Combining multiple ECG features does not improve prediction of defibrillation outcome compared to single features in a large population of out-of-hospital cardiac arrests

Table 4 Prediction power of combination methods and single features for the first defibrillations in the validation data (175 successful shocks/567 shocks)

Methods	AUC	Sensitivity (%)	Specificity (%)	NPV (%)	PPV (%)	PA (%)
LR	0.870	79.6	79.6	89.1	65.0	79.6
BP-C1	0.864	78.9	78.9	89.2	62.4	78.7
BP-C2	0.868	80.0	79.9	89.9	64.2	80.0
BP-C3	0.873	80.0	80.0	89.9	64.2	80.0
SVM-C1	N/A	N/A	69.0	100.0	0.0	69.0
SVM-C2	N/A	N/A	69.0	100.0	0.0	69.0
SVM-C3	N/A	67.0	76.2	92.0	36.0	74.6
MS	0.873	84.0	79.2	91.7	64.5	80.7
AMSA	0.870	73.1	82.8	87.3	65.6	79.8
MdS	0.872	76.0	82.0	88.4	65.5	80.1

C1, C2 and C3 represented combination of all features, combination of features with a high predictive power (AUC > 0.8) and combination of complementary features (MS and SFM) using BP neural network, respectively
AUC area under receiver operating characteristic curve, NPV negative predictive value, PPV positive predictive value, PA prediction accuracy, LR logistic regression method, BP back propagation neural network method, SVM support vector machine method, MS mean slope, AMSA amplitude spectral area, Mds median slope, N/A not existing

ISSN: 1364-8535