Development and validation of explainable machine-learning models for carotid atherosclerosis early screening

Table 3 Performance of GBDT model in five high-risk CAS subgroups

Disease subgroups	Datasets	auROC (95% CI)	auPR (95% CI)	Sensitivity	Specificity	PPV	NPV	PLR	NLR
Age ≥ 65	Training set (N = 239)	0.996(0.989–1)	NA	0.941	1.000	0.930	0.100	1.019	0.689
	Internal validation set (N = 64)	NA	NA	1.000	1.000	0.951	0.333	1.289	0.133
	External validation set (N = 252)	NA	NA	1.000	1.000	0.943	0.200	1.053	0.253
BMI ≥ 30	Training set (N = 341)	0.927(0.904–0.949)	0.939(0.917–0.96)	0.824	0.874	0.708	0.691	2.122	0.390
	Internal validation set (N = 68)	NA	NA	1.000	1.000	0.645	0.676	2.045	0.540
	External validation set (N = 213)	0.971(0.954–0.984)	0.972(0.954–0.985)	0.906	0.925	0.714	0.664	2.524	0.511
Dyslipidemia	Training set (N = 3027)	0.869(0.858–0.879)	0.866(0.852–0.878)	0.797	0.784	0.751	0.756	3.014	0.321
	Internal validation set (N = 754)	0.922(0.907–0.937)	0.928(0.911–0.943)	0.894	0.788	0.743	0.759	2.890	0.318
	External validation set (N = 2070)	0.877(0.864–0.888)	0.877(0.861–0.891)	0.822	0.775	0.761	0.773	3.184	0.293
Hypertension	Training set (N = 897)	0.87(0.85–0.89)	0.945(0.933–0.957)	0.710	0.855	0.793	0.611	1.474	0.244
	Internal validation set (N = 235)	0.978(0.962–0.99)	0.992(0.985–0.997)	0.948	0.903	0.802	0.491	1.453	0.372
	External validation set (N = 658)	0.895(0.875–0.916)	0.951(0.939–0.963)	0.804	0.829	0.789	0.667	1.621	0.217
Diabetes	Training set (N = 241)	0.973(0.954–0.989)	0.993(0.986–0.997)	0.911	0.939	0.824	0.361	1.198	0.452
	Internal validation set (N = 49)	NA	NA	1.000	1.000	0.870	0.000	0.930	NA
	External validation set (N = 128)	NA	NA	1.000	1.000	0.825	0.548	1.702	0.298

GBDT gradient boosting decision tree; auROC area under the receiver operating characteristic curve; auPR area under the Precision-Recall curve; PPV positive predictive value; NPV negative predictive value; PLR positive likelihood ratio; NLR negative likelihood ratio; NA not applicable

ISSN: 1479-5876