Genetic factors increase the identification efficiency of predictive models for dyslipidaemia: a prospective cohort study

Table 4 Performance of the conventional and conventional+GRS models in predicting dyslipidaemia

	AUC	△AUC	Continuous NRI, %	IDI, %
Cox
Conventional model	0.702(0.673, 0.729)
Conventional+GRS model	0.707(0.679, 0.734)	0.0049(P = 0.0549)	25.6 (13.8, 35.8)^*	2.3 (1.1, 3.7)^*
ANN
Conventional model	0.736(0.708, 0.762)
Conventional+GRS model	0.754(0.727, 0.779)	0.0183(P = 0.0031)^*	7.8 (−2.7, 18.5)	1.0 (−0.3, 2.4)
RF
Conventional model	0.787 (0.762, 0.811)
Conventional+GRS model	0.810 (0.762, 0.811)	0.0230(P = 0.023)^*	14.1 (1.1, 26.1)^*	2.5 (0.5, 4.2)^*
GBM
Conventional model	0.816(0.792, 0.839)
Conventional+GRS model	0.831(0.808, 0.853)	0.0151(P = 0.0135)^*	18.1 (4.4, 27.2)^*	1.8 (0.1, 3.5)^*

Abbreviations: AUC area under receiver operating characteristic curve, △AUC difference between AUC of conventional model and conventional+GRS model, NRI net reclassification improvement, IDI integrated discrimination improvement, ANN artificial neural network, RF random forest, GBM gradient boosting machine
^*Statistically significant values, P < 0.05

ISSN: 1476-511X