Sepsis, defined as a dysregulated host response to infection (1), carries high mortality (2). Optimal clinical management of this syndrome remains uncertain, although guidelines stress the importance of early recognition and timely resuscitation (3). Typically, this involves the administration of intravenous (i.v.) fluids and use of vasopressors (VP) to achieve specific hemodynamic targets (4). Three recent large randomized controlled trials of early goal-directed therapy (EGDT), as part of a resuscitation strategy in patients with early septic shock, demonstrated no mortality benefit in comparison to usual care (5–8).
Although i.v. fluid resuscitation is provided almost ubiquitously, the volume of fluid administered is highly variable, as is the use of VP (9, 10). Indeed, while the optimal timing of VP in septic shock is unknown, data from animal models (11) and human observational studies (12, 13) suggest this may modify clinical outcomes. In particular, if VP are delayed, persistent hypotension may lead to excessive i.v. fluid therapy, which has been associated with harm (14, 15). How VP are used in the early treatment of septic shock is not well described.
Accordingly, we aimed to describe the utilization of VP in patients presenting to the emergency department (ED) with septic shock and enrolled into the Australasian Resuscitation In Sepsis Evaluation (ARISE) trial. Specifically, we were interested in examining the timing, dose, predictors of, and variation in early VP support. We also hypothesized that earlier initiation of VP would be independently associated with less i.v. fluid therapy and differential 90-day mortality.
The ARISE trial was a large, multicenter randomized controlled trial of EGDT in comparison to usual care in patients presenting to the ED with septic shock (5). A detailed study protocol and statistical analysis plan have been published (16), and are discussed only briefly here. Ethical approval was provided by the coordinating center (Monash University, Australia), with individual institutional approval obtained at each participating site. Prior informed written consent or delayed consent was obtained from all patients or their legal surrogates.
Patients were eligible to be enrolled in the ARISE trial if they presented to the ED with suspected or confirmed infection, had 2 or more of the systemic inflammatory response syndrome criteria (17), and evidence of refractory hypotension or hypoperfusion within 6 h of presentation. Refractory hypotension was defined as a systolic blood pressure <90 mmHg or a mean arterial pressure <65 mmHg despite the administration of at least 1,000 mL of i.v. fluid therapy within a 60-min period. Hypoperfusion was defined as a blood lactate level of 4.0 mmol/L or more. The timing and choice of VP, and decisions regarding fluid administration prior to randomization, were at the discretion of the treating clinician.
Study procedures and outcomes
Following randomization, patients allocated to the usual care group had further i.v. fluid therapy and/or VP based on the judgment of the treating clinician. For those allocated to EGDT, fluid therapy and commencement of VP for the first 6 h post randomization was based on the published EGDT algorithm (16). Additional details are provided in Supplemental Digital Content 1 (http://links.lww.com/SHK/A827). Data were collected concerning i.v. fluid volumes, VP use (class, dose, duration, and timing), and other supportive therapy in both arms of the study.
The primary study outcome in ARISE was death from any cause within 90 days of randomization. The primary exposure variable in this post-hoc analysis was time from arrival in ED to commencing VP. Vasoactive agents considered as VP included: norepinephrine, epinephrine, metaraminol, or vasopressin. We explored the association between exposure and outcome by utilizing time to VP as both a continuous and categorical variable. In the latter case, we defined early VP use as commencing within 4 h of ED arrival, as this is a routinely employed metric in ED practice. In an attempt to limit confounding, a priori we decided to control for the following variables: ARISE group randomization, study inclusion criteria, total i.v. fluid (IVT) administered prior to commencing VP, age, gender, acute physiology and chronic health evaluation (APACHE) II score, plasma lactate, site of infection, and study institution. The study period extended to 72 h post ED presentation, and no imputation or assumptions were made about missing data.
Categorical data are summarized as n (%), continuous data as mean (SD) or median [interquartile range, (IQR)], depending upon their distribution. Between-group comparisons were performed by chi-squared, t test, or Wilcoxon rank-sum test as indicated. Time to VP was calculated from the time of commencement of first VP infusion minus the presentation time to the ED. IVT prior to VP was calculated as outlined in Supplemental Digital Content 2 (http://links.lww.com/SHK/A827), assuming a linear rate of administration over recorded time intervals.
