Vasopressin Initiation Timing and In-Hospital Mortality in Septic Shock: An Observational Study of Large Public Databases

Abstract

IMPORTANCE:

Vasopressin is used in one-third of patients with septic shock to augment hemodynamics and reduce overall catecholamine exposure. However, the optimal clinical context in which to initiate vasopressin is unknown.

OBJECTIVES:

To determine the association between norepinephrine-equivalent dose, lactate concentration, and time duration from shock onset at vasopressin initiation with in-hospital mortality

DESIGN, SETTING, AND PARTICIPANTS:

Retrospective, observational evaluation utilizing Medical Information Mart for Intensive Care-IV and electronic ICU Collaborative Research Database databases of adult patients with septic shock based on modified Sepsis-3 criteria receiving continuous infusion catecholamines.

MAIN OUTCOMES AND MEASURES:

The associations of norepinephrine-equivalent dose, lactate concentration, and time duration from shock onset at vasopressin initiation with in-hospital mortality were evaluated with multivariable regression models.

RESULTS:

In total, 1409 patients from 209 hospitals were included. At vasopressin initiation patients had a median (interquartile range) norepinephrine-equivalent dose 28.4 µg/min (16.4–42.6 µg/min), lactate concentration 3.7 mmol/L (2.5–6.2 mmol/L), and 5.6 hours (2.0–13.5 hr) had elapsed since shock onset. All three variables of interest were associated with in-hospital mortality. Three restricted cubic spline knots were identified where the relationship between norepinephrine-equivalent dose and in-hospital mortality changed substantially: 9, 28, and 72 µg/min. The odds of in-hospital mortality increased by 90% and 3.9-fold when comparing vasopressin initiation at norepinephrine-equivalent doses of 28 µg/min and 72 µg/min to 9 µg/min, respectively (adjusted odds ratio [OR], 1.90 [95% CI, 1.49–2.41] and 3.93 [95% CI, 2.74–5.64]). The odds of in-hospital mortality increased by 16% for every mmol/L in the lactate concentration at vasopressin initiation (adjusted OR, 1.16 [95% CI, 1.11–1.21]). Finally, the odds of in-hospital mortality increased by 3% for every hour in the time duration from shock onset to vasopressin initiation (adjusted OR, 1.03 [95% CI, 1.01–1.04]).

CONCLUSIONS AND RELEVANCE:

Earlier adjunctive vasopressin initiation may decrease mortality in patients with septic shock.

KEY POINTS

Question: What is the association between norepinephrine-equivalent dose, lactate concentration, and time duration from shock onset at vasopressin initiation with in-hospital mortality in patients with septic shock?

Findings: In 1409 included patients (209 hospitals), the odds of in-hospital mortality increased by 90% when vasopressin was initiated at a norepinephrine-equivalent dose of 28 µg/min compared with 9 µg/min.

Meaning: Lower norepinephrine-equivalent dose at vasopressin initiation, lower lactate concentration, and shorter time duration from shock onset to vasopressin initiation were associated with reduced in-hospital mortality in patients with septic shock.

Septic shock is a highly morbid disease with mortality rates exceeding 50% (1–3). Vasoactive agents, both catecholamine and noncatecholamine derived, are often required to augment hemodynamics and restore perfusion in this population. The first-line vasoactive agent recommended by the Surviving Sepsis Campaign (SSC) guidelines for patients with septic shock is norepinephrine (4). In patients who continue to remain below goal mean arterial pressure (MAP), vasopressin is suggested as the second-line adjunctive agent rather than continuing to escalate norepinephrine doses. To date, there remain conflicting data on the overall clinical impact of vasopressin and its analogues in patients with septic shock (5–12). Yet vasopressin is initiated in one-third of patients with this disease state, with frequencies continuing to rise (13–15).

Guidance on the clinical context to initiate vasopressin is limited. The 2021 SSC guidelines, for the first time, provide a remark (an “in our practice statement”) noting that, “…vasopressin is usually started when the dose of norepinephrine is in the range of 0.25–0.5 mcg/kg/min.” Additionally, the SSC guidelines noted that starting within this range “seems sensible” (4). Despite this remark, the optimal timing of vasopressin initiation has yet to be elucidated in clinical trials. The landmark Vasopressin and Septic Shock Trial (VASST) found beneficial associations with the utilization of vasopressin in subgroups of patients who were deemed less severely ill: those with norepinephrine doses less than 15 µg/min and those with lactate concentrations less than or equal to 1.4 mmol/L at the time of randomization (10). While these findings were not confirmed in a meta-analysis, they support the theory that vasopressin should be initiated early in the course of septic shock (8, 16).

