Table 2.
Performance characteristics of AI-ECG, by position
| Adequate ECG, n/N (%) | AUC |
Maximising Se and Sp equally (Youden index) |
Maximising Se and Sp with rule Se >81, Sp >67, Se >81, or maximising Sp |
|||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Threshold | Se | Sp | PPV | NPV | F1 score | Threshold | Se | Sp | PPV | NPV | F1 score | |||
| 1 | 846/1050 (80·6%) | 0·75 | 0·370 | 77·1 | 60·7 | 17·3 | 95·9 | 0·282 | 0·345 | 81·9 | 53·3 | 15·8 | 96·2 | 0·264 |
| 2 | 979/1050 (93·2%) | 0·85 | 0 ·443 | 71·7 | 86·5 | 37·0 | 96·3 | 0·486 | 0·341 | 84·8 | 69·5 | 23·6 | 97·4 | 0·369 |
| 3 | 946/1050 (90·1%) | 0·78 | 0·489 | 68·1 | 77·4 | 24·7 | 95·5 | 0·361 | 0·280 | 81·9 | 55·2 | 16·6 | 96·3 | 0·275 |
| 4 | 968/1050 (92·2%) | 0·78 | 0·420 | 62·9 | 80·6 | 26·2 | 95·0 | 0·368 | 0·312 | 81·4 | 58·4 | 17·7 | 96·4 | 0·290 |
| 5 | 916/1050 (87·2%) | 0·79 | 0·427 | 62·8 | 83·4 | 27·7 | 95·5 | 0·383 | 0·304 | 81·4 | 60·1 | 17·5 | 96·8 | 0·287 |
| 2 and 5 | 864/1050 (82·3%)* | 0·85 | 0·450 | 82·7 | 79·9 | 29·9 | 87·8 | 0·439 | 0·450 | 82·7 | 79·9 | 29·9 | 87·8 | 0·439 |
| 2 and 5, LR | 346/864 (40%)† | 0·91 | 0·497 | 91·9 | 80·2 | 35·1 | 98·4 | 0·503 | 0·497 | 91·9 | 80·2 | 35·1 | 98·4 | 0·503 |
AUC=area under the curve. 1=aortic. 2=pulmonary. 3=tricuspid. 4=mitral. 5=handheld. AI=artificial intelligence. LR=logistic regression. Se=sensitivity. Sp=specificity. PPV=positive predictive value. NPV=negative predictive value.
*
Number of patients who had adequate recordings at both position 2 and 5, where a positive AI-ECG result as per threshold was considered a positive test.
†
Representing 40% testing dataset from the original 864 participants with both position 2 and 5 recordings.