Continuous analysis employed time-to-event models, treating those not commencing VP as censored beyond the 72 h study window. As per Fine and Gray (18), a competing risks framework was used to adjust for early mortality. Unadjusted and multivariable models were constructed to estimate the associations with EGDT group assignment and 90-day mortality, as separate analyses. All covariates listed above were included in the initial multivariable model and removed by backward elimination; covariates were retained if they were significant at P < 0.1, or deemed clinically relevant. Primary analysis used the definition of VP as above, with subsequent sensitivity analysis using norepinephrine alone. Results are provided as sub-hazard ratios (SHR) with 95% confidence intervals (95% CI).
Categorical analysis was conducted defining early VP as a binary covariate (commencement within 4 h of ED presentation). Patients not commencing VP within the study period were included in analysis, assigned to the > 4-h group. Univariable and multivariable logistic regression models were constructed as per the competing risk framework above, to allow crude and adjusted assessment of the association with 90-day mortality. Results are provided as odds ratios (OR) with 95% CI.
In order to identify those factors associated with early VP use and to obtain a further estimate of any potential effect on 90-day mortality, a propensity-matched treatment effects model was constructed. Factors associated with early VP were first identified by multivariable logistic regression. Those covariates that were significant at P < 0.1, or deemed clinically relevant, were retained. A 1:1, nearest neighbor propensity-matched treatment effects model was then created. Matching was assessed by standardized covariate balance summary statistics and graphical assessment of propensity score overlap, a caliper distance was set less than 0.25 times the score SD and unmatched observations excluded from analysis. The clinical effect of early VP was estimated as the average 90-day mortality effect (95% CI) between groups; early VP versus otherwise. All analyses were performed in Stata/MP 15.1 (StataCorp LP, College Station, TX).
Sixteen hundred (n = 1,600) patients were enrolled into the ARISE trial, 10 declined or withdrew consent and 1 in each group was lost to follow-up, leaving 1,588 patients (792 EGDT/796 usual care) available for analysis. The median [IQR] time from ED arrival to randomization was 2.7 [2.0, 3.9] h, with isolated hypotension in 853 (54%), isolated hypoperfusion (lactate ≥ 4.0 mmol/L) in 476 (30%) and both criteria in 259 (16%). Sixteen patients (1%) were receiving VP on arrival in ED, with 261 (16%) receiving VP prior to randomization. A total of 1,102 (69%) received VP at any point, 1,088 (99%) of whom commenced within 72 h of arriving in ED. Demographic, illness severity, treatment, and outcome data for patients receiving VP compared with those who did not, are provided in Table 1. In those who received VP, 38% did so prior to central venous access.
Vasopressor and i.v. fluid administration
The median [IQR] time from ED presentation to commencing VP was 4.4 [2.7, 7.1] h and was not significantly different between the study groups (control 4.4 [2.4, 7.1] h, EGDT 4.6 [2.8, 7.0] h, P = 0.44). Norepinephrine was the most common first-line agent (89%), followed by metaraminol (6%), and epinephrine (4%). Summary statistics for the administration rates of norepinephrine and epinephrine for the first 72 h, the proportion of VP type at randomization, prior to central venous catheter placement and by order of inclusion in treatment, are provided in Supplemental Digital Content Table 3 and Supplemental Digital Content Table 4 (http://links.lww.com/SHK/A827).
The EGDT group had a higher event rate (i.e., initiation of VP), with an unadjusted SHR (95% CI) of 1.21 (1.07, 1.37), P = 0.002, Figure 1A. When adjusted for the significant covariates of age, study inclusion criteria, APACHE II score, IVT prior to VP, site of sepsis, and treating institution, the SHR remained significant; 1.17 (1.02, 1.32), P = 0.02, Figure 1B. The median [IQR] volume of i.v. fluid administered prior to commencing VP was 3.1 [2.3, 4.3] L and did not differ significantly between study (control vs. EGDT) groups, P > 0.5, or with survival status at day 90, P = 0.14. The early VP group (commenced within 4 h of ED arrival) received significantly less i.v. fluid, 3.0 [2.0, 3.7] versus 3.8 [2.7, 5.2] L, P < 0.001, in comparison to the remainder of the cohort.
