Table 4. Examples of Influential Predictive Terms Derived From Clinical Text.
| Clinical Term | Weighta |
|---|---|
| Pupils (fixed) | 7.78 |
| Gag | 6.74 |
| ECMO | 6.18 |
| Coagulopathy | 4.67 |
| Shock | 4.41 |
| Intubated | 4.28 |
| PEA | 3.68 |
| Chemotherapy | 3.49 |
| Ascites | 3.27 |
| CVVH | 2.78 |
| Sepsis | 2.27 |
| Meropenem | 2.09 |
| EtOH | −1.14 |
| OHNS | −1.15 |
| Alert | −1.51 |
| EBL | −2.10 |
| Diet | −2.68 |
| Awake | −3.11 |
| PERRL | −4.28 |
| Denies (pain) | −4.56 |
| POD | −4.70 |
| Extubated | −7.64 |
Abbreviations: CVVH, continuous venovenous hemofiltration; EBL, expected blood loss; ECMO, extracorporeal membrane oxygenation; EtOH, ethanol (alcohol); OHNS, otolaryngology–head and neck surgery; PEA, pulseless electrical activity; PERRL, pupils equal, round, and reactive to light; POD, postoperative day.
a
Each term is associated with a β coefficient or weight in the logistic regression model, which represents its relative association with mortality. Positive weights indicate increased odds of mortality when the term is included in a clinical note. Negative weights indicate decreased odds of mortality.