Average AUC, sensitivity and specificity with 95% CI of delayed c erebral ischemia (DCI) prediction models for all approaches
| Data | Model | AUC (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) |
| Prior knowledge variables* | LR (Model 1) | 0.63 (0.62 to 0.63) | 0.67 (0.64 to 0.70) | 0.62 (0.59 to 0.65) |
| Prior knowledge variables† | LR (Model 2) | 0.59 (0.57 to 0.60) | 0.61 (0.57 to 0.65) | 0.64 (0.60 to 0.68) |
| All clinical variables | SVM | 0.64 (0.63 to 0.65) | 0.67 (0.63 to 0.70) | 0.64 (0.61 to 0.67) |
| RFC | 0.68 (0.65 to 0.69) | 0.78 (0.75 to 0.81) | 0.57 (0.54 to 0.61) | |
| LR | 0.61 (0.60 to 0.63) | 0.65 (0.61 to 0.68) | 0.62 (0.59 to 0.67) | |
| MLP | 0.63 (0.62 to 0.64) | 0.59 (0.56 to 0.62) | 0.79 (0.76 to 0.81) | |
| All clinical variables see combined with extracted image features | SVM | 0.68 (0.65 to 0.68) | 0.63 (0.59 to 0.66) | 0.73 (0.70 to 0.76) |
| RFC | 0.74 (0.72 to 0.75) | 0.67 (0.65 to 0.70) | 0.75 (0.72 to 0.78) | |
| LR | 0.65 (0.64 to 0.67) | 0.65 (0.62 to 0.67) | 0.69 (0.66 to 0.71) | |
| MLP | 0.67 (0.66 to 0.68) | 0.64 (0.60 to 0.67) | 0.72 (0.69 to 0.75) |
The first two columns specify which data (variables) were used to build each model.
*WFNS, age, treatment (clipping or coiling), intraparenchymal and intraventricular hemorrhage, (TBV).3
†Hypertension, diabetes mellitus, history of smoking, alcohol use, hyperglycemia and Hunt and Hess grade on admission.4
AUC, area under the curve; LR, Logistic Regression; SVM, Ssupport Vector Machine; RFC, Random Forest; MLP, Multilayer Perceptron; All Variables, see online supplement table IV.