We recently reported a single healthsystem observational study that found an association between higher in-hospital mortality and both higher norepinephrine-equivalent dose at the time of vasopressin initiation and higher lactate concentration at the time of vasopressin initiation (17). Although the study supports the subgroup findings from VASST, it is limited by its single healthsystem nature. Additionally, these findings have not been corroborated with larger, more diverse datasets. The current study sought to evaluate the association of catecholamine dose, lactate concentration, and timing of vasopressin initiation with mortality in patients admitted to hospitals available in publicly sourced U.S. databases with septic shock who received vasopressin. The hypothesis was that vasopressin initiation in less severe shock would be associated with lower in-hospital mortality.

STUDY DESIGN AND METHODS

This was a retrospective, observational evaluation of adult patients, age 18 years or greater, admitted to an ICU included in either the electronic ICU Collaborative Research Database (eICU-CRD; from 2014 to 2015) or Medical Information Mart for Intensive Care-IV (MIMIC-IV; Version 2.2; from 2008 to 2019) PhysioNet databases (18–22). Patients were included in the study if they had septic shock and received continuous infusion catecholamines and adjunctive vasopressin. There was a subset of patients in the eICU-CRD database with norepinephrine doses that could not be determined due to incomplete documentation strategies and concentration omissions. These patients were not included in this study to ensure reliable documentation of norepinephrine dosages. Additionally, patients were excluded if they had vasopressin and catecholamines initiated at the same time, vasopressin initiated more than 48 hours after catecholamine initiation, or had no lactate measurement available at the time of vasopressin initiation. Only the first patient encounter where inclusion criteria were met was included and subsequent/duplicate encounters were excluded. The primary objective of this study was to evaluate the independent association of catecholamine dose, lactate concentration, and timing from shock onset at vasopressin initiation with in-hospital mortality in patients in the eICU-CRD and MIMIC-IV databases.

Septic shock was defined based on a modified Sepsis-3 clinical criteria definition (23, 24) including the presence of continuous infusion catecholamines, lactate concentration greater than or equal to 2.0 mmol/L within 72 hours of vasoactive initiation, and antibiotic utilization within 72 hours of vasoactive initiation. Sensitivity analyses were conducted utilizing study populations based on three alternate definitions for septic shock case identification: 1) septic shock via clinical criteria (vasoactive initiation plus, within 24 hr, lactate concentration ≥ 2.0 mmol/L, and antibiotic utilization within 24 hr of vasoactive initiation) (23); 2) severe sepsis or septic shock defined by explicit criteria from International Classification of Diseases, 9th Edition (ICD-9) and International Classification of Diseases, 10th Edition (ICD-10) coding (ICD-10 codes: R65.2 severe sepsis or R65.21 septic shock; ICD-9 codes: 995.92 severe sepsis or 785.52 septic shock); and 3) the Angus method utilizing implicit criteria from ICD-9 and ICD-10 codes to identify severe sepsis or septic shock (eTable 16, https://links.lww.com/CCX/B526) (14, 15, 25–27). For the septic shock definition based on clinical criteria, the need for vasopressor therapy was used alone without the requirement for a MAP value less than 65, as it was assumed that patients receiving vasopressors had a clinically significant hypotensive state. Additionally, due to limited blood culture data available in the databases, microbiological culture data was unable to be collected. Norepinephrine-equivalent dose was calculated as follows: norepinephrine (µg/min) + epinephrine (µg/min) + dopamine (µg/kg/min)/2 + phenylephrine (µg/min)/10 (10). Acute kidney injury (AKI) was evaluated at the time of vasopressin initiation and was defined as meeting stage 2 or 3 AKI per the AKI Network definition with missing baseline serum creatinine values back calculated, as previously described (28). Baseline comorbidities including diabetes mellitus, cirrhosis, chronic obstructive pulmonary disease, and immune suppression were identified based on diagnostic coding in eICU-CRD and in past medical history documentation of the discharge notes for MIMIC-IV. Sequential Organ Failure Assessment (SOFA) score was calculated on each calendar based on the most extreme daily value of all SOFA score components (29). Shock onset was defined as the time of initiation of continuous infusion catecholamines. Last, hemodynamic response was defined as achievement of both a decrease in norepinephrine-equivalent dose and MAP greater than or equal to 65 mm Hg at 6 hours after vasopressin initiation in patients who survived until hour 6 (30).

