Sepsis is a common and potentially fatal condition, causing over 270,000 deaths annually in the U.S. Emergency treatment involves fluids and vasopressors, with norepinephrine as the first-line choice and vasopressin as a second-line option when blood pressure remains unresponsive. However, the timing of vasopressin use is unclear due to the dynamic nature of septic shock and limited clinical trial guidance.
The Optimal Vasopressin Initiation in Septic Shock (OVISS) study, presented at ISICEM 2025, applied reinforcement learning, a machine learning technique to develop and validate an optimised vasopressin initiation strategy aimed at improving outcomes in critically ill patients with septic shock.
Reinforcement learning was used to develop an optimal vasopressin initiation rule to improve patient outcomes, based on electronic health records from 3,608 septic shock patients across five California hospitals (2012–2023). The rule was evaluated in 628 California patients and 10,217 patients from 227 U.S. hospitals using weighted importance sampling and pooled logistic regression with inverse probability weighting.
The primary outcome was in-hospital mortality. The derivation cohort (n=3,608) had a median age of 63 years and a SOFA score of 5, while the validation cohorts (n=10,217) had a median age of 67 years and a SOFA score of 6. The model recommended vasopressin initiation in more patients (87% vs. 31%), earlier (median 4 vs. 5 hours after shock onset), and at lower norepinephrine doses (0.20 vs. 0.37 µg/kg/min) than clinicians. The rule was associated with improved outcomes, showing lower hospital mortality when followed, a result consistent across validation sets.
The study analysed data from 232 hospitals and 14,453 patients to develop and validate a reinforcement learning model for optimising vasopressin initiation in septic shock patients receiving norepinephrine. The model was associated with improved outcomes, including reduced in-hospital mortality, compared to standard clinician decisions.
Current guidelines recommend vasopressin as a second-line vasopressor, but optimal timing and dosing remain unclear due to a lack of randomised trials. The OVISS model suggested earlier vasopressin initiation at lower norepinephrine doses, aligning with prior subgroup analyses. Potential benefits include improved kidney function, reduced catecholamine exposure, and enhanced clinical vigilance. While findings suggest improved outcomes, possible confounding factors remain.
In adult septic shock patients receiving norepinephrine, vasopressin use varied. A reinforcement learning model, developed and validated across multiple datasets, recommended earlier and more frequent vasopressin initiation than standard care and was associated with reduced mortality.
Source: JAMA; ISICEM 2025 Presentation
Image Credit: ISICEM Congress 2025
