, 0.961)	0.864 (0.810, 0.917)	0.383 (0.325, 0.452)	0.593 (0.511, 0.690)