Baseline characteristics and clinical outcomes are detailed based on the database source (MIMIC-IV vs. eICU-CRD) and described as mean ± sd and n (%). The three variables (predictors) of interest defined a priori were: 1) norepinephrine-equivalent dose at the time of vasopressin initiation; 2) lactate concentration at the time of vasopressin initiation; and 3) time duration from shock onset to vasopressin initiation. Two methods were used to evaluate the association between each of the variables of interest and in-hospital mortality, preformed similarly as in our previously published study (17): logistic regression utilizing restricted cubic splines and logistic regression utilizing piecewise segmented regression (31). For the restricted cubic splines approach, which modeled a nonlinear relationship between variables using a series of polynomial functions, model selection by the analysis of variance combined with data visualization was performed to decide if the usage of restricted cubic splines improved model fit, and the number of knots (connecting points between segments) was selected to ensure the parsimoniousness of the model and avoid overfitting (32). All three variables of interest were evaluated for a nonlinear relationship with in-hospital mortality by testing the significance of the nonlinear component in a restricted cubic spline model using a Wald test. Norepinephrine-equivalent dose at the time of vasopressin initiation was the only variable that demonstrated a substantial nonlinear relationship with in-hospital mortality and, because of this, was analyzed by restricted cubic splines with three knots (p of nonlinear effect = 0.007). Selection of three knots was based on clinical interpretation and statistical evaluation. Restricted cubic spline models with three, four, and five knots were compared using Akaike Information Criterion, Bayesian Information Criterion, area under the curve, and R² values. Since these metrics were similar across models, the simplest model with three knots was selected. Restricted cubic splines were not used for the models evaluating lactate concentration at time of vasopressin initiation or for the time duration from shock onset because no substantial nonlinear effect was detected. Finally, piecewise segmented regression was used to determine the breakpoint at which the relationship between norepinephrine-equivalent dose at vasopressin initiation and in-hospital mortality changed. This method estimates the change point of slope in norepinephrine-equivalent dose at vasopressin initiation, or the point at which the odds ratio (OR) of the association between norepinephrine-equivalent dose at vasopressin initiation and in-hospital mortality changes. Owing to the linear relationship between in-hospital mortality at vasopressin initiation, in the models evaluating the lactate concentration, and the time duration from shock onset, segmented regression was not performed. Evaluation of the presence of interaction between each of the three variables of interest (norepinephrine-equivalent dose at vasopressin initiation, lactate concentration at vasopressin initiation, and time duration from shock onset to vasopressin initiation) was evaluated, and no substantial interaction between these three variables was detected (each p > 0.5). Both models included all three variables of interest and known confounders for the relationship between the variables of interest and in-hospital mortality based on directed acyclic graphs (eFig. 1, https://links.lww.com/CCX/B526), and report the direct effects of the three variables of interest (33, 34). Prior literature and clinical expertise guided the developed relationship structure of included variables. The independent confounding variables of interest included in the developed models were the database source (MIMIC-IV vs. eICU-CRD), age, weight, sex, race, immune suppression, ICU location, hydrocortisone, mechanical ventilation, AKI, SOFA score, fluid balance, and fluid bolus administration. Multicollinearity of included variables in the segmented regression model was assessed using pairwise correlations and variance inflation factors. No evidence of multicollinearity was detected as all pairwise correlations were below 0.5 and all variance inflation factors were less than 2. Because of the eligibility criteria no patients were included that had missing data for the three main variables of interest. The only cofounding variable with missing data was weight in three patients. For all analyses, complete case analyses was conducted and these three patients were removed from the regression models. The same approaches for logistic regression modeling were used in the sensitivity analyses using study populations based on alternative septic shock case identification definitions. Details regarding the utilization of restricted cubic splines and segmented regression are outlined in the Supplemental Digital Content for each sensitivity analysis (eTable 1, https://links.lww.com/CCX/B526). Associations between the three variables of interest were also evaluated for two secondary outcomes: hemodynamic response to vasopressin initiation and SOFA score change at 48 hours. Both secondary outcomes were evaluated in a similar approach detailed with the primary outcome of interest, in-hospital mortality. Secondary outcomes were only evaluated in the primary study population (patients identified using the modified Sepsis-3 clinical criteria). To evaluate consistency of results with vasopressors originally prescribed as µg/kg/min rather than µg/min, a post hoc analysis of patients with weight-based dosing of vasopressors was conducted on the primary study population. Only patients included in the MIMIC-IV database had vasopressors originally prescribed as µg/kg/min. All analyses were conducted at an overall significance level of 0.05, using R, Version 4.2 (R foundation for Statistical Computing, Vienna, Austria). The analytic code for cohort development, data cleaning, and all analyses reported in this article is publicly available on GitHub (https://github.com/gretchensacha/timing-of-vasopressin-initiation).

