FormalPara Take home message

Vasopressin has similar 28-day mortality to norepinephrine but a different side effect profile (more digital ischaemia, fewer arrhythmias). We found no clear evidence for differential subgroup effects.

Introduction

Vasopressor therapy is used in septic shock to increase vascular resistance, raise mean arterial pressure and maintain perfusion of critical body tissues and organ systems [1]. The traditional approach involves using catecholamines but these are associated with risks including myocardial ischaemia and tachycardia with beta-agonists [1, 2]. These adverse effects have led to interest in adjunctive therapeutic strategies.

In shock states vasopressin acts as a potent vasoconstrictor via V1a receptors on vascular smooth muscle [3]. A relative deficiency of vasopressin in septic shock has been described [4] and administration of exogenous vasopressin reduces catecholamine requirements with the hypothesis that it may therefore also diminish the likelihood of catecholamine-related side effects. Furthermore, vasopressin may have additional benefits in terms of organ perfusion due to the distribution of the family of vasopressin receptors in different vascular beds and additional immunological effects compared with norepinephrine [5].

The Vasopressin and Septic Shock Trial (VASST) was the first large randomised comparison of vasopressin with norepinephrine [6]. Although there was no significant mortality benefit to addition of low-dose vasopressin in the overall population, there was a lower mortality in the subgroup of patients with less severe shock (< 15 μg/min norepinephrine). Subsequent post hoc analyses and other studies suggested a potential reduction in renal dysfunction with higher dose vasopressin [7,8,9] as well as a potentially synergistic interaction with corticosteroid treatment [10, 11]. This was specifically investigated in the Vasopressin versus Norepinephrine as Initial Therapy in Septic Shock (VANISH) trial, which found that early vasopressin use compared with norepinephrine did not increase the number of renal failure-free days and there was no interaction with corticosteroids [12]. However, the investigators noted that the confidence interval for renal-failure-free days included a potentially clinically relevant benefit for vasopressin as well as a reduced use of renal replacement therapy (RRT).

Although trials have not demonstrated a clear benefit for vasopressin, it is uncertain whether particular subgroups of patients benefit from this treatment. In particular, the following subgroups are of interest: duration of shock before receiving therapy, severity of shock by lactate level or by norepinephrine requirement and severity of kidney injury. The Surviving Sepsis Campaign 2016 guidelines use the totality of evidence to date to recommend low-dose vasopressin as a potential adjunctive vasopressor to norepinephrine [13].

Individual patient data (IPD) meta-analyses are considered the gold standard for synthesising information from RCTs [14]. They provide a means to explore further some of the aforementioned uncertainties around the possibility of different effects for different subgroups and to standardise the analysis of outcomes. The provision of the IPD increases the statistical power for investigating differential treatment effects [15]. Therefore, the aim of this study was to use IPD meta-analyses to quantify the efficacy and safety of vasopressin therapy within RCTs for septic shock, both overall and in a priori defined subgroups. We hypothesised that there might be beneficial effects of vasopressin therapy in specific subgroups despite the overall outcomes reported thus far by large trials.

Methods

The protocol for this study was published in the online PROSPERO database (CRD42017071698) prior to the analysis. The protocol is available at: http://www.crd.york.ac.uk/PROSPERO/display_record.php?ID=CRD42017071698.

This manuscript has been prepared in line with the guidelines by the PRISMA-IPD group and a checklist is available with the Supplementary Appendix [16].

Trial identification, selection and acquisition of data

We performed a comprehensive search using MeSH and free-text terms for various forms of the terms ‘septic shock’ and ‘vasopressin’, including specific drug names. The exact search strategy is listed in Appendix 1 of the study protocol. The following electronic databases were searched from inception to January 2019: MEDLINE, Embase, Science Citation Index Expanded, Cochrane Central Register of Controlled Trials, Clinicaltrials.gov and the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) search portal. Additional articles or abstracts were retrieved by manually scrutinising the reference list of relevant publications. There were no restrictions on language. We also searched conference abstracts from major critical care conferences for the last 3 years (full details in study protocol).

Publications were selected for review if they satisfied the following inclusion criteria: non-crossover randomised controlled trial, human adults with septic shock requiring vasopressor therapy [as defined by the trial investigators or the International Consensus definitions for Sepsis (1992, 2001, 2016)], intervention (vasopressin) versus any other vasoactive comparator with minimum duration for therapy of 3 h and/or until ICU discharge.

After removal of clearly irrelevant records, two authors (MN, ACG) independently screened abstracts for potentially eligible studies. Full-text reports were then assessed for eligibility. Where there was not enough information to make a decision on inclusion from published information, study authors were contacted for further details. Authors of eligible studies were invited to supply anonymised individual patient data sets. The variables requested of authors are detailed in Appendix 2 of our study protocol. Risk of bias was assessed by applying the Cochrane Risk of Bias tool (performed by MN, who was not involved in the conduct of any included studies) [17].

