Table 4 Ablation study for CRISP on MIMIC III dataset. AUROC performance comparison over different feature sets. ‘Tabular’ denotes categorical variables and numerical variables. ‘Text’ denotes diagnoses, procedures, and medications texts. ‘CasualI’ indicates results with causal inference
From: CRISP: A causal relationships-guided deep learning framework for advanced ICU mortality prediction
| Â | AUROC | AUPRC | Recall | F1 |
|---|---|---|---|---|
Tabular | 0.9381 (0.9377–0.9386) | 0.7523 (0.7516–0.7531) | 0.7185 (0.7073–0.7297) | 0.6832 (0.6821–0.6844) |
Tabula|CasualI | 0.9390 (0.9387–0.9393) | 0.7570 (0.7544–0.7596) | 0.7388 (0.7124–0.7652) | 0.6911 (0.6886–0.6936) |
Text | 0.8963 (0.8958–0.8968) | 0.5783 (0.5762–0.5803) | 0.6677 (0.6514–0.6839) | 0.5594 (0.5578–0.5609) |
Text|CasualI | 0.8973 (0.8972–0.8974) | 0.5800 (0.5772–0.5828) | 0.7033 (0.6738–0.7327) | 0.5684 (0.5619–0.5748) |
Tabular + Text | 0.9484 (0.9478–0.9490) | 0.7597 (0.7578–0.7616) | 0.7175 (0.7093–0.7256) | 0.6928 (0.6903–0.6953) |
Tabular + Text|CasualI | 0.9480 (0.9477–0.9483) | 0.7611 (0.7594–0.7628) | 0.7612 (0.7154–0.8069) | 0.6939 (0.6909–0.6968) |