RESULTS

After screening of all patients included in the MIMIC-IV and eICU-CRD databases, 9871 patients who met criteria for septic shock based on the modified Sepsis-3 clinical criteria were screened for inclusion in this study. After eligibility criteria were applied (Fig. 1), 1409 patients from 209 hospitals were included in the primary analysis: 1007 from MIMIC-IV and 402 from eICU-CRD.

Figure 1.

Patient flow diagram. eICU-CRD = electronic ICU Collaborative Research Database, MIMIC-IV = Medical Information Mart for Intensive Care-IV.

Baseline characteristics of included patients are detailed in Tables 1 and 2. Included patients were, on average, 64 ± 15 years old with most patients being White (65.9%). A preponderance of patients were admitted to a mixed ICU (33.1%), with 22.9% admitted to a medical ICU and 23.5% admitted to a cardiovascular ICU. At the time of vasopressin initiation, patients were receiving a median norepinephrine-equivalent dose of 28.4 µg/min (16.4–42.6 µg/min), had a median lactate concentration of 3.7 mmol/L (2.5–6.2 mmol/L), and had vasopressin initiated at 5.6 hours (2.0–13.5 hr) after shock onset. At baseline, included patients in eICU-CRD were more severely ill. Additional baseline characteristics are detailed in eTables 2 and 3 (https://links.lww.com/CCX/B526). Of included patients, 57.6% did not survive the hospital admission and 38.1% had a positive hemodynamic response after vasopressin initiation (eTable 4, https://links.lww.com/CCX/B526).

TABLE 1.

Baseline Characteristics Based on Data Source

Characteristic	Total (n = 1409)	MIMIC-IV (n = 1007)	eICU-CRD (n = 402)
Characteristics at ICU admission
Age, yra	64 ± 15	64 ± 15	62 ± 15
Male, n (%)	840 (59.6)	607 (60.3)	233 (58.0)
Race, n (%)
White	928 (65.9)	600 (59.6)	328 (81.6)
Black	144 (10.2)	108 (10.7)	36 (9.0)
Other	337 (23.9)	299 (29.7)	38 (9.5)
ICU location, n (%)
Mixed ICU	467 (33.1)	203 (20.2)	264 (65.7)
Medical ICU	323 (22.9)	310 (30.8)	13 (3.2)
Cardiac ICU	331 (23.5)	248 (24.6)	83 (20.6)
Surgical ICU	275 (19.5)	245 (24.3)	20 (7.5)
Neurosciences ICU	13 (0.9)	1 (0.1)	12 (3.0)
Weight, kga,b	84.5 ± 25.9	84.3 ± 25.6	85.1 ± 26.6
Diabetes mellitus, n (%)	340 (24.1)	223 (22.1)	117 (29.1)
Cirrhosis or hepatic failure, n (%)	225 (16.0)	176 (17.5)	49 (12.2)
Chronic obstructive pulmonary disease, n (%)	206 (14.6)	131 (13.0)	75 (18.7)
Immune suppression, n (%)	345 (24.5)	254 (25.2)	91 (22.6)
Characteristics at vasopressin initiation
Mechanical ventilation, n (%)	848 (60.2)	539 (53.5)	309 (76.9)
Sequential Organ Failure Assessment scorea	9.8 ± 3.0	8.9 ± 2.6	10.4 ± 3.1
Initial vasopressin dose, U/min	0.039 ± 0.045	0.039 ± 0.052	0.038 ± 0.018
Initial vasopressin dose, n (%)
0.03 U/min	127 (9.0)	32 (3.2)	95 (23.6)
0.04 U/min	926 (65.7)	682 (67.7)	244 (60.7)
Other	356 (25.3)	293 (29.1)	63 (15.7)
Lactate at AVP initiation, mmol/L	3.7 (2.5–6.2)	3.5 (2.5–5.7)	4.1 (2.6–7.5)
Time duration from shock onset to AVP, hr	5.6 (2.0–13.5)	5.0 (1.5–13.1)	7.2 (3.5–14.8)
Norepinephrine-equivalent dose at AVP initiation, µg/min	28.4 (16.4–42.6)	27.6 (16.7–41.4)	30.0 (15.7–48.8)
Norepinephrine-equivalent dose at AVP initiation, µg/kg/minb	0.4 (0.2–0.5)	0.4 (0.2–0.5)	0.3 (0.2–0.6)

AVP = arginine vasopressin, eICU-CRD = electronic ICU Collaborative Research Database, MIMIC-IV = Medical Information Mart for Intensive Care-IV.

^aData are presented as mean ± sd.

^bData are available in 1406 patients.

Data are presented as median (interquartile range) or n (%), unless otherwise specified.