Our primary efficacy outcome was mortality at day 28 and our primary safety outcome was total number of serious adverse events (SAEs). Secondary outcomes included rates of use of RRT (in patients without end-stage renal failure), duration of RRT, duration of shock, duration of ventilation, renal failure-free days to day 28, shock-free days to day 28, ventilation-free days to day 28, ICU-free days to day 28, long-term mortality (maximum follow-up day 60–180), ICU length of stay and hospital length of stay. For duration of time outcomes, results were also reported for survivors and non-survivors separately because of the competing risk of death.

Statistical analysis

We estimated the overall intervention effects and generated forest plots using a conventional two-stage approach (trial summary measures that are then combined by standard meta-analytical methods) [18]. For dichotomous outcomes such as proportion dead at day 28, we used the numbers of events and patients to calculate the risk ratio. For continuous outcomes such as length of stay, we used the mean and standard deviation to calculate the mean difference. These estimates were then combined in a fixed-effect model (Mantel-Haenszel method) that was stratified by trial. Although some continuous data may not be normally distributed, this was explored in a sensitivity analysis (see below) and no variables were transformed for the main analysis.

We planned to assess the following a priori defined subgroups: early versus late onset of shock to study inclusion (early defined as ≤ 12 h, late defined as > 12 h), low versus high baseline vasopressor requirement (low defined as < 15 μg of norepinephrine per minute or equivalent, high defined as ≥ 15 μg), low versus high baseline lactate (low defined as lactate level ≤ 2 mmol/l, high defined as > 2 mmol/l), low versus high severity of acute kidney injury at baseline—low defined as no acute kidney injury (AKI) and stage 1 AKI [19] or Risk (RIFLE criteria) [20], high defined as stage 2 and 3 AKI [19] or Injury and Failure (RIFLE criteria) [20]—and studies considered low versus high trial risk of bias (low defined as all domains judged at low risk, high defined as any domain judged at unclear or high risk).

To explore the effect of patient characteristics on outcomes, we fitted one-stage regression models with single treatment covariate interactions. Three specifications of model were assessed as detailed in the Addendum to Tables S4–S7 in the Supplementary Appendix. Model fit was compared using the Akaike information criterion (AIC) [21].

We also planned sensitivity analyses to assess the impact of not transforming skewed data to approximate normality (see Addendum to Table S8 in the Supplementary Appendix for further details) and to assess for model robustness through both one- and two-stage models with either trial fixed or trial random-effects approaches. No adjustments were made for multiple comparisons and this should be kept in mind when interpreting 95% confidence intervals for secondary outcomes. All analyses were performed in Stata SE version 12.1 (College Station, TX).

Results

The electronic search was performed initially in July 2017 and then updated to January 2019. It yielded 5952 records in total for further assessment (see Fig. S1 in the Supplementary Appendix). There were no extra records identified by conference abstract searching that were not already selected in the electronic search. After screening, 21 full-text records were assessed of which 10 were excluded immediately. Reasons for exclusion of full-text records are detailed in Fig. S1 in the Supplementary Appendix. The authors for all 11 potentially includable studies were contacted to determine full eligibility and obtain IPD. Two replies confirmed that no outcomes of interest had been collected and these studies were then excluded [22, 23]. Owing to a lack of replies from five studies, only four remained for inclusion in the analysis [6, 12, 24, 25].

The samples from four of the five unavailable studies [8, 26,27,28] were relatively small (median 30 patients, range 23–42) but the trial by Oliveira et al. had a sample size of 387 [29]. However, there was no public protocol, no entry in a trial registry and no peer-reviewed manuscript for this trial (only two abstracts dating from 2011 to 2014). All five non-included studies were rated overall at high risk of bias (see Table S1 in the Supplementary Appendix for risk of bias assessment for all nine potentially includable studies). The three trials VASST (n = 779), VANISH (n = 409) and VANCS II (n = 250) were rated at low risk of bias in all domains. The trial by Dunser et al. (n = 15) was rated overall at high risk of bias. Individual patient data were available for these four studies and no important issues were identified during checking of the data.

Baseline patient characteristics of the combined data set are displayed in Table 1. There were a total of 1453 eligible patients and norepinephrine was the control in all four trials. Median age and APACHE II score were 64 years and 26, respectively. Approximately 71% were ventilated at baseline (Table 1). Although just under 40% of patients had a stage 2 or greater AKI at baseline, only 3% were receiving RRT (Table 1). Vasopressin dosing was lower in VASST compared with the other trials (VASST: 0.01–0.03 U/min, VANISH: up to 0.06 U/min, VANCS II: 0.01–0.06 U/min, Dunser et al.: up to 0.066 U/min). The trial by Dunser et al. enrolled 48 patients in total but we only included the 15 patients with septic shock for our analysis.

