Clinical performance of a machine-learning algorithm to predict intra-operative hypotension with noninvasive arterial pressure waveforms: A cohort study
Intra-operative hypotension is associated with adverse postoperative outcomes. A machine-learning-derived algorithm developed to predict hypotension based on arterial blood pressure (ABP) waveforms significantly reduced intra-operative hypotension. The algorithm calculates the likelihood of hypotens...
Gespeichert in:
| Veröffentlicht in: | European journal of anaesthesiology Jg. 38; H. 6; S. 609 |
|---|---|
| Hauptverfasser: | , , , , , , , , |
| Format: | Journal Article |
| Sprache: | Englisch |
| Veröffentlicht: |
England
01.06.2021
|
| ISSN: | 1365-2346, 1365-2346 |
| Online-Zugang: | Weitere Angaben |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Zusammenfassung: | Intra-operative hypotension is associated with adverse postoperative outcomes. A machine-learning-derived algorithm developed to predict hypotension based on arterial blood pressure (ABP) waveforms significantly reduced intra-operative hypotension. The algorithm calculates the likelihood of hypotension occurring within minutes, expressed as the Hypotension Prediction Index (HPI) which ranges from 0 to 100. Currently, HPI is only available for patients monitored with invasive ABP, which is restricted to high-risk procedures and patients. In this study, the performance of HPI, employing noninvasive continuous ABP measurements, is assessed.
The first aim was to compare the performance of the HPI algorithm, using noninvasive versus invasive ABP measurements, at a mathematically optimal HPI alarm threshold (Youden index). The second aim was to assess the performance of the algorithm using a HPI alarm threshold of 85 that is currently used in clinical trials. Hypotension was defined as a mean arterial pressure (MAP) below 65 mmHg for at least 1 min. The predictive performance of the algorithm at different HPI alarm thresholds (75 and 95) was studied.
Observational cohort study.
Tertiary academic medical centre.
Five hundred and seven adult patients undergoing general surgery.
The performance of the algorithm with invasive and noninvasive ABP input was similar. A HPI alarm threshold of 85 showed a median [IQR] time from alarm to hypotension of 2.7 [1.0 to 7.0] min with a sensitivity of 92.7 (95% confidence interval [CI], 91.2 to 94.3), specificity of 87.6 (95% CI, 86.2 to 89.0), positive predictive value of 79.9 (95% CI, 77.7 to 82.1) and negative predictive value of 95.8 (95% CI, 94.9 to 96.7). A HPI alarm threshold of 75 provided a lower positive predictive value but a prolonged time from prediction to actual hypotension.
This study demonstrated that the algorithm can be employed using continuous noninvasive ABP waveforms. This opens up the potential to predict and prevent hypotension in a larger patient population.
Clinical trials registration number NCT03533205. |
|---|---|
| Bibliographie: | ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 23 ObjectType-Undefined-3 |
| ISSN: | 1365-2346 1365-2346 |
| DOI: | 10.1097/EJA.0000000000001521 |