TABLE 2.

Baseline Characteristics Based on Hospital Survival

Characteristic	Total (n = 1409)	Survivor (n = 597)	Nonsurvivor (n = 812)
Characteristics at ICU admission
Age, yra	64 ± 15	62 ± 15	65 ± 14
Male, n (%)	840 (59.6)	377 (63.1)	463 (57.0)
Race, n (%)
White	928 (65.9)	395 (66.2)	533 (65.6)
Black	144 (10.2)	68 (11.4)	76 (9.4)
Other	337 (23.9)	134 (22.4)	203 (25.0)
ICU location, n (%)
Mixed ICU	467 (33.1)	188 (31.5)	279 (34.4)
Medical ICU	323 (22.9)	112 (18.8)	211 (26.0)
Cardiac ICU	331 (23.5)	168 (28.1)	163 (20.1)
Surgical ICU	275 (19.5)	126 (21.1)	149 (18.3)
Neurosciences ICU	13 (0.9)	3 (0.5)	10 (1.2)
Weight, kga,b	84.5 ± 25.9	84.9 ± 23.4	84.2 ± 27.6
Diabetes mellitus, n (%)	340 (24.1)	151 (25.3)	189 (23.3)
Cirrhosis or hepatic failure, n (%)	225 (16.0)	61 (10.2)	164 (20.2)
Chronic obstructive pulmonary disease, n (%)	206 (14.6)	70 (11.7)	136 (16.7)
Immune suppression, n (%)	345 (24.5)	140 (23.5)	205 (25.2)
Characteristics at vasopressin initiation
Mechanical ventilation, n (%)	848 (60.2)	360 (60.3)	488 (60.1)
Sequential Organ Failure Assessment scorea	9.8 ± 3.0	8.9 ± 2.6	10.4 ± 3.1
Initial vasopressin dose, U/min	0.039 ± 0.045	0.040 ± 0.067	0.038 ± 0.015
Initial vasopressin dose, n (%)
0.03 U/min	127 (9.0)	49 (8.2)	78 (9.6)
0.04 U/min	926 (65.7)	375 (62.8)	551 (67.9)
Other	356 (25.3)	173 (29.0)	183 (22.5)
Lactate at AVP initiation, mmol/L	3.7 (2.5–6.2)	3.1 (2.4–4.5)	4.4 (2.8–7.6)
Time duration from shock onset to AVP, hr	5.6 (2.0–13.5)	4.8 (1.5–11.5)	6.2 (2.3–15.0)
Norepinephrine-equivalent dose at AVP initiation, µg/min	28.4 (16.4–42.6)	22.5 (13.1–33.2)	31.9 (20.0–50.0)
Norepinephrine-equivalent dose at AVP initiation, µg/kg/minb	0.4 (0.2–0.5)	0.3 (0.2–0.4)	0.4 (0.3–0.6)

AVP = arginine vasopressin.

^aData are presented as mean ± sd.

^bData are available in 1406 patients.

Data are presented as median (interquartile range) or n (%), unless otherwise specified.

The multivariable logistic regression model utilizing restricted cubic splines for the norepinephrine-equivalent dose at the time of vasopressin initiation identified three knots where the relationship between with in-hospital mortality changed substantially: 9, 28, and 72 µg/min. The multivariable-adjusted predicted in-hospital mortality rate at each of the three knots was: 47.9% (95% CI, 36.3–59.6%), 63.5% (95% CI, 53.2–72.7%), and 78.3% (95% CI, 69.6–85.0%), respectively. There was an independent association between the norepinephrine-equivalent dose at vasopressin initiation and in-hospital mortality after adjustment for known confounders (Fig. 2). When the norepinephrine-equivalent dose was 28 µg/min (compared with 9 µg/min) at the time of vasopressin initiation, the odds of in-hospital mortality increased by 90% (adjusted OR, 1.90 [95% CI, 1.49–2.41]). When the norepinephrine-equivalent dose was 72 µg/min (compared with 9 µg/min) at the time of vasopressin initiation, the odds of in-hospital mortality increased by 3.9-fold (adjusted OR, 3.93 [95% CI, 2.74–5.64]). When the norepinephrine-equivalent dose was 72 µg/min (compared with 28 µg/min) at the time of vasopressin initiation, the odds of in-hospital mortality increased by two-fold (adjusted OR, 2.07 [95% CI, 1.68–2.57]).

Figure 2.