Factors associated with early vasopressor use
Early VP (within 4 h of arrival in ED), occurred in 480 (30%) of 1,588 patients. A comparison of demographic, comorbidity, illness severity, treatment, and outcome data for patients receiving early versus late/no VP is provided in Supplemental Digital Content Table 5 (http://links.lww.com/SHK/A827). Factors independently associated with early use were assessed by multivariable logistic regression. Increasing age and volume of i.v. fluid therapy were associated with a lower OR, while increasing APACHE II score was associated with a higher OR, P < 0.001, respectively. There was no difference in median Charlson Comorbidity Index between groups. The hemodynamic criteria for study entry (refractory hypotension, hypoperfusion, or both) had a variable effect, with greatest early VP use in those patients where both criteria were met, OR 2.2 (1.5, 3.1), P < 0.001, followed by isolated refractory hypotension (reference OR 1.0) and lowest use in those with isolated high lactate, OR 0.4 (0.29, 0.57), P < 0.001. Study institution had a significant, but variable effect as demonstrated graphically in Figure 2. The full regression table is provided in Supplemental Digital Content Table 6 (http://links.lww.com/SHK/A827). Select data concerning those patients who received VP peripherally, prior to central venous access, are provided in the Supplemental Digital Content Table 7 (http://links.lww.com/SHK/A827).
Association between timing of vasopressor support and 90-day mortality
In those who subsequently died within 90 days, the SHR (95%CI) for commencing VP was 1.7 (1.46, 1.95), P < 0.001 unadjusted and 1.4 (1.20, 1.68), P = 0.001, adjusted for age, APACHE II score, EGDT group, study inclusion criteria, IVT prior to VP, study institution, and site of infection – Figure 3. Similar findings were noted when this analysis was limited to patients commenced on norepinephrine alone (SHR 1.6 (1.40, 1.92), P < 0.001 unadjusted, and 1.4 (1.13, 1.63), P = 0.001, adjusted).
In categorical analysis, 90-day mortality was 27% (131 of 480) in the early VP group versus 15% (166 of 1,108), in the remainder of the study cohort, P < 0.001. Figure 4 depicts graphically the unadjusted relationship between time to VP and 90-day mortality. Factors independently associated with an increased odds of death in multivariable logistic regression included patient age, APACHE II score and plasma lactate, P < 0.001, and early administration of VP (within 4 h of ED presentation), P = 0.02. EGDT group assignment was non-significant. The full regression table is provided in Supplemental Digital Content Table 8 (http://links.lww.com/SHK/A827). Of note, there were no significant interaction effects between VP group (early compared with late) and EGDT assignment, for any of the clinical outcomes measured.
Finally, factors associated with early vasopressor administration (age, APACHE II score, EGDT group, study inclusion criteria, IVT prior to VP, and study institution – see Supplemental Digital Content Table 6, http://links.lww.com/SHK/A827) were used to build a propensity-matched treatment effects model. The estimated between group effect of early VP was a 9.8% (4, 16), P < 0.01, relative increase in 90 day mortality. Propensity score balance and overlap plots for the treatment effects model are provided in Supplemental Digital Content 9 (http://links.lww.com/SHK/A827).
In the ARISE EGDT trial, patients were commenced on VP a median of 4.4 h after arrival to ED, and following 3.1 L of i.v. fluid. In many instances (38%), this was prior to central venous access. Norepinephrine was most frequently employed first-line. Larger volume fluid resuscitation and increasing age were associated with a lower proportion receiving VP early, whereas increasing illness severity was associated with a greater proportion. Early VP use was not associated with greater chronic comorbidity, while study site had a significant, but variable effect. Consistent across time-to-event analysis, multivariable logistic regression, and propensity-matched treatment effects modeling, was an association between early VP administration and greater 90-day mortality. In addition, a significantly higher 90-day mortality rate was observed in those patients where VP were commenced 24 or more hours after presentation.
