Optimal Vasopressin Initiation in Septic Shock: The OVISS Reinforcement Learning Study
An erratum to this article was published (JAMA, 6 May 2025 ↗)
Main finding · Hospital mortalityResult — go to the resultAdjusted odds ratio 0.81 95% CI 0.73–0.91Authors' conclusion
In adult patients with septic shock receiving norepinephrine, the use of vasopressin was variable. A reinforcement learning model developed and validated in several observational datasets recommended more frequent and earlier use of vasopressin than average care patterns and was associated with reduced mortality.
Study context
- Population
- Adult critically ill patients receiving norepinephrine for septic shock (Sepsis-3 shock criteria); derivation at 5 California hospitals 2012–2023, validation in 3 external datasets from 227 US hospitalsderivation
- Participants
- 3,608 patients
- Exposure
- Rule vs Clinicians' actions
- Outcome
- In-hospital mortality
Methodological detailsreinforcement learning model derivation and validation, multicentre, external validation, weighted importance sampling, inverse probability weighting
- Reinforcement learning model derivation and validation
- Multicentre
- External validation
- Weighted importance sampling
- Inverse probability weighting
Main finding
Hospital mortality
Adjusted odds ratio
0.81
95% CI 0.73–0.91
As reported in the abstract
Observational data: associations, not effects.
Further findings
Vasopressin initiation (validation data)
Time from shock onset to vasopressin initiation (validation data)
Norepinephrine dose at vasopressin initiation (validation data)
| Outcome | Weighted importance sampling difference (95% CI) | |
|---|---|---|
| Expected reward (validation data) | 31 (15–52) | |
| −50050100Difference (linear scale) |
Original publication
Composed by AI from the published abstract and checked against it. How Briefs are made →