Predicted rate of in-hospital mortality based on norepinephrine-equivalent dose at vasopressin initiation. Predicted rate of in-hospital mortality by norepinephrine-equivalent dosage at time of vasopressin initiation using multivariable logistic regression with restricted cubic splines using three knots (at 9, 28, and 72 µg/min). The shaded gray area indicates the 95% CI of the predicted incidence of in-hospital mortality. Model adjusted for the following known confounders: lactate at vasopressin initiation (fixed at the median value of 3.7 mmol/L), time duration from shock onset (fixed at the median value of 5.6 hr), Medical Information Mart for Intensive Care-IV database, age (fixed at 65 yr), weight (fixed at 80 kg), male sex, Caucasian race, no immune suppression, medical ICU admission, receiving hydrocortisone, mechanical ventilation, no acute kidney injury, baseline Sequential Organ Failure Assessment score (fixed at a score of 10), fluid balance before vasopressin initiation (fixed at positive 8.5 L), and volume of fluid bolus administered before vasopressin initiation (fixed at 1 L).

There was also an independent association detected between the lactate concentration at the time of vasopressin initiation and in-hospital mortality (adjusted OR, 1.16 [95% CI, 1.11–1.21] for every 1 mmol/L increase in lactate concentration) (Fig. 3). Last, there was an independent association detected between the time duration from shock onset to vasopressin initiation and in-hospital mortality (adjusted OR, 1.03 [95% CI, 1.01–1.04] for every 1 hr increase in time duration) (Fig. 4).

Figure 3.

Predicted incidence of in-hospital mortality based on lactate concentration at vasopressin initiation. Predicted incidence of in-hospital mortality by lactate concentration at time of vasopressin initiation using multivariable logistic regression. The shaded gray area indicates the 95% CI of the predicted incidence of in-hospital mortality. Model adjusted for the following known confounders: norepinephrine-equivalent dose (incorporated as a restricted cubic spline with three knots in the model, with value fixed at the median value of 28.4 µg/min), time duration from shock onset (fixed at the median value of 5.6 hr), Medical Information Mart for Intensive Care-IV database, age (fixed at 65 yr), weight (fixed at 80 kg), male sex, Caucasian race, no immune suppression, medical ICU admission, receiving hydrocortisone, mechanical ventilation, no acute kidney injury, baseline Sequential Organ Failure Assessment score (fixed at a score of 10), fluid balance before vasopressin initiation (fixed at positive 8.5 L), and volume of fluid bolus administered before vasopressin initiation (fixed at 1 L).

Figure 4.

Predicted incidence of in-hospital mortality based on time duration from shock onset to vasopressin initiation. Predicted incidence of in-hospital mortality by time duration from shock onset to vasopressin initiation using multivariable logistic regression. The shaded gray area indicates the 95% CI of the predicted incidence of in-hospital mortality. Model adjusted for the following known confounders: norepinephrine-equivalent dose (incorporated as a restricted cubic spline with three knots in the model, with value fixed at the median value of 28.4 µg/min), lactate at vasopressin initiation (fixed at the median value of 3.7 mmol/L), Medical Information Mart for Intensive Care-IV database, age (fixed at 65 yr), weight (fixed at 80 kg), male sex, Caucasian race, no immune suppression, medical ICU admission, receiving hydrocortisone, mechanical ventilation, no acute kidney injury, baseline Sequential Organ Failure Assessment score (fixed at a score of 10), fluid balance before vasopressin initiation (fixed at positive 8.5 L), and volume of fluid bolus administered before vasopressin initiation (fixed at 1 L).

Results of the logistic regression model utilizing piecewise segmented regression are detailed in the Supplemental Digital Content and were similar to the results of the logistic regression model utilizing restricted cubic splines (eTable 5 and eFig. 2, https://links.lww.com/CCX/B526). Figures depicting the unadjusted in-hospital mortality incidence based on the combinations of quartile divisions of the three independent variables of interest are presented in eFigure 3 (https://links.lww.com/CCX/B526). Results were similar when evaluating the patients prescribed vasopressors dosed as µg/kg/min (eTable 6, https://links.lww.com/CCX/B526). Additionally, results were similar in the sensitivity analyses using alternative definitions for septic shock case identification (eFigs. 4–10 and eTables 7–15, https://links.lww.com/CCX/B526). Because of the unexpected range in values in fluid balance seen in the included patients (interquartile range up to 18 L positive fluid balance), a post hoc sensitivity analysis removing fluid balance as a covariate was conducted and all results remained identical to the analyses with the inclusion of fluid balance.

