Editorial 

Vasopressin in Septic Shock—Is Early the Right Question?

Romain Pirracchio, Christopher W. Seymour

JAMA Published Online: October 14, 2026

doi: 10.1001/jama.2026.21703

Septic shock remains associated with unacceptably high mortality rates and, while new treatments continue to be investigated, hemodynamic support remains the primary therapeutic strategy. The 2026 Surviving Sepsis Campaign¹ recommends adding vasopressin for hemodynamic support when a patient requires escalating doses of norepinephrine. But the optimal timing for vasopressin initiation remains unknown.

In this issue of JAMA, Peltan and colleagues² report the results of the Vasopressin for Septic Shock Pragmatic (VASSPR) trial. The VASSPR trial is the first randomized trial comparing 2 vasopressin initiation strategies. Peltan and colleagues² used a cluster-crossover pragmatic design, randomizing initiation of a fixed dose of vasopressin either when the total dose of other vasopressors reached 0.1 µg/kg/min of norepinephrine or equivalent (the lower-threshold [earlier] strategy) or when the total dose of other vasopressors reached 0.4 µg/kg/min of norepinephrine or equivalent (the higher-threshold [later] strategy). The primary outcome was 28-day mortality.

Thirteen hospitals were randomized for the first trial month to an initial strategy and then crossed over to the alternate strategy monthly thereafter. A total of 3959 unique patients with septic shock were enrolled in the study (1975 patients in the lower-threshold strategy and 1984 in the higher-threshold strategy). The trial was pragmatic by design to mimic the real-life implementation of a new policy at a given hospital, leaving the clinician with the opportunity to follow or not follow the assigned initiation threshold.

The primary result was neutral. Within 28 days, 628 of the 1975 patients (31.8%) assigned to the lower-threshold vasopressin initiation strategy died and 655 of the 1984 patients (33.0%) assigned to the higher-threshold strategy died (adjusted marginal odds ratio, 0.98 [95% CI, 0.87 to 1.06], P = .62; adjusted risk difference, −0.4 [95% CI, −2.7 to 1.3] percentage points).² Similarly, no clinically meaningful difference was found for any of the secondary outcomes. In the context of a neutral result, the authors examined the experimental separation between study groups among the strategy-eligible patients (≥0.07 µg/kg/min of norepinephrine or equivalent) (1521 patients in the lower-threshold group and 1526 in the higher-threshold group) and found that, as expected, vasopressin was administered more frequently in the lower-threshold group (1034 [68.0%] vs 651 [42.7%] in the higher-threshold group), with a lower maximum dose of vasopressor administered prior to vasopressin initiation in those who received vasopressin (difference in medians, −0.20 [95% CI, −0.22 to −0.16] µg/kg/min of norepinephrine or equivalent) and a faster initiation from the time of enrollment (difference in medians, −0.8 hours [95% CI, −1.4 to −0.4 hours]).² The authors concluded that a hospital-level strategy recommending a lower vasopressor threshold for initiating vasopressin in patients with septic shock, which resulted in starting vasopressin sooner, did not decrease 28-day mortality.

How should the findings from the VASSPR trial² be incorporated into practice? Despite international clinical practice guidelines intended to promote standardized approaches to septic shock,¹ practices remain quite heterogeneous across institutions and intensivists, including when to add a second-line vasopressor such as vasopressin.^3,4 Much of the rationale for earlier vasopressin initiation derives from the VASST trial.⁵ Although the VASST trial⁵ was neutral overall for 28-day mortality (35.4% in vasopressin group vs 39.3% in norepinephrine group; P = .26), a prespecified subgroup with less severe shock (defined as treatment with <15 µg/min of norepinephrine or the equivalent) showed improved survival with adjunctive vasopressin (26.5% vs 35.7%, respectively; P = .05). This post hoc observation raised the possibility that patients with less severe shock were essentially presenting earlier in their course. Hence, the notion was posited that earlier initiation may be better. Given that agents such as norepinephrine are associated with significant adverse effects, earlier initiation of vasopressin might be norepinephrine sparing. After the VASST trial,⁵ observational studies also reported an association between early vasopressin use and lower mortality.⁶ Recently, 2 target trial emulation studies^7,8 reported findings consistent with a possible mortality reduction with early vasopressin initiation.

