Table 3. Confusion Matrix for Classification of Hyperkalemia, Deep-Learning Model vs Serum Potassium.
| Validation Data Set | No. (%)a | ||||
|---|---|---|---|---|---|
| True Positiveb | False Positive | False Negative | True Negative | Accuracyc | |
| Minnesota (n = 50 099) | |||||
| SN = SP, 2 ECG leads | 1032 (2.1) | 9578 (19.1) | 250 (0.5) | 39 239 (78.3) | 40 271 (80.4) |
| SN90, 2 ECG leads | 1153 (2.3) | 17 809 (35.5) | 129 (0.3) | 31 008 (61.9) | 32 161 (64.2) |
| SN = SP, 4 ECG leads | 1058 (2.1) | 8495 (17.0) | 224 (0.4) | 40 322 (80.5) | 41 380 (82.6) |
| SN90, 4 ECG leads | 1153 (2.3) | 15 372 (30.7) | 129 (0.3) | 33 445 (66.7) | 34 598 (69.0) |
| Florida (n = 6011) | |||||
| SN = SP, 2 ECG leads | 225 (3.7) | 1374 (22.9) | 62 (1.0) | 4350 (72.4) | 4575 (76.1) |
| SN90, 2 ECG leads | 262 (4.4) | 2499 (41.6) | 25 (0.4) | 3225 (53.6) | 3487 (58.0) |
| SN = SP, 4 ECG leads | 239 (4.0) | 1308 (21.8) | 48 (0.8) | 4416 (73.4) | 4655 (77.4) |
| SN90, 4 ECG leads | 263 (4.4) | 2149 (35.7) | 24 (0.4) | 3575 (59.5) | 3838 (63.9) |
| Arizona (n = 5855) | |||||
| SN = SP, 2 ECG leads | 209 (3.6) | 1322 (22.6) | 61 (1.0) | 4263 (72.8) | 4472 (76.4) |
| SN90, 2 ECG leads | 237 (4.0) | 2434 (41.6) | 33 (0.6) | 3151 (53.8) | 3388 (57.8) |
| SN = SP, 4 ECG leads | 223 (3.8) | 1275 (21.8) | 47 (0.8) | 4310 (73.6) | 4533 (77.4) |
| SN90, 4 ECG leads | 249 (4.3) | 2095 (35.8) | 21 (0.3) | 3490 (59.6) | 3739 (63.9) |
Abbreviations: SN, sensitivity, SP, specificity; SN90, sensitivity at 90%; ECG, electrocardiogram.
SI conversion factor: To convert potassium to millimoles per liter, multiply by 1.
a
Percentage value is value divided by the total sample size.
b
Hyperkalemia defined as serum potassium ≥5.5 mEq/L.
c
Accuracy calculated as number of true positives and true negative divided by total sample size.