Comparison with previous studies
In a retrospective cohort (n = 213) of patients admitted to ICU, norepinephrine was commenced on average 3.1 ± 2.5 h after the onset of septic shock, with 40.4% starting this within 2 h (12). Time to initiation of norepinephrine was significantly longer in those who died, and continued to be independently associated with 28-day mortality (OR 1.39 (1.14–1.70) for every hour delay), when adjusted for patient, treatment, and illness severity characteristics (12).
In a much larger cohort of septic shock patients (n = 6,514), Beck et al.(13) identified time to VP (from onset of hypotension) as having a small, but independent effect on hospital mortality [OR 1.02 (1.01–1.03) for every hour delay], although this was primarily driven by patients who had the longest delays (>14.1 h). Norepinephrine was used first-line in approximately two-thirds, and the median time to VP was 3 h (13). Our data also suggest that very late initiation of VP (at or beyond 24 h from arrival to ED), may be associated with greater mortality. Indeed, this observation is consistent with data from Huang et al.(19), where delayed onset of septic shock after ICU admission was independently associated with greater in-hospital mortality. However, substantial caution must be exercised, as this group represents <5% of our study cohort (n = 38) and the 75th percentile was > 136 h, by which time factors unrelated to the primary illness are likely to have become more relevant.
In their retrospective analysis of 2,849 septic shock patients (48.8% of whom were admitted to the ICU from the ED), Waechter et al.(20) noted a complex interaction between i.v. fluid resuscitation and initiation of VP. The fluid/vasopressor combinations associated with the lowest in-hospital mortality included large volume fluid resuscitation early (0–6 h post onset hypotension), combined with delaying VP (at least 1 h after the onset of shock). Similarly, in our study, where VP were not initiated early, a survival advantage was noted.
Our study has identified VP as a common intervention provided to septic shock patients in the ED. In the setting of a clinical trial, this implies that the requisite resources and staff are generally available to provide such treatment early. In addition, a significant proportion of patients received VP without central venous access, implying that clinicians appear comfortable providing this therapy peripherally, at least in the short term. This has important implications for future clinical trials, in that the use of short-term peripheral VP may represent “standard” practice in many institutions. It should be noted, however, that complications associated with peripheral VP administration were not specifically collected as part of the ARISE trial, and therefore this practice may carry certain risks, that were not accurately captured.
Using 3 different statistical methods, a consistent finding was the association between early VP and greater 90-day mortality. Interpreting this observation is complex, as the decision to commence VP likely introduces a substantial selection bias, whereby the clinician has adjudged the patient to be significantly more unwell, than someone who is not deemed to require such therapy. In this respect, early VP may simply be an independent maker of greater illness severity. The variable effect of study site is also noteworthy, and implies significant heterogeneity in clinician and/or institutional attitudes to commencing VP, particularly in a well-resourced healthcare setting. In this manner, it may be that there is genuine equipoise about the role of early VP in septic shock, which provides a unique opportunity to consider further clinical trials, such as a “fluid-sparing approach” to sepsis resuscitation. Finally, the observation that very late initiation of VP was also associated with a greater risk of death, suggests this patient sub-group may not be ideally suited to inclusion in such a trial, given the likelihood that different mechanisms are involved.
Strengths and limitations
This study has utilized data from a large, multicenter, randomized controlled trial. All patients met a robust set of inclusion criteria, with the administration of antibiotic therapy mandated prior to randomization. In addition, outcome assessment was blinded to study intervention, with minimal loss to follow-up. As such, the data have high internal validity and are likely to be reflective of clinical practice across Australia and New Zealand. We have also employed varying statistical analysis techniques, which have generated highly concordant results, suggesting a robust assessment of the association between the timing of VP support and 90-day mortality.
As a post-hoc analysis, our findings are limited, particularly in terms of any causal inference. In particular, the decision to commence VP is a complex one, such that we are substantially limited by the inherent selection bias associated with initiating this therapy early. In determining the volume of fluid administered prior to commencing VP, we have assumed a linear rate of fluid administration over specific time periods, a constraint imposed by the manner in which the data were collected. This may not necessarily reflect clinical practice, where clinicians are likely to prescribe fluid boluses. In addition, we are unable to determine if any “delay” in providing VP represents an active clinical decision, or is a reflection of the infrastructure/processes of care at specific institutions.
