Mathematical models of amino acid panel for assisting diagnosis of children acute leukemia

Table 4 The validation of models on AL diagnosis for Out-Sample Test

	Diagnosis (model/clinical diagnosis)				χ²	Kappa value	p value	AUCⁱ
	+/+ ^a	∓^b	±^c	−/−^d	χ²	Kappa value	p value	AUCⁱ
Result-SVM^h	237	43	46	262	0.1011	0.697	0.751	0.788
Sensitivity^e (%)	84.64
Specificity^f (%)	85.06
Accuracy^g (%)	84.86
Result-RF^h	231	49	38	270	1.3908	0.703	0.238	0.803
Sensitivity^e (%)	82.50
Specificity^f (%)	87.66
Accuracy^g (%)	85.20
Result-XGB^h	252	28	34	274	0.2903	0.789	0.446	0.830
Sensitivity^e (%)	90.00
Specificity^f (%)	88.96
Accuracy^g (%)	89.46

SVM: support vector machine; RF: random forest; XGB: XGBoot; FN: false negative; FP: false positive; AUC: area under curve
^aOur model or clinical diagnosis were both positive-children were with leukemia
^bOur model diagnosed children as normal, but the clinical diagnosis of them was leukemia
^cOur model diagnosed children as leukemia, but the clinical diagnosis of them was normal
^dOur model or clinical diagnosis were both negative, and children were normal
^eNumber of +/+ for each model/(number of +/+ for each model plus number of ∓ for each model) × 100%
^fNumber of −/− for each model/(number of −/− for each model plus number of ± for each model) × 100%
^g(Number of −/− for each model plus number of +/+ for each model)/588 × 100%
^hMcNemar’s test
ⁱROC analysis

ISSN: 1479-5876