TREWS
2022
Prospective, multi-site study of patient outcomes after implementation of the TREWS machine learning-based early warning system for sepsis
In 6,877 patients with sepsis flagged before antibiotics, provider confirmation of the machine-learning sepsis alert within 3 h reduced in-hospital mortality compared with alerts not confirmed by a provider within 3 h.
Result
Primary endpoint: In-hospital mortality
In-hospital mortality adjusted absolute reduction 3.3% (CI 1.7–5.1%), adjusted relative reduction 18.7% (CI 9.4–27.0%), with less organ failure and shorter stay; among patients additionally flagged as high risk, absolute mortality reduction 4.5% (CI 0.8–8.3%). 590,736 patients monitored.
- Subtopic
- Sepsis alerts
- Design
- Prospective multi-site cohort study, non-randomised (5 hospitals)
- Relevance
- Supporting: context rather than must-know.
- Source
- Adams R, et al. Nat Med. 2022;28(7):1455-1460.doi 10.1038/s41591-022-01894-0
Discussion & critique
One of the first machine-learning sepsis systems evaluated at scale in live clinical use, showing that sepsis can be flagged before antibiotics and that acting on the alert tracks with better outcomes. The comparison is not randomised: it contrasts patients whose alert a clinician confirmed within 3 h with those whose alert was not, so confounding by clinician attention and patient presentation, only partly removed by adjustment, can explain part of the effect.