It’s Time to Consider How We Should Use Vasopressors, Rather Than Just Which We Should Use

Septic shock is one of the most common forms of circulatory failure encountered in the ICU. Norepinephrine is recommended in the Surviving Sepsis Campaign guidelines as the first-choice vasopressor to raise mean arterial pressure in septic shock (1). However, when ineffective, clinicians often use secondary agents with the intent to minimize catecholamine exposure and their associated side effects, or to achieve synergism from different blood pressure regulatory mechanisms (2). The Surviving Sepsis Campaign guidelines suggest vasopressin as the second-choice vasopressor when norepinephrine alone fails to yield an adequate mean arterial pressure and remark that vasopressin is “typically started” at norepinephrine dosages of 0.25–0.5 µg/kg/min (1). Indeed, the addition of vasopressin to catecholamines has been associated with reduced risk of atrial fibrillation compared with catecholamines alone (3), and a growing body of literature has suggested that the addition of vasopressin to norepinephrine at lower norepinephrine dosages and/or earlier may yield improvements in clinical outcomes (4–7). Accordingly, the popularity of vasopressin has grown over the years, and in the United States, it is used in approximately one-third of patients with septic shock (8) with international survey studies demonstrating even more frequent utilization (9). Despite this, guideline recommendations regarding when vasopressin should be used in septic shock are vague (1).

In this issue of Critical Care Explorations, Sacha et al (10) report on the timing of vasopressin addition in septic shock and its association with clinical outcomes using secondary data from 209 hospitals in the Medical Information Mart for Intensive Care-IV (MIMIC-IV) and electronic ICU Collaborative Research Database (eICU-CRD) databases. In a multivariable model including data from 1409 patients, they found in-hospital mortality to be higher when vasopressin was initiated at higher norepinephrine dosages, higher lactate concentrations, and when longer durations of time had lapsed from the onset of shock. Specifically, the odds of in-hospital death increased by 39% for every 10 µg/min increase in norepinephrine-equivalent dose at the time vasopressin was initiated (0.125 µg/kg/min in a 80 kg patient), with this association persisting up to a dose of 39.5 µg/min. Beyond this threshold, the association persisted but was attenuated, with an 11% increase in the odds of in-hospital death for each additional 10 µg/min delay in norepinephrine-equivalent dose. In the restricted cubic spline analysis, three nodes of norepinephrine dose thresholds were identified where the in-hospital mortality increased dramatically. Compared with when vasopressin was added at just 9 µg/min norepinephrine, the odds of in-hospital mortality increased by 90% if this was delayed to 28 µg/min norepinephrine and further increased by 393% if delayed to 72 µg/min norepinephrine. Similarly, the odds of in-hospital mortality increased by 16% for every 1 mmol/L increase in lactate concentration and 3% for every 1 hour increase in time delay at the time vasopressin was added to norepinephrine.

The authors’ work is commendable in confronting the very important question of vasopressin timing in septic shock with a comprehensive and well-thought-out analysis. While several publicly available critical care databases exist, the MIMIC-IV and eICU-CRD databases are well-suited to address this question (11). Although risk adjustment and clinical trajectories in septic shock can be complex and nuanced, both databases capture comorbid conditions and severity of illness surrogates, allowing for covariate adjustment and inclusion of relevant outcomes information to facilitate modeling. Ultimately, higher resolution physiologic and laboratory variables would be necessary to facilitate more complex modeling, which are better captured in databases like the Amsterdam University Medical Center database and the high time-resolution ICU dataset (11). Additionally, the proportion of patients receiving organ supportive therapies in MIMIC-IV and eICU-CRD is less than these other databases, likely signaling inclusion of generally “less sick” ICU patients. MIMIC has facilitated myriad investigations over the past 10 years; however, its derivation from a single center is now better appreciated as a barrier to generalizability. Including data not just from MIMIC-IV but also eICU-CRD allowed Sacha et al (10) to examine a more representative sample encompassing over 200 U.S. hospitals, increasing the applicability of their findings.