Our data reinforce the imperative to conduct controlled clinical trials in this area (21), but also highlight the potential challenges. Specifically, variation in the site of infection, illness severity, physiological reserve, and clinician preferences are all likely to influence any potential effect from early VP. Indeed, the “right” combination of fluids and/or VP may be highly variable between patients, necessitating a substantial sample size.
In this post-hoc analysis of the ARISE trial, VP were commenced a median of 4.4 h after arriving to ED, following administration of a median of 3.1 L of i.v. fluid, and norepinephrine was most frequently employed first-line. Many patients received VP prior to central venous access. Early VP use was associated with greater 90-day mortality in adjusted and propensity-matched analysis.
A.U. gratefully acknowledges salary support from the National Health and Medical Research Council of Australia (Early Career Fellowship; GNT1124532).
The ARISE study was a collaboration of the Australian and New Zealand Intensive Care Society Clinical Trials Group (ANZICS CTG), The Australasian College for Emergency Medicine and the Australian and New Zealand Intensive Care Research Center (ANZIC-RC), Monash University. The trial was endorsed by the Irish Critical care Trials Group and the College of Intensive Care Medicine. The ARISE Management and Steering Committee: S. Peake (Chair), A. Delaney, R. Bellomo, P. A. Cameron, D. J. Cooper, A. Cross, C. Gomersall, C. Graham, A .M. Higgins, A. Holdgate, B.D. Howe, I. Jacobs, S. Johanson, P. Jones, P. Kruger, C. McArthur, J. Myburgh, A. Nichol, V. Pettila[Combining Diaeresis], D. Rajbhandari, S.A.R. Webb, A. Williams, J. Williams, P. Williams.
The ARISE site investigators: (alphabetically by institution and all in Australia unless specified to New Zealand, NZ, Finland, FL, or Hong Kong, HK or Ireland, IRE). The Alfred Hospital, Melbourne – V. Bennett, J. Board, P. McCracken, S. McGloughlin, V. Nanjayya, A. Teo. Auckland City Hospital, Auckland, NZ – E. Hill, P. Jones. E. O’Brien, F. Sawtell, K. Schimanski, D. Wilson. Austin Health, Melbourne – R. Bellomo, S. Bolch, G. Eastwood, F. Kerr, L. Peak, H. Young. Bendigo Hospital, Bendigo J. Edington, J. Fletcher, J. Smith. Blacktown Hospital, Blacktown - D. Ghelani, K. Nand, T. Sara. Box Hill Hospital, Melbourne – A. Cross, D. Flemming, M. Grummisch, A. Purdue. Canberra Hospital, Canberra - E. Fulton, K. Grove, A. Harney, K. Milburn, R. Millar, I. Mitchell, H. Rodgers, S. Scanlon. Central Gippsland Health Service, Sale – T. Coles, H. Connor, J. Dennett, A. Van Berkel. Christchurch Hospital, Christchurch, NZ – S. Barrington- Onslow, S. Henderson, J. Mehrtens. Coffs Harbour Base Hospital, Coffs Harbour – J. Dryburgh, A. Tankel. Dandenong Hospital, Melbourne – G. Braitberg, B. O’Bree, K. Shepherd, S. Vij. Frankston Hospital, Melbourne – S. Allsop, D. Haji, K. Haji, J. Vuat. Geelong Hospital, Geelong – A. Bone, T. Elderkin, N. Orford, M. Ragg. Gosford Hospital, Gosford – S. Kelly, D. Stewart, N. Woodward. Helsinki University Hospital, Helsinki, FL – V- P. Harjola, M. Okkonen V. Pettila[Combining Diaeresis], S. Sutinen, E. Wilkman. Hornsby Ku-ring-gai Hospital, Hornsby – J. Fratzia, J. Halkhoree, S. Treloar. Ipswich Hospital, Ipswich – K. Ryan, T. Sandford, J. Walsham. John Hunter Hospital, Newcastle – C. Jenkins, D. Williamson. Joondalup Health Campus, Joondalup – J. Burrows, D. Hawkins, C. Tang. Liverpool Hospital, Liverpool – A. Dimakis, A. Holdgate, S. Micallef, M. Parr. Logan Hospital, Meadowbrook – H, White, L. Morrison, K. Sosnowski. Lyell McEwin Hospital, Elizabeth Vale – R. Ramadoss, N. Soar, J. Wood. Manly Hospital, Manly – M. Franks. Middlemore Hospital, Auckland, NZ – A. Williams, C. Hogan, R. Song, A. Tilsley. Modbury Hospital, Modbury – D. Rainsford, N. Soar, R. Wells, J. Wood. Monash Medical Centre, Clayton – J. Dowling, P. Galt, T. Lamac, D. Lightfoot, C. Walker. Nepean Hospital, Penrith – K. Braid, T. DeVillecourt, H. S. Tan, I. Seppelt. Pamela Youde Nethersole Eastern Hospital, Chai Wan, HK – L. F Chang, W. S Cheung, S. K Fok, P. K Lam, S. M Lam, H. M So, W. W Yan. Port Macquarie Base, Port Macquarie – A. Altea, B. Lancashire. Chinese University of Hong Kong, Prince of Wales Hospital, Shatin, HK – C. D. Gomersall, C. A. Graham, P. Leung. Prince of Wales Hospital, Sydney – S. Arora, F. Bass, Y. Shehabi. Princess Alexandra, Woolloongabba – J. Isoardi, K. Isoardi, D. Powrie, S. Lawrence. Royal Adelaide Hospital, Adelaide – A. Ankor, L. Chester, M. Davies, S. O’Connor, A. Poole, T. Soulsby, K. Sundararajan. Royal Brisbane and Women's Hospital, Brisbane – J. Williams, J. H. Greenslade. Royal Melbourne Hospital, Melbourne – C. MacIsaac, K. Gorman, A. Jordan, L. Moore. Royal North Shore Hospital, St Leonards – S. Ankers, S. Bird, A. Delaney, J. Dowling, T. Fogg, E. Hickson, T. Jewell, K. Kyneur, A. O’Connor, J. Townsend, E. Yarad. Royal Perth Hospital, Perth – S. Brown, J. Chamberlain, J. Cooper, E. Jenkinson, E. McDonald, S. Webb. Royal Prince Alfred Hospital, Camperdown – H. Buhr, J. Coakley, J. Cowell, D. Hutch, D. Gattas, M. Keir, D. Rajbhandari, C. Rees. Sir Charles Gairdner Hospital, Nedlands – S. Baker, B. Roberts. St. Vincent's Hospital, Melbourne – E. Farone, J. Holmes, J. Santamaria, C. Winter. St. Vincent's Hospital, Sydney – A. Finckh, S. Knowles, J. McCabe, P. Nair, C. Reynolds. St. Vincent's University Hospital, Dublin/University College Dublin, IRE – B. Ahmed, D. Barton, E. Meaney, A. Nichol. Sydney Adventist Hospital, Wahroonga – R. Harris, L. Shields, K. Thomas. Tampere University Hospital, Tampere, FL – S. Karlsson, A. Kuitunen, A. Kukkurainen, J. Tenhunen, S. Varila. Tamworth Hospital, Tamworth – J. Burrows, N. Ryan, C. Trethewy. Toowoomba Hospital, Toowoomba – J. Crosdale, J. C Smith, M. Vellaichamy. Townsville Hospital, Townsville – J. Furyk, G. Gordon, L. Jones, S. Senthuran. Western Hospital, Footscray – S. Bates, J. Butler, C. French, A. Tippett. Westmead Hospital, Westmead – J. Kelly, J. Kwans, M. Murphy, D. O’Flynn. The Queen Elizabeth Hospital, Woodville South – C. Kurenda, T. Otto, S. Peake, V. Raniga, P. Williams. The Queen Elizabeth Hospital, HK – H. F. Ho, A. Leung, H. Wu.