The only variable that was found to associate with hemodynamic response after vasopressin’s initiation was the lactate concentration at the time of vasopressin initiation (adjusted OR, 0.94 [95% CI, 0.91–0.97]). No association was detected between norepinephrine-equivalent dose at the time of vasopressin initiation (adjusted OR, 0.99 [95% CI, 0.99–1.00]) or the time duration from shock onset (adjusted OR, 0.99 [95% CI, 0.99–1.01]) and hemodynamic response after vasopressin initiation. Increasing norepinephrine-equivalent dose at the time of vasopressin initiation was independently associated with increased SOFA score change at 48 hours: adjusted beta-coefficient, 0.20 (95% CI, 0.12–0.27) for each 10 µg/min increase in norepinephrine-equivalent dose. Similarly, the lactate concentration at the time of vasopressin initiation was independently associated with an increase SOFA score change at 48 hours: adjusted beta-coefficient, 0.12 (95% CI, 0.05–0.19) for each 1 mmol/L increase in lactate concentration. The time duration from shock onset was not associated with SOFA score change at 48 hours.

DISCUSSION

This large, retrospective, evaluation of 1409 patients with septic shock admitted to multiple U.S. hospitals found an association between higher in-hospital mortality and higher norepinephrine-equivalent dose at vasopressin initiation, increased lactate concentration at vasopressin initiation, and with longer duration between shock onset to vasopressin initiation. The odds of in-hospital mortality were 90% higher when vasopressin was initiated at a norepinephrine-equivalent dose of 28 µg/min compared with 9 µg/min, and 3.9 times higher when initiated at 72 µg/min compared with 9 µg/min. The odds of in-hospital mortality increased by 16% for every 1 mmol/L increase in the lactate concentration at the time of vasopressin. Finally, the odds of in-hospital mortality increased by 3% for every hour increase in the time from shock onset to vasopressin initiation. These results indicate that patients with septic shock, in whom adjunctive vasopressin will ultimately be initiated, will benefit from vasopressin initiation earlier in their shock presentation.

These results align with the subgroup analyses from the landmark VASST study. Patients included in the VASST study who required norepinephrine doses between 5 and 14 µg/min at the time of randomization had improved mortality when randomized to receive adjunctive vasopressin compared with patients who received norepinephrine alone (90-d mortality 35.8% vs. 46.1%; absolute difference, –10.4% [95% CI, –20.3% to –0.4%]) (10). Further, patients with lactate concentrations less than or equal to 1.4 mmol/L at the time of randomization had lower all-cause mortality rates when randomized to receive vasopressin (18.9% vs. 33.8%; absolute difference, –14.9% [95% CI, –27.9% to –1.5%]). Additionally, these results corroborate the recently reported single healthsystem observational study of 1610 patients with septic shock (17). In the single healthsystem study, the odds of in-hospital mortality increased 20.7% for every 10 µg/min increase in norepinephrine-equivalent dose at vasopressin initiation, up to a norepinephrine-equivalent dose of 60 µg/min (adjusted OR, 1.21; 95% CI, 1.09–1.34) (17). These results are similar to the current evaluation, which found the odds of in-hospital mortality increased by 39% for every 10 µg/min increase in norepinephrine-equivalent dose at vasopressin initiation, up to a norepinephrine-equivalent dose of 39.5 µg/min (adjusted OR, 1.39; 95% CI, 1.21–1.61). These results also corroborate a recent evaluation of the MIMIC-III and MIMIC-IV databases in which initiation of vasopressin when the norepinephrine dose was less than 0.25 µg/kg/min was associated with reduced 28-day mortality compared with initiation when the norepinephrine dose was greater than or equal to 0.25 µg/kg/min (aOR, 0.66; 95% CI, 0.52–0.84) (35). Last, a recent reinforcement learning model provided similar results suggesting vasopressin initiation at lower norepinephrine doses (0.20 µg/kg/min [95% CI, 0.08–0.45 µg/kg/min] vs. 0.37 µg/kg/min [95% CI, 0.17–0.69 µg/kg/min]), at lower lactate concentrations (2.5 mmol/L [95% CI, 1.7–4.9 mmol/L] vs. 3.6 mmol/L [95% CI, 1.8–6.8 mmol/L]), and earlier relative to shock onset (4 hr [95% CI, 1–8 hr] vs. 5 hr [95% CI, 1–14 hr]) compared with clinician observed action (36).

