Table 2.
Predictive values of the prediction algorithms, including 95% CIs, for the prediction of good and poor outcome at 12 and 24 h after cardiac arrest for the internal and external validation tests
| Parameter | Internal validation | External validation | ||||
|---|---|---|---|---|---|---|
| Logistic regression | Random forest | Convolutional neural network | Logistic regression | Random forest | Convolutional neural network | |
| Prediction of good outcome | ||||||
| Predictive threshold for > 90% specificity at 12 h | 0.79 | 0.91 | 0.62 | 0.79 | 0.91 | 0.62 |
| Sensitivity at 12 h in % (CI) | 51 (32 to 70) | 51 (10 to 92) | 67 (34 to 100) | 83 (83 to 83) | 1 (0 to 4) | 78 (52 to 100) |
| FPR at 12 h in % (CI) | 9 (0 to 19) | 12 (0 to 29) | 13 (0 to 29) | 3 (3 to 3) | 0 (0 to 0) | 12 (0 to 24) |
| Predictive threshold for > 90% specificity at 24 h | 0.71 | 0.94 | 0.62 | 0.71 | 0.94 | 0.62 |
| Sensitivity at 24 h in % (CI) | 56 (39 to 73) | 48 (20 to 75) | 71 (59 to 83) | 66 (55 to 76) | 0 (0 to 0) | 81 (72 to 90) |
| FPR at 24 h in % (CI) | 10 (0 to 21) | 14 (8 to 20) | 14 (3 to 25) | 17 (12 to 22) | 0 (0 to 0) | 22 (14 to 30) |
| Prediction of poor outcome | ||||||
| Predictive threshold for > 99% specificity at 12 h | 0.02 | 0.06 | 0.16 | 0.02 | 0.06 | 0.16 |
| Sensitivity at 12 h in % (CI) | 51 (30 to 72) | 28 (0 to 63) | 49 (18 to 81) | 75 (75 to 75) | 56 (31 to 81) | 57 (43 to 71) |
| FPR at 12 h in % (CI) | 1 (0 to 5) | 4 (0 to 16) | 1 (0 to 4) | 3 (3 to 3) | 3 (3 to 3) | 3 (3 to 3) |
| Predictive threshold for > 99% specificity at 24 h | 0.10 | 0.03 | 0.17 | 0.10 | 0.03 | 0.17 |
| Sensitivity at 24 h in % (CI) | 40 (24 to 55) | 15 (5 to 24) | 55 (34 to 76) | 33 (33 to 33) | 13 (0 to 50) | 54 (44 to 64) |
| FPR at 24 h in % (CI) | 1 (0 to 4) | 3 (0 to 9) | 2 (0 to 9) | 0 (0 to 0) | 0 (0 to 2) | 0 (0 to 2) |
CI, Confidence interval, FPR, false positive rate