Table 5 Features of clinical, radiological and combined models in detecting ICU candidate COVID-19 patients
Model features | Training and validation set** | Test set*** | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
Type | Parameters | p | Odds Ratio* (95% CI) | C-index 95% CI§ | AUC | Spec% | Sens% | Acc% | Precision | F1 |
Clinical model | Age | 0.002 | 1.057 (1.023–1.092) | Training 0.846 0.824–0.866 Validation 0.805 0.724–0.869 | 0.736 | 79.17 | 66.67 | 74.36 | 66.67 | 0.67 |
Platelet count | 0.117 | 0.995 (0.990–0.999) | ||||||||
eGFR | 0.385 | 0.989 (0.966–1.014) | ||||||||
AST | 0.322 | 1.016 (0.998–1.034) | ||||||||
LDH | < 0.001 | 1.007 (1.003–1.011) | ||||||||
Procalcitonin | < 0.001 | 8.542 (1.892–38.559) | ||||||||
Fever > 38 °C | 0.092 | 3.418 (1.439–8.121) | ||||||||
Radiological model | Percent infected lung | 0.010 | 1.043 (1.009–1.078) | Training 0.841 0.819–0.860 Validation 0.815 0.746–0.877 | 0.708 | 79.17 | 60.00 | 71.79 | 64.28 | 0.62 |
First order skewness | < 0.001 | 0.029 (0.006–0.138) | ||||||||
First order RMAD | 0.030 | 0.969 (0.942–0.997) | ||||||||
GLCM sumentropy | 0.002 | 7.181 (1.974–26.128) | ||||||||
GLDM LDLGLE | 0.005 | 5.486 (1.650–18.245) | ||||||||
Combined model | Percent infected lung | 0.015 | 1.048 (1009–1088) | Training 0.897 0.882–0.914 Validation 0.853 0.787–0.905 | 0.794 | 87.50 | 73.33 | 82.05 | 78.57 | 0.76 |
First order skewness | 0.001 | 0.012 (0.001–0.166) | ||||||||
GLCM clustershade | 0.058 | 1.000 (1.000–1.001) | ||||||||
GLSZM LALGLE | 0.002 | 1.525 (1.174–1.982) | ||||||||
N/L ratio | 0.051 | 1.088 (1.000–1.184) | ||||||||
Platelet count | 0.009 | 0.993 (0.987–0.998) | ||||||||
eGFR | < 0.001 | 0.962 (0.942–0.983) | ||||||||
AST | 0.024 | 1.021 (1.003–1.040) | ||||||||