The current study found an association between higher lactate concentration at vasopressin initiation and higher in-hospital mortality, similar to the single healthsystem study. Further, the point estimate in the current study for the association between the time duration from shock onset to vasopressin initiation and in-hospital mortality (adjusted OR 1.03) was similar to the point estimates observed in the single healthsystem study (adjusted OR range from 1.01 to 1.04 due to statistical interaction with lactate concentration), but the 95% CI in the current study was narrower than the 95% CIs in the previous study (17). The reason for this difference is unclear but could be due to dissimilarities in study populations between the studies (e.g., predominantly medical ICU and higher SOFA scores in the previous study). Regardless, the finding in the current study in which the odds of in-hospital mortality increased by 3% for every hour delay in the time of vasopressin initiation after shock onset (adjusted OR, 1.03; 95% CI, 1.01–1.04) may not be clinically meaningful. Overall, it is noteworthy that the data from the current study continue to support the narrative of earlier vasopressin initiation in the course of septic shock (16, 36).

The current study independently evaluated three variables to assess their impact on the timing of vasopressin initiation. While the potential interplay between these variables cannot be ruled out, no statistical interaction (effect modification) was detected in this analysis. However, integrating all three variables into bedside clinical decision-making remains a complex challenge. Focusing on the norepinephrine-equivalent dose as the indicator for vasopressin initiation may be the most practical for several reasons. The time from shock onset to vasopressin initiation has not consistently been associated with clinical outcomes in clinical trials, and thus may not accurately reflect patient severity of illness or septic shock pathobiology, and alternative markers like norepinephrine-equivalent dose or lactate concentration may be better indicators. However, in practice, lactate concentrations may not be routinely monitored at frequent intervals in all patients with septic shock, particularly in resource poor settings. Further, lactate concentrations in patients with septic shock have several causes and confounders (37, 38). Although clinical prediction models incorporating a multitude of data points may prove useful in the future, norepinephrine dose is the most pragmatic bedside clinical marker for vasopressin initiation and should be considered as the indicator for future studies. In fact, clinical trials are currently underway evaluating different timing strategies of vasopressin initiation based on the norepinephrine dose (39–41).

This study also evaluated two secondary outcomes and found that the clinical context for vasopressin initiation is associated with both hemodynamic response and organ dysfunction progression via SOFA score change. A prior evaluation of vasopressin recipients found that 45% of patients who received vasopressin had a positive hemodynamic response to its initiation (30). In that study, only two factors were found to be associated with hemodynamic response: lactate concentration at the time of vasopressin initiation and ICU location. The current evaluation corroborates these findings indicating that initiation of vasopressin at lower lactate concentrations is associated with improved hemodynamic response to vasopressin initiation. Additionally, both higher norepinephrine-equivalent dose and increasing lactate concentration at the time of vasopressin initiation were found to be associated with increased SOFA score change at 48 hours, reflecting worse organ dysfunction. These findings further indicate delaying initiation of vasopressin, compared with earlier initiation, is associated with worsened patient outcomes.

This study is limited by its sole inclusion of vasopressin recipients and inherent inability to compare early adjunctive vasopressin initiation to a counterfactual cohort receiving norepinephrine monotherapy, emphasizing the need for future clinical trials in this area. To account for differences in severity of illness at the time of vasopressin initiation, multiple methods of regression modeling with adjustment for confounders was conducted. However, residual differences in severity of illness may remain that impact the results. Specifically, the same composite scores for hospital mortality risk prediction, such as the Acute Physiology and Chronic Health Evaluation, were not available in both databases used, and therefore were not incorporated into regression models. Unmeasured confounders such as antibiotic selection and timing, as well as infectious source control, may have influenced analyses and were unable to be accounted for in this study. Additionally, charting of vasopressor doses in the eICU-CRD database is not consistent with regards to units of measure (µg/min vs. µg/kg/min) and concentration of the vasopressor product. Because of this, in some cases assumptions and conversions were required to make the data as homogenous as possible. To account for cases where extreme assumptions were needed, patients with uncertain vasopressor doses were excluded from this study, which may have led to a selection bias. The norepinephrine salt formulation used for included patients was unknown due to the large number of included centers. As all patients admitted to U.S. hospitals, it is assumed that all norepinephrine dosages are in base molecule equivalency (42). Last, while this study did not identify nonlinear relationships between the dependent variables and in-hospital mortality, the possibility of a nonlinear, time-varying association cannot be ruled out given the dynamic nature of septic shock and treatment response. Strengths of this study include its large sample size and inclusion of diverse patients from multiple U.S. health systems. Ultimately, results from this study corroborated findings from existing literature and support the need for this question to be answered in a prospective trial.

In patients with septic shock who ultimately receive vasopressin, this study emphasizes the benefit of early initiation of vasopressin, at lower norepinephrine-equivalent doses, at lower lactate concentrations, and earlier in the course of the patients’ shock. Randomized clinical trials should be conducted to determine the effect of early vasopressin initiation.

Supplementary Material

cc9-7-e1284-s001.pdf ^{(1.4MB, pdf)}