Table 5.
Sensitivity Analyses for external validation eICU dataset. Systolic Blood Pressure, BUN, and hemoglobin cutoffs were derived from the Glasgow Blatchford Score.
| Total encounters N = 1526 | LSTM neural network | Logistic regression | p-value | Regularized logistic regression with elastic net | p-value |
|---|---|---|---|---|---|
|
Female N = 607 |
0.64 (0.57–0.71) | 0.54 (0.46–0.62) | 0.002 | 0.49 (0.42–0.56) | 0.02 |
|
Age > 65 (mean) N = 820 |
0.61 (0.55–0.67) | 0.54 (0.47–0.60) | 0.008 | 0.56 (0.50–0.61) | 0.01 |
|
Systolic Blood Pressure < 110 N = 849 |
0.64 (0.58–0.69) | 0.57 (0.50–0.63) | 0.03 | 0.55 (0.49–0.61) | 0.002 |
|
BUN > 18.2 N = 1110 |
0.64 (0.58–0.69) | 0.56 (0.50–0.62) | 0.003 | 0.56 (0.50–0.61) | < 0.001 |
|
Hemoglobin < 10 N = 1303 |
0.64 (0.60–0.69) | 0.56 (0.51–0.62) | 0.001 | 0.56 (0.51–0.61) | < 0.001 |
Hemoglobin cutoff was chosen due to the matched risk for both men and women.