Table 5.
Performance of the logistic regression model versus two simple threshold alerts.
| Hospital 1 | Hospital 2 | |||||
|---|---|---|---|---|---|---|
| Model | Threshold Alert | Model | Threshold Alert | |||
| 60 mmHg | 65 mmHg | 60 mmHg | 65 mmHg | |||
| Statistics: | ||||||
| Number of sustained episodes of hypotension, n | 259 | 259 | 259 | 26 | 26 | 26 |
| Proportion of episodes that were undetected, % | 0.4% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| Advance forecast time, median (IQR), min | 12.0 (0.0–30.0) | 0.0 † (0.0–0.0) | 5.0 (2.0–18.8) | 22.0 (2.0–30.0) | 0.0 † (0.0–0.0) | 8.0 (4.0–26.0) |
| Number of false alarms per 24 hours, median per stay (IQR) | 6.9 (2.0–12.4) | 11.9 * (4.2–20.9) | 25.9 † (12.6–42.6) | 2.1 (0.4–7.5) | 3.8 * (1.9–10.2) | 20.1 † (9.0–44.0) |
| Duration of false alarms, median per stay (IQR), min | 8. 0 (3.0–16.0) | 2.0 † (1.0–3.0) | 3.0 † (1.0–7.0) | 5. 0 (2.3–12.0) | 2.0 † (1.0–3.0) | 2.0 † (1.0–4.0) |
Hypotension is defined as MAP < 60 mmHg and sustained episode of hypotension is defined as ≥15 min. An episode of hypotension was “detected” if, directly upon its onset, there had been an alarm episode in the preceding 30 min. An alarm episode (defined as the continuous time interval when the alarm condition was true) was a “false alarm” if, directly upon its onset, there was no sustained hypotension commencing within 30 min. *Significant difference between model and threshold alert, P < 0.05. †Significant difference between model and threshold alert, P < 0.001.