Predicting the serum digoxin concentrations of infants in the neonatal intensive care unit through an artificial neural network

Table 4 Classification performance of prediction to differentiate toxicity concentrations (i.e., equal and above 1.5 ng/ml) or not, as compared to the observed serum digoxin concentrations, for each ANN model on validation dataset

Model	No. of parameters	Parameters	TP	TN	FP	FN	RCP(%)	SE(%)	SP(%)
1	11	All Variables	1	22	2	4	79.3	20	91.7
2	10	-Sex	2	22	2	3	82.8	40	91.7
3	9	-Sex-DCM	3	21	3	2	82.8	60	87.5
4	8	-Sex-DCM -PH	2	22	2	3	82.8	40	91.7
5	7	-Sex-DCM -PH -Captopril	2	18	6	3	69.0	40	75.0
6	6	-Sex-DCM -PH -Captopril -Furosemide	2	20	4	3	75.9	40	83.3
7	5	-Sex-DCM -PH -Captopril -Furosemide -VSD	3	16	8	2	65.5	60	66.7
8	4*	-Sex-DCM -PH -Captopril -Furosemide -VSD -ibuprofen	2	21	3	3	79.3	40	87.5

“- “in the column of parameters refers to “exclude” that specific variable from the model 1, which contain all variables. TP true positive (correctly classified to be ‘positive’); TN true negative (correctly classified to be ‘negative’); FP false positive (incorrectly classified to be ‘positive’); FN false negative (incorrectly classified to be ‘negative’), respectively; RCP rate of correct prediction; SE sensitivity; SP specificity
All variables include: dose per total body weight, weight, gender, postmenstrual age (PMA), Congestive heart failure (CHF), dilated cardiomyopathy (DCM), pulmonary hypertension (PH), Ventricular septal defect (VSD), with captopril, with furosemide, with ibuprofen
*the common variables used in population pharmacokinetics were dose per total body weight, weight, PNA, CHF

ISSN: 1471-2431