How should these earlier findings be reconciled with the VASSPR trial? Given the nonrandomized nature of the evidence prior to the VASSPR trial,² one possible explanation is that the previous results were biased due to residual confounding by indication. Another possibility is that the optimal timing of vasopressin initiation varies across patients. Peltan and colleagues² did explore treatment effect heterogeneity using a prespecified subgroup analysis and did not find any significant interaction. However, this approach to heterogeneity of treatment effect exploration is known to carry substantial limitations, especially when the trial uses a cluster design, including limited power and risk of both false-negative (due to low precision) and false-positive (from multiple testing) results.^9,10 Given the reported heterogeneity of patients with sepsis and septic shock,^11,12 it seems reasonable to assume that different sepsis presentations may require different treatment strategies, including different vasopressor strategies.¹³ In the VASSPR trial,² limited experimental separation and imperfect adherence could have attenuated the effect. Indeed, the VASSPR trial² was designed as a pragmatic trial, with the goal of estimating the effectiveness of a hospital-wide strategy for vasopressor management under real-world conditions. Practically, this design allowed the bedside clinician to follow or not follow the assigned vasopressin initiation threshold. Consistent with the previously reported variability in clinical practices,^3,4 strict adherence to an assigned hospital-level vasopressin initiation strategy was observed in only 45.1% of the lower-threshold groups and 65.6% of the higher-threshold groups. The low adherence could be explained by different factors, including the personal convictions or habits of the bedside clinicians regarding vasopressor management.

Importantly, the VASSPR trial² defined vasopressin initiation using dose of norepinephrine or equivalent alone. The norepinephrine or equivalent dose is a potential reason to diversify vasopressor support based on concerns that high-dose norepinephrine may be harmful.^14,15 However, the concept of norepinephrine sparing remains debated and does not necessarily drive vasopressor decisions at the bedside.^16,17 Although the norepinephrine or equivalent dose requirement provides important information, it represents only one dimension of the patient’s evolving hemodynamic state. At the bedside, clinicians also integrate response to initial resuscitation, markers of tissue perfusion, organ dysfunction, and other physiological features when deciding whether to add vasopressin. Thus, a threshold based on a single norepinephrine or equivalent dose may incompletely capture the clinical states in which vasopressin is most likely to provide benefit. The incomplete adherence to assigned thresholds may itself reflect this complexity in that bedside clinicians were making treatment decisions based on a richer, evolving clinical picture rather than norepinephrine or equivalent dose alone.

Perhaps the biggest problem is that the dynamic nature of vasopressor treatment decisions and their effect on the clinical trajectory is difficult to assess using a fixed randomized design of 2 decision thresholds, even with pragmatic and cluster-based designs. The treatment of patients with critical illnesses is a sequential decision process taken under uncertainty, a process that would be called a dynamic treatment regime in the biostatistical literature.¹⁸ Although a dose threshold treatment rule is typically static, a dynamic treatment regime defines a sequential decision process and maps the patient state into an optimal action over successive time points to identify the overall optimal treatment policy. In the typical care of a patient with septic shock, the decision to add vasopressin is one that could be revisited many times over the course of 1 or 2 days. Conventional randomized trials artificially compress this continually evolving time course into a single comparison at a single time point and, therefore, have limited ability to identify optimal sequences of treatment decisions over time.¹⁹ Other approaches (such as sequential, multiple assignment, randomized trial [SMART] design¹⁹) that provide sequential, response-adaptive randomization frameworks to directly operationalize a dynamic treatment regime would be better suited; however, these approaches have not been tried in this context.

Of course, multiple sequential randomizations within the first hours or days of care for a patient with septic shock will be very challenging to execute. An alternative approach to estimate the best dynamic treatment regime is to use observational data with causal inference methods or reinforcement learning. The Optimal Vasopressin Initiation in Septic Shock (OVISS) study²⁰ used a reinforcement learning approach to identify the optimal initiation rule for vasopressin in adult, critically ill patients receiving norepinephrine for septic shock. The identified rule recommended more frequent and earlier use of vasopressin than observed care patterns and was associated with reduced mortality. However, the rule derived from the OVISS study²⁰ was learned and evaluated using retrospective data, which leaves the analysis vulnerable to residual confounding. In addition, because the data were retrospective, the OVISS study²⁰ had to use an offline reinforcement learning approach (ie, a modeling scenario where the policy is learned using previously collected data without the possibility for the agent to explore the action space and evaluate patient responses to actions that were infrequently observed in the historical data).

Ultimately, the aspiration will be to deploy study designs, leveraging principles of online reinforcement learning, dynamic treatment regime optimization, adaptive randomization, and SMART designs¹⁹ that match the complexity of the bedside decision-making. Whether early or late, the optimal strategy is likely to be personalized and dynamic, highlighting the need for study designs capable of identifying optimal, dynamic, and individualized treatment regimens.