Table 1 Patient characteristics in combined individual patient data set
Full size table

Crude outcome data are displayed in Table 2. Overall 28-day mortality was 38% (49% when restricting to the 904 patients with a lactate > 2 mmol/l satisfying the septic shock 3.0 definition [30]) and varied between trials [VASST (37%), VANISH (29%), VANCS II (55%), Dunser et al. (60%)]. For the primary outcomes, there was no evidence of an effect of vasopressin on 28-day mortality [relative risk (RR) 0.98, 95% CI 0.86–1.12] or SAEs (RR 1.02, 95% CI 0.82–1.26). Both analyses had no evidence of significant between-trial heterogeneity (I2 0% and 8%, respectively) and forest plots are shown in Fig. 1a, b.

Table 2 Primary and secondary outcomes
Full size table
Fig. 1
figure 1

Forest plot of primary outcomes: (a) 28-day mortality, (b) serious adverse events and secondary outcome, and (c) requirement for renal replacement therapy (RRT)

Full size image

Serious adverse events, stratified by type, are displayed in Table 3. Although the overall number of events was similar, the side effect profile between vasopressin and norepinephrine was different. Vasopressin led to more digital ischaemia [absolute risk difference (ARD) 1.7%, 95% CI 0.3%–3.2%] but fewer arrhythmias (ARD − 2.8%, 95% CI − 0.2% to − 5.3%). Mesenteric ischaemia and acute coronary syndrome events were similar between groups.

Table 3 Serious adverse events
Full size table

For the secondary outcomes, vasopressin reduced the requirement for RRT (RR 0.86, 95% CI 0.74–0.99) with no evidence of between-trial heterogeneity (I2 27%, see Fig. 1c). There was no evidence for an effect of vasopressin on any other secondary outcome (see Tables S2a in the Supplementary Appendix).

There were no significant treatment covariate interactions in the a priori defined subgroups that we investigated (see Tables S4–7 in the Supplementary Appendix). Forest plots stratified by subgroup for the two primary outcomes and for two exploratory secondary outcomes are shown in Figs. 2 and 3, respectively.

Fig. 2
figure 2

Subgroup analysis for primary outcomes (28-day mortality and serious adverse events)

Full size image
Fig. 3
figure 3

Subgroup analysis for exploratory secondary outcomes [90-day mortality and requirement for renal replacement therapy (RRT)]

Full size image

In sensitivity analyses, there was no material difference in primary outcome results with a random-effects specification or one-stage analyses (full results in Tables S2b, S3a and S3b in the Supplementary Appendix). There was no material difference in the relative risk of 28-day mortality when using aggregate data from the five non-included studies where mortality was reported (see Figures S2 in the Supplementary Appendix). For secondary outcomes, there was no material difference in results with one-stage analyses but the reduced RRT requirement with vasopressin was non-significant with a random-effects specification (RR 0.83, 95% CI 0.67–1.01, see Table S2b in the Supplementary Appendix). Where skewed continuous data were transformed to approximate normality, the only material difference in results pertained to duration of shock (defined as an ongoing requirement for vasopressors). Here, vasopressin consistently demonstrated a longer duration of shock in all patients (1.2 h, 95% CI 1.1–1.4) and in only survivors (1.2 h, 95% CI 1.0–1.4) that was statistically though not clinically significant: median shock duration in the total population was 50 h [interquartile range (IQR) 28–91 h)], see Table S8 and the Addendum to Table S8 in the Supplementary Appendix for further details.

Discussion

This is the first individual patient data meta-analysis of vasopressin in septic shock and there are four main findings. First, we found no evidence of a statistically significant reduction in 28-day mortality with the use of vasopressin therapy in adults with septic shock. In our analysis, observed 95% CIs were consistent with an effect that ranges between a 14% relative reduction and a 12% relative increase in the risk of 28-day mortality with vasopressin therapy. Second, vasopressin appears safe regarding serious adverse events overall and in all planned subgroups but with a different side effect profile to norepinephrine borne out by more digital ischaemia and fewer arrhythmias. Third, there is weak evidence for vasopressin resulting in a reduced requirement for RRT. Fourth, we had hypothesised that that there might be beneficial effects of vasopressin therapy in specific subgroups but the subgroup interactions were not statistically significant although the 95% confidence intervals imply considerable uncertainty.