While this approach has strengths, there are also other relevant limitations to consider. Hospitals represented in eICU-CRD are typically small- to medium-sized (< 250 beds), patients may have a generally low severity of illness, and most ICUs are likely staffed by noncritical care providers, although the data have been adjudicated and appear to be complete and of good quality (12). While eICU-CRD and other databases offer a rich set of clinical variables, they often lack key elements along the causal pathway to mortality. In the study by Sacha et al (10), and in the context of septic shock, these gaps include means by which to judge the adequacy of antimicrobial therapy or the timeliness of source control, which are known but unmeasured confounders. In causal inference, these are critical links between the exposure and the outcome. When such factors are unmeasured, models may misattribute effects, leading to residual confounding. For example, in the study by Sacha et al (10), the association between vasopressor dose and mortality may reflect unmeasured treatment delays or other nuanced but important clinical phenomena rather than a direct causal effect. Confounding by known and unknown variables is an inherent limitation of retrospective analyses using routinely recorded healthcare data. While these gaps should temper interpretation, the authors’ use of careful adjustment for known confounders, a large sample size, and thoughtful sensitivity analyses helps strengthen confidence in their findings.

More broadly, secondary data, such as those used by Sacha et al (10), must always be cautiously approached and interpreted. Erroneous data that cannot be manually confirmed may be present in secondary databases, such as the grossly discrepant fluid balance between MIMIC-IV (10.5 L) and eICU-CRD (1 L) at the time vasopressin was added. Thus, the possibility of misclassification bias should always be considered with secondary data, which may distort or obscure true associations. To address this, Sacha et al (10) conducted sensitivity analyses to test the robustness of their primary findings. Those included examination of important variables that are also prone to misclassification, such as the definition chosen for septic shock, fluid balance status, and vasopressor dosing units, and these sensitivity analyses demonstrated nearly identical point estimates. Finally, we commend Sacha et al (10) for making their analytic code publicly available to increase the transparency and reproducibility of their analysis, a recommended practice that is often omitted.

The findings of Sacha et al (10), taken in concert with other relevant literature, have important clinical implications for the care of patients with septic shock. While randomized trial evidence has not consistently demonstrated an improvement in survival with the addition of vasopressin in septic shock (13), observational studies have suggested that the manner in which it is applied is potentially consequential. A multicenter study in Australia found that vasopressin addition in the first 6 hours of septic shock was associated with 31% less odds of in-hospital death compared with its later addition (14). Another analysis by Sacha et al (15) using their local data found that the odds of in-hospital death increased by nearly 21% for every 10 µg/min increase in norepinephrine dose at the time vasopressin was added. Recently, a reinforcement learning model that was trained on clinician behavior suggested initiating vasopressin in more patients with septic shock, at lower dosages of norepinephrine, and sooner after the onset of shock (4). Finally, a multicenter target trial emulation demonstrated reduced risk of 30-day mortality when vasopressin was added in the first 6 hours of septic shock and when norepinephrine dosages were less than 0.25 µg/kg/min (7).

While norepinephrine dose may be viewed as merely a surrogate of shock severity or progression, its use at the bedside to inform decision-making appears both practical and commonplace (9). Of note, the Surviving Sepsis Campaign guidelines state vasopressin is typically started when norepinephrine dosages are around 0.25–0.5 µg/kg/min (1). Importantly, this equates to 20–40 µg/min in an 80 kg patient, which is considerably higher than the optimal norepinephrine dosages identified by Sacha et al (10) and other observational analyses. Additionally, the seminal Vasopressin and Septic Shock Trial (VASST) trial found that patients randomized to vasopressin when the norepinephrine dose was less than 15 µg/min were afforded a greater than 20% lower risk of death compared with placebo, an effect not observed in patients initiated on vasopressin when above this threshold (16). While it remains to be seen whether the current evidence is sufficient to shift clinical guidelines, the study by Sacha et al (10) adds to a growing body of literature suggesting that the timing of vasopressin initiation may be a modifiable and consequential aspect of clinical care for patients with septic shock.