Comparing our findings to the existing literature requires caution for several reasons. Most existing non-IPD meta-analyses assess not only vasopressin but also its analogues and these are assessed in distributive shock states other than just sepsis (e.g., vasoplegia post cardiac surgery). Also, the as yet unpublished VANCS II trial has not been included in any meta-analyses to date [24]. McIntyre and colleagues found a reduced 28-day mortality in the subgroup of septic trials [31]. However, these trials also included terlipressin and limiting to only vasopressin resulted in a non-significant estimate. Vasopressin did not have a material impact on ICU or hospital length of stay, and this was in line with our results. A recent non-IPD meta-analysis by Nedel et al. found no significantly reduced incidence of RRT in the subgroup of septic patients receiving vasopressin (OR 0.75, 95% CI 0.54–1.04) but the effect estimate was broadly similar to ours [32]. The variation in event rates among the four included trials is not unexpected given the different inclusion/exclusion criteria and trial time frames that they encompass [33].

The high number of critical care trials that have found no mortality benefit for a proposed intervention has resulted in increasing attention toward enrichment strategies and identification of specific subgroups that disproportionately benefit from a therapy. Our IPD meta-analysis did not show significant interactions in the subgroups. In a re-analysis of the VASST trial, Russell and colleagues assessed the impact of the new septic shock 3.0 definitions on the original trial results [34]. They found that vasopressin was not efficacious in the new septic shock 3.0 cohort but was beneficial in patients with a lactate level ≤ 2 mmol/l. Bhatraju et al. investigated the response to vasopressin in two distinct AKI sub-phenotypes. Three hundred twenty-eight patients from the VASST trial who had IL-8, Ang-1 and Ang-2 measured were classified into two groups with a significantly improved 90-day mortality only found in one of the groups [35]. However, Antcliffe and colleagues performed a post hoc analysis of 176 VANISH trial patients with blood samples enabling categorisation into two groups according to their transcriptomic sepsis response signatures and found no significant interaction between vasopressin/norepinephrine and 28-day mortality for the two groups [36]. These studies highlight the differential responses to vasopressin for some but not all phenotypes.

The data on serious adverse events also merit attention. It may surprise clinicians that vasopressin, at these low doses, does not increase the incidence of mesenteric ischaemia. Vasopressin results in a decreased incidence of arrhythmias, possibly by decreasing the use of adrenergic vasopressors. This result is consistent with the SOAP2 trial where dopamine led to more arrhythmias than norepinephrine (dopamine has far greater beta-adrenergic activity than norepinephrine) [2]. Our findings provide actionable evidence for clinicians when choosing a vasopressor for septic shock patients: greater concern for arrhythmia will favour vasopressin while concern about digital ischaemia may favour norepinephrine.

Our findings have several implications. First, they may inform future versions of the Surviving Sepsis Campaign guidelines, which currently recommend vasopressin only as an adjunct agent [13]. In practice, this means that vasopressin is often used only as a rescue therapy. Additional guidance could recommend earlier use in those with dysrhythmias or a tachycardia. Second, our results align with the finding of reduced RRT use initially identified in the VANISH trial (although the overall evidence is weak considering the sensitivity analysis that does not achieve statistical significance and the multiple secondary outcomes assessed in this analysis) [12]. Therefore, this finding should be viewed as hypothesis generating only. RRT is an expensive critical care therapy [37] and future trials should incorporate health economic assessments that investigate the cost-effectiveness of vasopressin in this regard. Third, long-term mortality may be a more appropriate outcome for future vasopressin trials (at least 90 days). The ubiquitous provision of RRT in modern intensive care units ensures that renal failure is rarely a direct, early cause of mortality but renal injury may contribute to downstream longer term effects such as secondary episodes of sepsis and spiralling multi-organ failure, which are only apparent in mortality beyond the first month [38, 39].

The main strengths of our study are: a systematic and comprehensive search (with a pre-published protocol and analysis plan), the explicit inclusion of only randomised trial data, collection of individual patient data with which to facilitate standardised subgroup analyses and the use of models for assessing treatment-covariate interactions that accounted for aggregation bias where applicable. Our findings must also be considered in light of several limitations. First, not all trials reported data on every outcome of interest. Second, there are multiple comparisons with no adjustment made. Third, there were five trials that may potentially have provided additional data but that we excluded because of lack of further information from the study authors [8, 26,27,28,29]. Our analysis included 1453 patients from a total pool of 1975 (74%) if these five trials had also been included. As mentioned in the results, four of the studies have very small sample sizes [8, 26,27,28] and the fifth remains unregistered and unpublished [29]. Only one of these five trials included data on the RRT requirement [8]. It is unclear how inclusion of data from these studies may have altered our results. Fourth, the trials included a span of 14 years in which there have been major shifts in the management of septic shock that may have differentially affected trial populations [40, 41].

Conclusions

We found vasopressin therapy in septic shock had no effect on 28-day mortality although the confidence intervals are wide. It appears safe but with a different side effect profile to norepinephrine. The finding of reduced RRT should be interpreted cautiously given the multiple secondary outcomes assessed and the fact that the result was not robust to sensitivity analyses. Future trials should focus on long-term outcomes in select patient groups as well as incorporating cost-effectiveness analyses regarding possible reduced RRT requirements.